Danielle C. Breen* I. Introduction Fake news is a term deeply

______________________________________________________________________________

SILENT NO MORE: HOW DEEPFAKES WILL FORCE COURTS TO

RECONSIDER VIDEO ADMISSION STANDARDS

Danielle C. Breen

I. Introduction

Fake news is a term deeply embedded into our everyday

vocabulary.

The spread of disinformation online has led to a rapid

change in the way that stories are consumed and viewed by the public.

There are even media outlets designed exclusively for spreading false

news—many under the guise of well-known networks.

It is more

J.D. Candidate, Suffolk University Law School, 2021; B.A. French Language &

Literature, B.A. Sociology, University of Colorado Boulder, 2016. Danielle can be

reached at [email protected].

See Fernando Nuñez, Note, Disinformation Legislation and Freedom of

Expression, 10 U.C. IRVINE L. REV. 783, 786 (2020) (describing fake news as

information that is “routinely used to describe subjectively unfavorable content or

inaccurate content that is the result of a mistake.”); Holly Kathleen Hall, Deepfake

Videos: When Seeing Isn’t Believing, 27 CATH. UNIV. J. L. & TECH. 51, 53 (2018)

(defining fake news as “information that is invented by people or governments for

their own purposes”).

See Nuñez, supra note 1, at 786 (noting the speed at which false information

spreads online). “Research from Massachusetts Institute of Technology (MIT)

suggests that false content spreads up to six times faster than factual content on social

media sites and false news stories are seventy percent more likely to be shared.” Id.

See Don’t get fooled by these fake news sites, CBS NEWS (Nov. 17, 2019)

[hereinafter Don’t get fooled], archived at https://perma.cc/ARS8-7JMY (listing

various fake news websites that the public should be aware of); Maxwell Library,

Evaluating Websites: Identifying Fake News Sources, BRIDGEWATER STATE UNIV.

(Nov. 17, 2019), archived at https://perma.cc/5E6M-GPA9 (identifying different

ways to determine if a website is fake). For example, the website abcnews.com.co

is designed to fool users into believing it is the authentic ABC News site. Don’t get

fooled, supra.

______________________________________________________________________________

2021] SILENT NO MORE

123

important than ever before to aptly identify disinformation.

A large

portion of the public is skeptical of previously trusted content.

However, even with growing awareness of disinformation, an

alarmingly high percentage of people still admit to knowingly sharing

false information online.

Modern technology and computer-generated imagery further

complicate the ability to decipher true information.

Through artificial

imagery known as “deepfake,” it is now possible to take

disinformation to the next level by creating doctored videos of events

See Nuñez, supra note 1, at 785–86 (explaining the difference between

disinformation and misinformation). “Disinformation is a more serious threat to

freedom of expression because it is information that is deliberately created to mislead

and influence the public, unlike misinformation, which may be shared under a

genuine belief that its contents are truthful.” Id. See Cat Zakrzewski, Report urges

social media companies to take down ‘provably’ false information, BOS. GLOBE

(Sept. 3, 2019), archived at https://perma.cc/5VVT-SNBQ (providing nine

recommendations from a New York University report on how social media

companies should combat the spread of disinformation on their platforms).

See Joshua Benton, Here’s how much Americans trust 38 major news organizations

(hint: not all that much!), NIEMANLAB (Oct. 5, 2018), archived at

https://perma.cc/2P5W-RCBP (asserting that 72% of Americans trusted the media in

1972 compared to 32% in 2016). See also Ashley Smith-Roberts, Article, Facebook,

Fake News, and the First Amendment, 95 DENV. L. REV. 118, 120 (2018) (describing

the goal behind the spread of disinformation as “‘erod[ing] trust in mainstream

media, public figures, government institutions”’); Madeline Lamo & Ryan Calo,

Regulating Bot Speech, 66 UCLA L. REV. 988, 993–94 (2019) (discussing the

prevalence of bots and how quickly they can generate false content online).

See Michael Barthel et al., Many Americans Believe Fake News Is Sowing

Confusion, PEW RSCH. CTR. (Dec. 15, 2016), archived at https://perma.cc/DZB9-

NN58 (providing statistics showing that 23% of Americans admitted to sharing a

fake news story). See also Patrick Huston & Eric Bahm, Deepfakes 2.0: The New

Era of “Truth Decay”, JUST SEC. (Apr. 14, 2020), archived at

https://perma.cc/92WS-6Y2G (declaring that “[o]ver half of Generation Z gets its

news and information primarily from social media and messaging apps on their

smartphones.”).

See Rebecca Delfino, Pornographic Deepfakes: The Case for Federal

Criminalization of Revenge Porn’s Next Tragic Act, 88 FORDHAM L. REV. 887, 889–

90 (2019) (stating that deepfakes and technology used to create them have become

“widely available”); Douglas Harris, Article, Deepfakes: False Pornography is Here

and the Law Cannot Protect You, 17 DUKE L. & TECH. REV. 99, 99–100 (2019)

(quoting Google’s CEO who believes that “artificial intelligence will change

humanity more profoundly than fire.”). See also USPTO Launches Fake Specimen

Informant Program, GERBEN (Jan. 23, 2020) [hereinafter USPTO], archived at

https://perma.cc/4CZF-WLZP (reporting on fake images infiltrating patent and

trademark applications).

______________________________________________________________________________

JOURNAL OF HIGH TECHNOLOGY LAW [Vol. XXI: No. 1

124

or people.

Deepfake creators use existing images and manipulate

them to construct completely different impressions of what occurred.

Although some deepfakes are identifiable with the naked eye, the

public’s ability to recognize them is likely to decline as technological

innovation enables the creation of more realistic doctored videos.

This new technology threatens the traditional treatment of video

evidence by courts as a trustworthy representation of events.

The

See Bill Hochberg, YouTube Won’t Take Down A Deepfake of Jay-Z Reading

Hamlet – “To Sue Or Not To Sue”, FORBES (May 18, 2020), archived at

https://perma.cc/XT8S-9GZD (highlighting the issue that deepfakes pose when an

individual’s likeness is used without their consent); David Frum, The Very Real

Threat of Trump’s Deepfake, ATLANTIC (Apr. 17, 2020), archived at

https://perma.cc/XA5S-9PDJ (reporting how Donald Trump intentionally tweeted an

obvious deepfake video of Joe Biden while serving as president). See also Benjamin

Goggin, From porn to ‘Game of Thrones’: How deepfakes and realistic-looking fake

videos hit it big, BUS. INSIDER (July 23, 2019), archived at https://perma.cc/X3LB-

8GP3 (outlining the long history of individuals creating deepfake videos of

politicians). In April 2018, BuzzFeed created a deepfake video of Barack Obama

saying things he never actually did to warn the public about the dangers of deepfake

technology. Id. Another deepfake video in July 2018 made Alexandria Ocasio-

Cortez appear as though she was unable to answer interview questions. Id. In May

2019, a deepfake video made Nancy Pelosi appear drunk at a political engagement.

Id.

See Bobby Chesney & Danielle Citron, Deep Falls: A Looming Challenge for

Privacy, Democracy, and National Security, 107 CALIF. L. REV. 1753, 1756 (2019)

(recounting how a deepfake video of a Parkland High School shooting survivor’s

speech quickly went viral).

See id. at 1757 (noting that sophisticated deepfake technologies “are maturing

rapidly.”); Alex Engler, Fighting deepfakes when detection fails, BROOKINGS INST.

(Nov. 14, 2019), archived at https://perma.cc/5RRB-EQ2K (describing different

ways to identify deepfakes); Kevin Stankiewicz, ‘Perfectly real’ deepfakes will

arrive in 6 months to a year, technology pioneer Hao Li says, CNBC (Sept. 20,

2019), archived at https://perma.cc/GKX5-2GAG (warning that deepfake

technology will soon be so realistic that humans will not be able to spot manipulated

videos with the naked eye). “Carefully made deepfakes can already be very realistic,

though only under certain circumstances—an attentive observer will notice that

convincing deepfakes focus on individuals who don’t wear glasses or have beards,

and typically use a stationary camera.” Engler, supra.

See Jane A. Kalinski, Jurors at the Movies: Day-In-The-Life Videos as Effective

Evidentiary Tool or Unfairly Prejudicial Device?, 27 SUFFOLK UNIV. L. REV. 789,

789–90 (1993) (illustrating the value behind video evidence as a tool for lawyers to

present information in a more digestible form to juries). See also Jonathan Mraunac,

The Future Of Authenticating Audio And Video Evidence, LAW360 (July 26, 2018),

archived at https://perma.cc/VUZ2-AWGE (examining the inherent trust that juries

______________________________________________________________________________

2021] SILENT NO MORE

125

justice system views the ability of jurors to make determinations using

visual evidence so highly that a District of Columbia Appeals Court

held it constitutionally permissible for prosecutors to exclude blind

persons from juries.

People like to believe what they see with their

own eyes, and the prevalence of deepfake imagery makes this an

incredibly dangerous assumption.

The growing use of deepfake technology will require courts to

re-evaluate the typical treatment of video evidence. A large number

of jurisdictions have traditionally allowed the admission of video

evidence in jury trials without a witness testifying before the jury that

the video is a fair and accurate representation of what occurred. The

evidence is deemed reliable behind the scenes by a judge and then

presented to the jury. With deepfakes causing many people to question

the authenticity of seemingly reliable videos, it is urgent that this

practice changes. Deepfake videos will force courts to always require

a witness to testify before the jury about the accuracy and authenticity

and society as a whole put into video evidence); David Dorfman, Decoding

Deepfakes: How do Lawyers Adapt When Seeing Isn’t Always Believing?, 80 OR.

STATE BAR BULL. 18, 20 (2020) (warning that all deepfakes cannot be identified with

detection technology).

See Yael Granot et al., In the Eyes of the Law: Perception versus Reality in

Appraisals of Video Evidence, 24 PSYCH. PUB. POL’Y & L. 93, 93 (2017) (describing

the holding in United States v. Watson, 483 F.3d 828 (D.C. 2007)). Prosecutors

striking blind persons from the jury demonstrates the idea that one must be able to

see in order to fully comprehend all of the evidence in a case. Id. Visual evidence

is so highly regarded in the justice system that it is seen as a way to mitigate juror

bias towards other pieces of evidence in a case. Id.

See Dorfman, supra note 11, at 20 (predicting that “in the not-so-distant future, as

manipulated media becomes more prevalent, an equal concern may be the impact of

deepfakes on trust in visual and audio recordings generally.”); Pakinam Amer,

Deepfakes are getting better. Should we be worried?, BOS. GLOBE (Dec. 13, 2019),

archived at https://perma.cc/WE47-U9R7 (warning how “[v]ideo is not a substitute

for truth . . . at least not anymore.”); Drew Harwell, Top AI researchers race to detect

‘deepfake’ videos: ‘We are outgunned’, WASH. POST (July 12, 2019), archived at

https://perma.cc/3N8Y-C484 (discussing the dangers of highly realistic deepfakes

on perception and efforts to combat them); Grace Shao, Fake videos could be the

next big problem in the 2020 elections, CNBC (Oct. 15, 2019) [hereinafter Deepfakes

in 2020 elections], archived at https://perma.cc/25MP-9AEQ (cautioning how the

public will need to consider whether videos of politicians are deepfakes during the

2020 election). See also Philip Ewing, What You Need To Know About Fake Video,

Audio, And the 2020 Election, NPR (Sept. 2, 2019), archived at

https://perma.cc/S8JW-F5MX (describing how deepfakes are often detected after

having a widespread negative impact on the subject’s image).

______________________________________________________________________________

JOURNAL OF HIGH TECHNOLOGY LAW [Vol. XXI: No. 1

126

of video evidence in order to mitigate jurors’ doubt about the reliability

of evidence.

II. History

A. The Pictorial Evidence Theory

There are traditionally two different approaches to admitting

video evidence in court: the pictorial evidence theory and the silent

witness theory.

Under the pictorial evidence theory, visual evidence

is only admissible when a witness can testify before the jury that the

evidence is a fair and accurate representation of what occurred.

This

theory rests on the idea that any photographic or video evidence is just

a “graphic portrayal of oral testimony,” and therefore must be verified

See Tracey Bateman Farrell, Construction and Application of Silent Witness

Theory, 116 AM. L. REV. 1, 2 (2019) (detailing the two approaches that courts take

towards admitting video evidence). See also FED. R. EVID. 401 (providing the test

for relevant evidence); FED. R. EVID. 402 (providing circumstances under which

relevant evidence is admissible); FED. R. EVID. 403 (providing circumstances under

which a judge may decide to keep relevant evidence out). Under both the silent

witness and pictorial evidence theory, the judge completes the 403-weighing test to

determine whether to allow the visual evidence in for the jury’s consideration. FED.

R. EVID. 403. See also James Alexander Tanford, THE PREJUDICE RULE, IND.

UNIV. MAURER SCH. L. (2014), archived at https://perma.cc/QYJ5-WJ6H

(illustrating how the 403 balancing inquiry works). Relevant evidence will be

excluded if its probative value is substantially outweighed by the risk of unfair

prejudice. Id.

See Farrell, supra note 14, at 2 (explaining the pictorial evidence standard). See

also Ex parte Rieber, 663 So. 2d 999, 1009 (Ala. 1995) (holding that a witness’s

knowledge must be verified under pictorial evidence theory). The court held that the

prosecution must demonstrate that the witness testifying about the events in the

security tape had intimate knowledge of what occurred. Id. at 1011. See also Fisher

v. State, 643 S.W.2d 571, 573 (Ark. Ct. App. 1982) (holding that videotape evidence

could not be admitted without a witness under the pictorial evidence theory). The

Fisher Court held that absent a witness verifying the events in the videotape were

accurately represented and the video was not tampered with prior to trial, it could not

be admitted into evidence. Id. at 573–74. See also Ex parte Fuller, 620 So. 2d 675,

679 (Ala. 1993) (providing an example of questioning that appropriately laid the

foundation for the pictorial evidence theory). The prosecution questioned an

investigator who was present while the Defendant gave a recorded statement. Id.

The questions included the investigator’s experience with the particular tape recorder

used, whether he had played the recording before trial, and whether it was an accurate

representation of what occurred. Id.

______________________________________________________________________________

2021] SILENT NO MORE

127

as correct by a witness.

The witness must testify that the video

accurately represents the subject discussed, but they do not need to

have been present at the time it was created.

There is also no

requirement that the witness be an expert in photography or

videography.

The witness only needs to have personal knowledge of

the subject material to reliably confirm that the events presented are

authentic.

The classic example of the pictorial evidence theory is a

medical examiner testifying before the jury during a murder trial about

the nature of a victim’s wound.

The pictorial evidence theory became increasingly relevant

with the development of photo-editing technology such as Adobe,

which gave parties the ability to easily alter evidence.

This theory

See Jordan S. Gruber, Foundation for Contemporaneous Videotape Evidence, 16

AM. JURIS. PROOF FACTS 493, § 5 (2019) [hereinafter Contemporaneous Videotape

Evidence] (setting forth why photographic and video evidence must be verified by a

sponsoring witness at trial). Under the pictorial evidence theory, the “witness must

testify that witness has sufficient personal knowledge of scene or events pictured or

sounds recorded and that item offered accurately and reliably represents actual scene

or sounds.” Id.

See Benjamin V. Madison III, Scientific Evidence Symposium: Note: Seeing can

be Deceiving: Photographic Evidence in a Visual Age — How much Weight does it

Deserve?, 25 WM. & MARY L. REV. 705, 708 (1984) (introducing requirements of

witnesses under the pictorial evidence theory). The pictorial evidence theory is used

with fingerprint and other evidence that is not meaningful to an “untrained eye”

without explanation. Id. The pictorial evidence theory is also commonly used to

depict conditions described by a witness, such as how far away something was at the

time of an accident. Id. at 710.

See Contemporaneous Videotape Evidence, supra note 16, § 5 (reiterating that the

witness need only have “sufficient personal knowledge” under the pictorial evidence

theory).

See id. (maintaining that a witness may have any background so long as they have

personal knowledge of events).

See Madison, supra note 17, at 710 (observing how medical examiners are

frequently required to provide context in murder cases because images of wounds

cannot be fully understood as accurate alone). “Photographic displays allow an

examiner to illustrate wounds that are difficult to conceptualize, such as numerous

stab wounds, multiple bruises, or extensive damage resulting from a gunshot

wound.” Id.

See Brian Barakat & Bronwyn Miller, Authentication of Digital Photographs

Under The “Pictorial Testimony” Theory: A Response to Critics, 78 FLA. BAR J. 38,

38 (2004) (discussing the importance of the pictorial evidence theory in litigation).

Manipulation of an image requires human action and allowing a person to testify to

a jury about the content of an image prior to and after its manipulation allows the

jury to still understand the image as “a true and accurate representation of what he

______________________________________________________________________________

JOURNAL OF HIGH TECHNOLOGY LAW [Vol. XXI: No. 1

128

permits a jury to hear witness testimony regarding any alterations

made to a photograph or video before concluding that the image is an

accurate representation of what the witness saw.

This allows the

introduction of video evidence that may be altered for reasons deemed

permissible by a judge, such as lightening the video to better show an

object.

B. The Silent Witness Theory

In contrast to the pictorial evidence theory, the silent witness

theory admits visual evidence absent a qualifying witness.

Instead,

a judge deems whether there is a sufficient foundation to admit the

evidence without a witness testifying before the jury.

The theory was

or she saw.” Id. See also Ashley Brown, Article, Picture

ImPerfect: Photoshop Redefining Beauty in Cosmetic Advertisements, Giving False

Advertising a Run For the Money, 16 TEX. REV. ENT. & SPORTS L. 87, 90–92 (2015)

(summarizing the developments of Photoshop since its creation in 1987).

See Bakarat & Miller, supra note 21, at 40 (reiterating how a witness need only to

testify to the foundational facts of the photograph in front of the jury for

authentication). Foundational facts can be as simple as the witness being able to

identify the subject in the photograph or video. Id. It is the responsibility of the

adverse party to challenge the photo or video evidence admitted. Id. See Timothy

Williams et al., Police Body Cameras: What Do You See?, N.Y. TIMES (Apr. 1,

2016), archived at https://perma.cc/JK6X-LCNA (exposing issues of perception

with video evidence). The New York Times study showed how police body cameras

only show the officer’s perspective and may not be an accurate depiction of what

actually occurred. Id. When the officer wore his chest camera, as many officers do

in the United States, it appeared he was in a threatening interaction with another

individual. Id. However, when the film of the interaction is shown from another

angle, the viewer sees that the interaction was actually two people dancing together.

Id. The reporter witnessing the two officers verified the actual interaction. Id.

See Madison, supra note 17, at 709 (asserting how altered evidence may be

admitted under the pictorial evidence theory if “the jury can understand the changes

in appearance that occurred between the relevant time and the time the photograph

was taken.”).

See Farrell, supra note 14, at 2 (indicating the components of the silent witness

theory). The silent witness theory allows for admission absent a sponsoring witness

because of the idea that the video evidence speaks for itself and its authenticity is

corroborated. Id. There are no set requirements to authenticate video evidence

because the facts of each case vary so widely. Id.

See id. (describing how evidence is authenticated under the silent witness theory).

Evidence under the silent witness theory draws its reliability from circumstances

other than a witness testifying that the image or video is an authentic portrayal of

______________________________________________________________________________

2021] SILENT NO MORE

129

originally proposed to more easily allow X-ray images and

surveillance videos into evidence.

The silent witness theory is a

representation of the inherent trust that society places in video

evidence because it demonstrates the belief that the video is a non-

biased account of events.

Under this theory, evidence is admissible

at the trial court judge’s discretion upon a showing that the video was

created under reliable processes and untampered with between the time

it was taken and presented to the court.

Judges admit visual evidence

under the silent witness theory as a trusted substitute for a qualifying

witness’s account of what happened; in other words, the process in

which the evidence was obtained is deemed sufficiently reliable for

admission.

Most jurisdictions apply the silent witness theory,

what they saw. Id. See Jordan S. Gruber, Videotape Evidence, 44 AM. JURIS. TRIALS

171, § 60 (2020) [hereinafter Videotape Evidence] (noting that foundational

requirements for the admission of evidence are more relaxed under the silent witness

theory). If there is no authenticating witness, the offering party can use “some other

testimony, such as a chain-of-custody argument or ‘the testimony of a photographic

expert who has determined that it had not been altered in any way and was not built-

up or faked,’ which clearly establishes the authenticity and competency of the

photographic evidence.” Id.

See Contemporaneous Videotape Evidence, supra note 16, § 5 (outlining the

original justification for the adoption of the silent witness theory). X-rays do not

provide any observations directly to a witness, which spurred courts to adopt the

silent witness theory to have them more easily admitted into evidence. Id. See also

Videotape Evidence, supra note 25, § 60 (indicating that the silent witness theory

can also come into play when there is no verifying witness to verify the accuracy of

the video).

See Granot et al., supra note 12, at 94 (emphasizing the trust placed in video

evidence by stating that viewers may be disinclined to question the creation of

images and what information is excluded); Zachariah B. Parry, Note, Digital

Manipulation and Photographic Evidence: Defrauding the Courts One Thousand

Words at a Time, 2009 ILL. J. L. TECH. & POL’Y 175, 176 (2009) (commenting on

the persuasive power of photographic evidence despite the fact that it “has never

been easier for photos to misrepresent the truth”).

See Contemporaneous Videotape Evidence, supra note 16, § 5 (outlining policy

considerations for using the silent witness theory). If the process behind the creation

of the video is deemed inherently reliable by the judge, the evidence may “speak for

itself.” Id. See Madison, supra note 17, at 711 (discussing why courts apply the

silent witness theory). Courts are generally reluctant to limit the use of photographic

evidence, which is a large underlying policy reason behind the application of the

silent witness theory. Id.

See Ex parte Fuller, 620 So. 2d 675, 678 (Ala. 1993) (explaining why the silent

witness theory allows evidence to be admitted absent a witness).

______________________________________________________________________________

JOURNAL OF HIGH TECHNOLOGY LAW [Vol. XXI: No. 1

130

showcasing the insurmountable value placed in video evidence to

show what actually happened in a case.

This assumption of

reliability can create issues in the courtroom, as case law demonstrates

that even multiple Supreme Court Justices have watched the same

video and concluded that different versions of events occurred.

Courts have adopted a variety of different approaches under the

silent witness theory.

This is because judges emphasize the facts in

each case vary so widely that it is difficult to apply one uniform

standard to evaluate all video evidence.

Some courts set general

guidelines in the application of the silent witness theory rather than

mandatory standards.

Other courts created a standard simply

[T]he process or mechanism substitutes for the witness’s senses,

and because the process or mechanism is explained before the

photograph, etc., is admitted, the trust placed in its truthfulness

comes from the proposition that, had a witness been there, the

witness would have sensed what the photograph, etc., records.

Id. See Madison, supra note 17, at 710–11 (articulating the weight evidence is given

after admission under the silent witness theory). “In practical terms, such

photographic evidence assumes greater significance than photographic evidence

authenticated by testimony. Instead of supplementing testimony on an issue, the

photographic evidence forms an independent basis upon which the proponent may

establish a fact or occurrence.” Id. at 711.

See Farrell, supra note 14, at 2 (asserting that while most jurisdictions have not

expressly adopted the silent witness theory, very few have explicitly rejected it);

Wagner v. State, 707 So. 2d 827, 830 (Fla. App. 1998) (setting forth the rationale

behind the silent witness theory).

See Granot et al., supra note 12, at 94 (reiterating what occurred in Scott v. Harris,

550 U.S. 372 (2007)). In Scott, the driver of a vehicle had a dashboard camera that

showed him crashing. Id. The majority of Supreme Court Justices agreed that

summary judgment was appropriate because no reasonable juror could watch the

dashboard video and conclude that the driver was not reckless. Id. However, a few

of the Justices disagreed. Id. See also Kalinski, supra note 11, at 798–99 (stressing

the dangers of using video evidence in jury trials). Although video evidence is a

powerful tool to condense tedious information into a more digestible format for

jurors, it creates a strong possibility for misrepresentation and misunderstanding. Id.

See generally Farrell, supra note 14 (listing cases in different jurisdictions

outlining various approaches to the silent witness theory in civil and criminal cases).

See id. at 2 (indicating that most courts generally have not taken a specific

approach to authenticating evidence under the silent witness theory).

See id. (introducing specific silent witness theory guidelines listed by an Indiana

appeals court); State v. Anglemeyer, 691 N.W.2d 153, 161–62 (Neb. 2005) (holding

that evidence is admissible when “it is a correct reproduction of what it purports to

depict.”); Kindred v. State, 524 N.E.2d 279, 298 (Ind. 1988) (holding that video

______________________________________________________________________________

2021] SILENT NO MORE

131

requiring verification of the chain of custody of the photographic or

video evidence.

A few courts have outlined specific step-by-step

processes that must be taken in order to verify video evidence.

One

of the more restrictive step-by-step approaches is the seven-prong

standard, adopted by Alabama courts, which lays out an enumerated

procedure of verifying the creation process of a video before it can be

admitted into evidence.

The goal of the seven-prong standard is to

ensure that video evidence is properly deemed authentic by the court

evidence just needs a strong showing of authenticity under the facts of the case to be

admitted under the silent witness theory).

See Mraunac, supra note 11 (discussing the chain of custody approach to the silent

witness theory). Evidence should be established that the camera was working

properly, the system was reliable, and that the evidence was properly handled until

the start of trial. Id. See also Mendoza v. Mashburn, 747 So. 2d 1159, 1172 (La. Ct.

App. 1999) (reiterating that “[t]he purpose of the chain of custody rule is to assure

the integrity of the evidence.”); Meador v. State, 664 S.W.2d 878, 880 (Ark. Ct. App.

1984) (maintaining that “[t]he purpose of the rule requiring a chain of custody is to

guard against the introduction of evidence which is not authenticated.”); Nelson v.

State, 687 P.2d 744, 746 (Okla. Crim. App. 1984) (asserting how “[t]he purpose of

the chain-of-custody rule is to ensure that the physical evidence against the accused

has not been tampered with or altered.”).

See Mraunac, supra note 11 (outlining the five-step test used by some jurisdictions

in applying the silent witness theory). A recording’s authenticity is determined by a

set of five factors: “(1) evidence of time and date, (2) the presence or absence of

evidence of tampering, (3) the operating condition and capability of the equipment

as it relates to the accuracy and reliability of the product, (4) operating, testing and

security procedures, and (5) the identification of participants depicted in the

recording.” Id.

See Bohannon v. State, 222 So. 3d 457, 494–95 (Ala. Crim. App. 2015) (applying

the seven-prong standard for the silent witness theory).

[The seven-prong] standard requires: (1) a showing that the device

or process or mechanism that produced the item being offered as

evidence was capable of recording what a witness would have seen

or heard had a witness been present at the scene or event recorded,

(2) a showing that the operator of the device or process or

mechanism was competent, (3) establishment of the authenticity

and correctness of the resulting recording, photograph, videotape,

etc., (4) a showing that no changes, additions, or deletions have

been made, (5) a showing of the manner in which the recording,

photograph, videotape, etc., was preserved, (6) identification of the

speakers, or persons pictured, and (7) for criminal cases only, a

showing that any statement made in the recording, tape, etc., was

voluntarily made without any kind of coercion or improper

inducement.

Id.

______________________________________________________________________________

JOURNAL OF HIGH TECHNOLOGY LAW [Vol. XXI: No. 1

132

prior to it ever reaching the jury.

This practice assures that jurors

evaluate video evidence in light of the circumstances of the case rather

than question whether or not it is authentic.

The jurisdictional

discrepancies regarding application of the silent witness theory

demonstrate both the complexity of visual evidence and the value

behind it.

In allowing this authentication process to occur outside of

the courtroom, the court system reinforces the inherent trust built into

video evidence by juries and society.

However, the silent witness

theory has the potential for serious error because not all judges are

familiar enough with modern photo and video editing technology to

fully understand the processes through which evidence may be

created—let alone to rightfully evaluate authenticity.

See id. at 494 (articulating the rationale behind adopting the seven-prong standard

in Alabama criminal courts). But see Videotape Evidence, supra note 25, § 61

(arguing that the seven-prong test to verify chain of custody and foundational

requirements is an unnecessarily high standard for the admission of video evidence).

See Pressley v. State, 770 So. 2d 115, 132–33 (Ala. Crim. App. 1999) (providing

an example of a court applying the seven-prong test to verify video evidence prior to

its admission). The court held the seven-prong test was satisfied when the officer

testified how the video surveillance system worked, the video was kept in his sole

custody, the video was in the same condition at trial, and that there were no changes

made to the video since it was in his custody. Id. See Videotape Evidence, supra

note 25, § 62 (conceding that there may be a “revival of certain foundational

requirements” with the prevalence of video editing).

See Farrell, supra note 14, at 2 (listing different circumstances where visual

evidence was admitted under the silent witness theory). “Courts have held police

surveillance videotapes, bank, store, or business security surveillance videotapes,

and videotapes from a bank automatic teller machine admissible . . . [o]ne court has

also held that enhanced still prints made from a videotape were admissible . . . . ” Id.

See Patterson Dubois, Some Observations on the Psychology of Jurors and Juries,

53 PROC. AM. PHIL. SOC’Y 307, 316 (1914) (emphasizing the impact that visual

evidence has on the perception of truth in cases); Mraunac, supra note 11

(highlighting common phrases used to demonstrate trust in video and photographic

evidence such as “seeing is believing”). “The very fact that the photographs are

handled about and continually referred to, that witnesses have seen that conditions

for years . . . gradually works upon the minds of the jurors because no one can say

that none of these things are so.” Dubois, supra.

See Melissa Whitney, How to improve technical expertise for judges in AI-related

litigation, BROOKINGS INST. (Nov. 7, 2019), archived at https://perma.cc/U8BB-

PQD9 (suggesting a need to have technical advisers to educate judges on technology

and AI-related issues in litigation to ensure that they properly consider evidence);

Judge Herbert B. Dixon Jr., Deepfakes: More Frightening Than Photoshop on

Steroids, A.B.A. (Aug. 12, 2019), archived at https://perma.cc/85NZ-T2QK

______________________________________________________________________________

2021] SILENT NO MORE

133

C. Deepfake Technology

Deepfakes are currently the most advanced form of digital

image manipulation, but they are certainly not the first.

Standard

photoshop technology developed in 1987, and over the course of its

growth has been put to increasingly deceptive use.

Modern examples

include Adobe’s Project VoCo, which makes doctored audio with a ten

minute sample of the subject speaking so realistic that it is referred to

as “Photoshop for the human voice.”

Similar to photoshop,

deepfakes can have harmful consequences on public perception, but

(cautioning about the challenges deepfakes will bring to the courtroom when parties

have conflicting testimony about the authenticity of a video); Debra Cassens Weiss,

Should there be a duty of tech competence for judges? Survey raises questions,

A.B.A. J. (May 10, 2019) [hereinafter Tech Competence for Judges], archived at

https://perma.cc/SYA7-LCDY (quoting a 2019 survey where two-thirds of judges

stated that they need more e-discovery training); Riana Pfefferkorn, ‘Deepfakes’: A

New Challenge for Trial Courts, WASH. STATE BAR ASS’N (Mar. 13, 2019), archived

at https://perma.cc/4B5A-KNN7 (warning how trial courts will need to become apt

at confronting deepfakes).

See Brown, supra note 21, at 90 (summarizing the evolution of photo editing

software). Countries such as England, France, and Brazil all enacted regulation

surrounding photo editing because it is so common. Id. at 93.

See id. at 90 (tracing the development of photoshop back to a PhD student at the

University of Michigan in 1987); Michael Scott Henderson, Note, Applying Tort Law

to Fabricated Digital Content, 2018 UTAH L. REV. 1145, 1148–49 (2018) (discussing

how courts have already had to confront edited photos in the context of child

pornography); Ewing, supra note 13 (noting how Photoshop makes it extremely

simple for the public to manipulate still images); Parry, supra note 27, at 182–83

(emphasizing the ease at which novice photo editors can manipulate images using

Photoshop). “With a moderate amount of expertise” users can significantly alter

photos to add things such as water or snow, open a door, or change what someone is

wearing. Parry, supra.

See Nicholas Mirra, Putting Words in Your Mouth: The Evidentiary Impact of

Emerging Voice Editing Software, 25 RICH. J. L. & TECH. 1, 8 (2018) (explaining

how VoCo creates fake audio). “The VoCo user can individually adjust each

phoneme within any word in the sentence in order to create a sentence that flows as

naturally as a real human statement.” Id. The user can also alter duration and pitch

of words to make the recording more realistic. Id. See also Debra Cassens Weiss,

Eckert Seamans lawyer warns about voice fakers after he nearly wired $9k to

scammer, A.B.A. J. (Mar. 10, 2020) [hereinafter Eckert Seamans], archived at

https://perma.cc/QFC6-WLDE (proffering an example of an attorney falling victim

to a voice-editing technology scam). A lawyer nearly wired $9,000 to a scammer

after receiving a call that he believed was from his son. Id.

______________________________________________________________________________

JOURNAL OF HIGH TECHNOLOGY LAW [Vol. XXI: No. 1

134

that is not the technology’s sole purpose.

The first use of deepfakes

to alter images can actually be attributed to harmless fun on social

media, such as using filters to add features like dog ears to someone’s

face.

In recent years, deepfake technology has transitioned from

innocent use to more problematic purposes.

Technological advances

creating more convincing doctored photos and videos make deepfakes

even more dangerous than previous digital editing technology.

The mainstream term “deepfake” derives from a Reddit user’s

username who first began using the technology to create fake

See Danielle S. Van Lier, The People vs. Deepfakes: California AB 1903 Provides

Criminal Charges for Deepfakes Activity to Guard Against Falsified Defaming

Celebrity Online Content, 43 L.A. LAW. 16, 18 (2020) (arguing that deepfake

technology could have positive impacts on the entertainment industry by making

special effects cheaper and less time-consuming); Chesney & Citron, supra note 9,

at 1769–71 (providing examples of possible beneficial uses of deepfakes in

education, art, and autonomy); Huston & Bahm, supra note 6 (stating that early

deepfakes were “largely used for entertainment purposes.”); Carlos Melendez, It's

All Fun And Games Until Someone Gets Hurt: The Implications Of Deepfakes,

FORBES (Dec. 27, 2019), archived at https://perma.cc/S6QN-4SPD (acknowledging

several uses of deepfakes and noting that “until recently, it was harmless”).

See Melendez, supra note 46 (citing the origins of deepfakes in social media).

See Charlotte Jee, An Indian politician is using deepfake technology to win new

voters, MIT TECH. REV. (Feb. 19, 2020), archived at https://perma.cc/6BST-NDNZ

(reporting on how the president of India’s Bharatiya Janata Party appeared in the first

deepfake video used for political campaigning purposes). The first video showed

the Bharatiya Janata Party president speaking in English, and the second video, a

deepfake, showed him speaking in Haryanvi, the Hindi dialect spoken by the political

party’s target voters. Id. See Melendez, supra note 46 (illustrating examples of

recent uses of deepfake technology to portray American public figures in a negative

light). An example of this is the editing of an interaction between a CNN reporter

and an intern to make it appear as though the reporter attacked the intern by removing

a portion of the tape where the CNN reporter said “pardon me, ma’am.” Id.

See Chesney & Citron, supra note 9, at 1771–86 (warning about the various harms

deepfakes can have on individuals, organizations, and society); Grace Shao, What

‘deepfakes’ are and how they may be dangerous, CNBC (Oct. 13, 2019) [hereinafter

What Deepfakes Are], archived at https://perma.cc/6T9N-5XQ4 (addressing the

danger of deepfakes’ accuracy deceiving people into believing falsities); Engler,

supra note 10 (describing how “the use of this technology to manipulate facial

expressions and speech, or face-swap an individual into a video, has garnered the

greatest concern.”). See also Van Lier, supra note 46, at 18 (warning that deepfakes

can be created “in a matter of hours using an adequately powerful home computer.”).

______________________________________________________________________________

2021] SILENT NO MORE

135

pornographic videos of celebrities.

It was this use of deepfake videos

that originally captured public attention.

The word “deepfake” is

derived from a combination of the phrases “deep learning” and

“fake.”

Deep learning refers to the process of training technology to

become more intelligent by continuously feeding it information.

Deepfakes can create a false representation of events by superimposing

a person’s face on another’s body or by changing the contents of what

a person is saying.

Currently, deepfakes are largely used to create

See Russell Spivak, “Deepfakes”: The Newest Way to Commit One of the Oldest

Crimes, 3 GEO. L. TECH. REV. 339, 339 (2019) (referencing the beginnings of

deepfake technology). An anonymous user on Reddit “superimposed images of

celebrities such as Gal Gadot (Wonder Woman), Masie Williams (Game of Thrones),

and Daisy Ridley (Star Wars) onto the bodies of adult video stars in pornographic

films.” Id. at 339–40. This led to the shocking realization that online users could

create anything they wanted through the manipulation of images. Id. See Rachel

Metz, The number of deepfake videos online are spiking. Most are porn, CNN (Oct.

7, 2019), archived at https://perma.cc/CFX3-9GWV (recounting the rise of

deepfakes in 2017 on Reddit). See also Van Lier, supra note 46, at 17 (distinguishing

the difference between “cheap fakes” and “deepfakes”). “Deepfakes should not be

confused with the more prevalent ‘shallow fakes’ or ‘cheap fakes,’ videos

manipulated through traditional video editing techniques or simply deceptively

labeled to convey a narrative different from that actually depicted in the video.” Id.

See Hayley Duquette, Note, Digital Fame: Amending the Right of Publicity to

Combat Advances in Face-Swapping Technology, 20 SUFFOLK UNIV. J. HIGH TECH.

L. 83, 105 (2019) (detailing how the creation of celebrity deepfake pornography

caught public attention). See also Delfino, supra note 7, at 893–94 (focusing on the

development of “Fake App” to make video editing easier after a Reddit user first

created deepfake pornography). “Before FakeApp's development, the production of

realistic doctored videos was an expensive and arduous process confined to

Hollywood movie studios. FakeApp's creator achieved the goal of ‘mak[ing]

deepfakes’ technology available to people without a technical background or

programming experience.’” Id.

See Metz, supra note 50 (explaining the origin of the word deepfake).

See Harris, supra note 7, at 100 (chronicling how deep learning works in the

context of pornographic images).

See Spivak, supra note 50, at 339 (defining deepfakes); Duquette, supra note 51,

at 104–05 (describing how deepfakes first became controversial through

superimposing celebrities’ faces on pornographic actors); What Deepfakes Are,

supra note 49 (refining further the definition of deepfakes). “Deepfakes refer to

manipulated videos, or other digital representations produced by sophisticated

artificial intelligence, that yield fabricated images and sounds that appear to be real.”

What Deepfakes Are, supra.

______________________________________________________________________________

JOURNAL OF HIGH TECHNOLOGY LAW [Vol. XXI: No. 1

136

pornography, but they are also used in other avenues such as politics.

Examples of political deepfake videos include doctored videos of

Nancy Pelosi appearing drunk at a speaking engagement, Italy’s prime

minster speaking in a hoarse whisper, an Indian politician speaking in

a different dialect, and Barack Obama calling Donald Trump a

“dipshit.”

These examples of foreign and domestic deepfake use

demonstrate the technology’s ability to manipulate global public

perception.

To fully grasp how deepfakes work, it is helpful to understand

that a video is simply a series of still images strung together in a

sequence.

Deepfake technology uses an intelligent algorithm to

See Metz, supra note 50 (offering statistics on how deepfakes are used). “While

much of the coverage about deepfakes has focused on its potential to be a tool for

information warfare in politics, the Deeptrace findings show the more immediate

issue is porn.” Id. Deeptrace said 96% of identified deepfakes featured pornographic

content. Id.

See Jee, supra note 48 (reporting on how an Indian political candidate used a

deepfake video of him speaking in a Hindi dialect to influence a key demographic of

voters); Siddharth Venkataramakrishnan, Can you believe your eyes? How deepfakes

are coming for politics, FIN. TIMES (Oct. 24, 2019), archived at

https://perma.cc/3VQQ-CR5T (delving into various ways deepfakes were recently

used in politics to create false impressions including examples of Nancy Pelosi and

the Italian Prime Minister); Kaylee Fagan, A viral video that appeared to show

Obama calling Trump a 'dips---' shows a disturbing new trend called ‘deepfakes’,

BUS. INSIDER (Apr. 17, 2018), archived at https://perma.cc/VFD4-5HVF (explaining

how a deepfake video of Barack Obama calling Donald Trump a “dipshit” went

viral). Comedian Jordan Peele created the video of Barack Obama calling Donald

Trump a “dipshit” in just fifty-six hours with the help of a professional video editor

to demonstrate the danger behind deepfake technology. Fagan, supra. See also

Sharon D. Nelson & John W. Simek, 96 Percent of Deepfake Videos Are Women

Engaged in Sexual Acts, SLAW (Mar. 25, 2020), archived at https://perma.cc/34ZS-

LBKE (discussing the fact that most deepfakes online are pornography).

See Duquette, supra note 51, at 105 (emphasizing the immorality of using

someone’s likeness to create deepfakes); Venkataramakrishnan, supra note 56

(inferring the power altered videos of politicians can have on public perception of

events); Daniel Thomas, Deepfakes: A threat to democracy or just a bit of fun?, BBC

NEWS (Jan. 23, 2020), archived at https://perma.cc/M643-ER3A (critiquing how

politicians can use deepfakes as a defense to their behavior caught on tape). In 2018,

the governor of Sao Paulo, Brazil, claimed that a video of him engaged in an orgy

was a deepfake as a defense. Thomas, supra. The public was unable to conclusively

refute his claims. Id.

See Mraunac, supra note 11 (summarizing the basics of video evidence). Digital

video cameras capture light and turn it into digital information by stringing a series

______________________________________________________________________________

2021] SILENT NO MORE

137

manipulate these images, otherwise referred to as deep learning.

Two different types of algorithms are used to create deepfake videos:

discriminative and generative algorithms.

Discriminative algorithms

classify data by looking at the subject’s features and assigning it a

category or label.

An example of this would be determining whether

an email message is spam or not.

In contrast to discriminative

algorithms, generative algorithms operate in the reverse: they first

assume the category, and then determine what features make the data

of images along to create a video. Id. Sound waves are captured by a microphone

and then turned into an electrical signal that is stored on a video tape. Id. Although

most cameras record sound and images at the same time, it should be noted that two

separate devices capture sound and images. Id.

See Spivak, supra note 50, at 344 (articulating how deepfake technology is created

through algorithm manipulation); What Deepfakes Are, supra note 49 (defining deep

learning). Deepfake technology uses deep learning, which refers to algorithm

arrangements that are capable of training themselves in order to make independent

decisions. What Deepfakes Are, supra. Deep learning is a “subset of AI.” Id. See

Bernard Marr, What Is Deep Learning AI? A Simple Guide With 8 Practical

Examples, FORBES (Oct. 1, 2018) [hereinafter What is Deep Learning AI?], archived

at https://perma.cc/R5RN-6379 (refining further the definition of deep learning).

Deep learning is when machines become capable of performing tasks that typically

require human intelligence. Id. Deep learning occurs through an algorithm

repeatedly performing tasks and slightly changing them in order to become more

intelligent. Id. Deep learning gets its name because “the neural networks have

various (deep) layers that enable learning.” Id.

See Spivak, supra note 50, at 342–43 (naming the two types of algorithms used to

create deepfake videos). See generally Andrew Ng, Lecture, Generative Learning

Algorithms, STAN. (2015), archived at https://perma.cc/DW6X-LLUJ (analyzing

discriminative and generative algorithms at a high level).

See Spivak, supra note 50, at 342 (explaining the function of discriminative

algorithms). “‘Discriminative algorithms try to classify input data; that is, given the

features of a data instance, they predict a label or category to which that data

belongs.’” Id.

See id. (using the email spam example to demonstrate how discriminative

algorithms work). See also Ng, supra note 60 (setting forth another way to think

about discriminative algorithms using animal classifications).

First, looking at elephants, we can build a model of what elephants

look like. Then, looking at dogs, we can build a separate model of

what dogs look like. Finally, to classify a new animal, we can match

the new animal against the elephant model, and match it against the

dog model, to see whether the new animal looks more like the

elephants or more like the dogs we had seen in the training set.

Id.

______________________________________________________________________________

JOURNAL OF HIGH TECHNOLOGY LAW [Vol. XXI: No. 1

138

more likely to fall into that particular group.

Using the same email

example, if a message is identified as spam, the generative algorithm

attempts to predict what features of the email increase the probability

that it is considered spam.

Discriminative and generative algorithms became more

advanced in 2004 when University of Montreal researcher Ian

Goodfellow created the Generative Adversary Network (“GAN”) to

produce realistic fake photos.

The use of GANs later expanded to

encompass video editing.

GANs simultaneously train discriminative

and generative algorithm models.

A neutral network within the

GAN, called the generator, generates new and artificial images, while

the discriminator network evaluates the authenticity of those images.

See Spivak, supra note 50, at 343 (discussing how generative algorithms work).

“[A] generative model provides a way to generate data that looks like it came from

the dataset. Instead of predicting a label given certain features, it attempts to predict

features given a certain label.” Id.

See id. (applying the spam e-mail message example to generative algorithms).

See id. (crediting Ian Goodfellow with the invention of the GAN, which “pits

[discriminative and generative] algorithms against one other.”); George Lawton,

Generative adversarial networks could be most powerful algorithm in AI,

TECHTARGET (June 6, 2018), archived at https://perma.cc/9EBM-GA63 (examining

why the development of the GAN was important for artificial intelligence research).

See also Jordan Novet, One of Google’s top A.I. people has joined Apple, CNBC

(Apr. 4, 2019), archived at https://perma.cc/3ZMX-JW4F (describing Ian

Goodfellow’s background as a widely respected researcher in artificial intelligence).

The article describes Goodfellow as “the father of an AI approach known as general

adversarial networks, or GANs, [whose] research is widely cited in AI literature.”

Id.

See Spivak, supra note 50, at 344–45 (explaining how GANs have evolved

throughout the years to encompass video editing, including infamous deepfake

pornography).

See id. at 343–45 (outlining the training processes necessary for GANs to train

discriminatory and generative algorithms); Chesney & Citron, supra note 9, at 1760

(emphasizing how powerful GANs are). “The result far exceeds the speed, scale,

and nuance of what human reviewers could achieve. Growing sophistication of the

GAN approach is sure to lead to the production of increasingly convincing deep

fakes.” Chesney & Citron, supra. See generally IAN J. GOODFELLOW ET AL.,

GENERATIVE ADVERSARIAL NETS (Univ. Montreal, ed. 2014) (discussing GANs at a

high technical level).

See Chesney & Citron, supra note 9, at 1759 (defining a neutral network “as a kind

of tabula rasa featuring a nodal network controlled by a set of numerical standards

set at random”); Lawton, supra note 65 (asserting that “[t]he main benefit of GANs

is a neural network only interprets the world through the lens of its training data”);

______________________________________________________________________________

2021] SILENT NO MORE

139

The goal of this training is to eventually make the generator images so

convincing that the discriminator network believes that they belong in

the dataset.

The model repetition allows the GAN to become

“smarter” and make more decisions on its own.

This is the process

that creates deepfake videos.

To summarize deepfake technology in the simplest way

possible, consider this analogy from a professor at the University of

Pennsylvania.

Think of the process for creating deepfakes as a

personal trainer for computer software.

The algorithm compares

images to one another to identify characteristics that it then uses to

Spivak, supra note 50, at 343–44 (explaining how a generator and discriminator

network work together); Mika Westerlund, The Emergency of Deepfake Technology:

A Review, 9 TECH. INNOVATION MGMT. REV. 39, 41 (2019) (developing further how

the generator and discriminator networks interact with each other).

These two networks called “the generator” and “the discriminator”

are trained on the same dataset of images, videos, or sounds. The

first then tries to create new samples that are good enough to trick

the second network, which works to determine whether the new

media it sees is real. That way, they drive each other to improve.

Westerlund, supra.

See Spivak, supra note 50, at 344 (noting the point at which images become a

deepfake). “After enough of this ‘training,’ the algorithm is refined enough to

‘convincingly manipulat[e] video on the fly,’ meaning it will generate images into

each individual video frame such that when played regularly, the video appears

seamless. This process produces a deepfake.” Id. See also Lawton, supra note 65

(discussing how the training process gives GANs a better understanding of the world

over time); What is Deep Learning AI?, supra note 59 (explaining how GANs can

better perform and assess tasks when they are given more data to learn from).

See Lawton, supra note 65 (highlighting how GANs make it possible for “‘AI [to

generate] previously unseen data that may be completely novel and unique, but

agrees with the same type and class of data in the real world”).

See Spivak, supra note 50, at 344–45 (illustrating how GANs and repetitious

training create deepfake videos).

With deepfakes, the generator constructs new video frames, while

the discriminator tries to discern whether the frame, with its

superimposed subject, is authentic (say, an actual video frame of

the original actor) or fake (a doctored video frame of the actor in a

compromising position). If the discriminator cannot tell the real

images from the false images, a human may not be able to either.

Id. at 345.

See Demystifying Deepfakes: 3 Truths About AI-Generated Videos, UNIV. PA.

(Nov. 14, 2019) [hereinafter Demystifying Deepfakes], archived

at https://perma.cc/K8T9-TTEF (crediting professor Michael Kearns with the

personal trainer deepfake analogy).

See id. (analogizing the creation of deepfakes to personal training).

______________________________________________________________________________

JOURNAL OF HIGH TECHNOLOGY LAW [Vol. XXI: No. 1

140

create a fake image.

The computer continues to identify the fake

image from the real one, and each time it identifies the fake, the next

false image that the computer creates is more authentic, or in “better

shape” under the personal trainer analogy.

D. Identifying Deepfake Videos

Before deepfakes became as sophisticated as they are today, it

was fairly easy to spot a doctored video.

One could look for simple

tells in a video suggesting that it was fake such as irregular blinking,

differences in quality of certain segments, and unnatural movements.

More complex and modern technology is making it increasingly

difficult to spot such simple changes.

Even if it is possible to identify

alterations with the naked eye, doing so would require familiarity with

the video subject’s typical mannerisms in order to notice deviations

that would expose the video as a deepfake.

The concern with the

See id. (describing how GANs work in simpler terms).

See id. (noting that computers train themselves on how to create deepfake videos).

See Bernard Marr, The Best (And Scariest) Examples Of AI-Enabled Deepfakes,

FORBES (July 22, 2019) [hereinafter Examples of AI-Enabled Deepfakes], archived

at https://perma.cc/F2RR-LHEY (detailing examples of deepfakes since 2017).

Some of the earliest examples of deepfakes occurred in 2017 when a Reddit user

superimposed celebrities’ faces on porn actors to create deepfake pornography. Id.

These uses of deepfake technology were far less sophisticated than later uses such as

the University of Washington-created deepfake editing the contents of Barack

Obama’s speech. Id.

See Connor Levesque, Deepfakes Explained: What, Why and How to Spot Them,

LEXLYTICS (Mar. 7, 2019), archived at https://perma.cc/MZL7-63WB (outlining

factors to consider in determining whether a video is a deepfake); What Deepfakes

Are, supra note 49 (listing ways to spot a deepfake video). Viewers can also look

for tells such as a face appearing too smooth, eyes and ears not matching, and

fuzziness in the video. What Deepfakes Are, supra.

See Chesney & Citron, supra note 9, at 1759 (stressing the effect technological

advances have on identifying deepfakes); What Deepfakes Are, supra note 49

(emphasizing the increase in access and sophistication of deepfake videos). “While

the detection of doctored audio and video was once fairly straightforward, the

emergence of generative technology capitalizing on machine learning promises to

shift this balance.” Chesney & Citron, supra.

See Adrienne LaFrance, The Technology That Will Make It Impossible for You to

Believe What You See, ATLANTIC (July 11, 2017), archived at

https://perma.cc/2MU9-SX6R (warning how many people may not even think to

question a video’s authenticity if it is realistic enough). People may be unaware of

______________________________________________________________________________

2021] SILENT NO MORE

141

public’s inability to detect deepfakes is prompting many researchers to

generate deepfake identification tools, such as databases of identified

deepfake videos and applications to verify digital media authenticity.

However, the number of researchers working on identifications tools

is severely outpaced by the amount of fake videos created, with some

experts estimating that the ratio of deepfake creators to researchers is

as high as one hundred to one.

the deepfake identification databases to verify a video’s authenticity, and people

“shar[ing] misinformation unintentionally will likely exacerbate the increasing

distrust in experts who can help make sense of things.” Id. See David Ingram &

Jacob Ward, How do you spot a deepfake? A clue hides within our voices,

researchers say, NBC NEWS (Dec. 14, 2019), archived at https://perma.cc/XT7J-

VQGV (commenting on the public difficulty in spotting a deepfake video). Members

of the public can be easily misled by content if are unfamiliar with the person

speaking in the video. Id. Examples of what this may look like include Elizabeth

Warren’s head movements while public speaking and Barack Obama’s frowning tell

when he delivers bad news. Id.

See Davey Alba, Tool to Help Journalists Spot Doctored Images Is Unveiled by

Jigsaw, N.Y. TIMES (Feb. 4, 2020), archived at https://perma.cc/B9M5-89JS

(reporting on how a new tool called “Assembler” will enable reporters to spot

doctored images through “analyz[ing] an image and highlight[ing] where it thinks

those [signs of manipulation] are.”); Amer, supra note 13 (discussing research at

DeepTrace Labs and the University of California to combat deepfake videos); Press

Release, PR Newswire, Tamperproof Media Startup Announces Seed Round and

Company Launch; Combats Digital Fraud And Deepfakes With Data Validation

Platform (Mar. 18, 2020) (on file with author) [hereinafter Attestiv, Inc.]

(announcing the development of Attestiv, Inc., whose “platform uses AI and

blockchain technology to validate and secure the authenticity of digital media”);

What Deepfakes Are, supra note 49 (describing initiatives to identify and detect

deepfakes). Microsoft and Facebook are detecting and removing deepfake videos

and are collaborating with universities to create a database of fake videos. What

Deepfakes Are, supra. See also Elizabeth Culliford, Facebook, Microsoft launch

contest to detect deepfake videos, REUTERS (Sept. 9, 2019), archived at

https://perma.cc/9TDC-52SK (reporting on how Facebook invested $10 million in a

Deepfake Detection Challenge). See generally SUPASORN SUWAJANAKORN ET AL.,

SYNTHESIZING OBAMA: LEARNING LIP SYNC FROM AUDIO (Univ. Wash., ed. 2017)

(introducing the first major study conducted on deepfakes to warn the public about

the technology’s power to influence perception).

See Harwell, supra note 13 (providing estimates on the number of researchers

versus the number of deepfake creators); Amer, supra note 13 (describing the fight

to create deepfake detection technology as a “‘cat and mouse game’” with deepfake

creators); Chesney & Citron, supra note 9, at 1762 (cautioning that “[f]or better or

worse, deep-fake technology will diffuse and democratize rapidly.”). See also

Nelson & Simek, supra note 56 (articulating how the four biggest deepfake sites

received “a combined 134 million views”).

______________________________________________________________________________

JOURNAL OF HIGH TECHNOLOGY LAW [Vol. XXI: No. 1

142

III. Premise

A. Video Evidence and Courtroom Perception

With the rising prevalence of technology in the courtroom and

in crime television shows, jurors almost expect to see some form of

photo or video evidence in court.

Even prosecutors emphasize how

powerful video evidence can be in the courtroom.

Video evidence

can visually tell the jurors what happened in a case as well as have an

important emotional effect.

Some scholars compare the weight that

jurors give to video evidence to the “CSI effect,” which is the concept

that the commonality of police body cameras and video surveillance

leave jurors surprised—and perhaps disappointed—if there is no video

evidence offered at trial.

See GLOBAL JUSTICE INFORMATION SHARING INITIATIVE, VIDEO EVIDENCE: A

PRIMER FOR PROSECUTORS 1–2 (Bureau Just. Assistance, ed. 2016) (discussing the

impact of video evidence in the courtroom). Some estimates say that video evidence

is used in as many as 80% of criminal cases. Id. It is also noted that juries are

influenced by what they see on television and expect to see video evidence more

frequently as a result. Id.

See John Schwartz & Katie Zezmia, With Video Everywhere, Stark Evidence Is on

Trial, N.Y. TIMES (Dec. 8, 2010), archived at https://perma.cc/5GCH-F829

(reiterating how videos are typically a “main feature” of many cases). A Cornell law

professor notes that video evidence is so important that during jury selection

prosecutors often consider how potential jurors will able to process video evidence

in highly emotional cases. Id.

See id. (contrasting different opinions on the use of video evidence in court and its

impact). Legal experts say the rise of accessible technology such as cell phone

camera will force judges to reevaluate the way they admit video evidence. Id. The

District Attorney in Suffolk County, Massachusetts said it is “a powerful tool for us

in determining the truth.” Id. A defense attorney noted that the power of video

evidence requires defense attorneys to regularly file motions to challenge video

evidence. Id. A juror noted in a sexual molestation case that the video was not

necessary to convict but still had an intense emotional impact on him while

considering the case. Id.

See Paula Hannaford-Agor, Are Body-Worn Cameras the New CSI Effect?, 30 CT.

MANAGER 3 (2015) (inferring that jurors expect to see some form of video evidence

due to the growing prevalence of surveillance and cell phone video in society). See

also Honorable Donald E. Shelton, Juror Expectations for Scientific Evidence in

Criminal Cases: Perceptions and Reality about the “CSI Effect” Myth, 27 W. MICH.

UNIV. COOLEY L. REV. 1 (2010) (noting that jurors who watch CSI have higher

expectations of seeing some form of forensic evidence in criminal cases). A study

______________________________________________________________________________

2021] SILENT NO MORE

143

Disagreements can arise over what occurred even with

untampered video evidence.

A study conducted by The New York

Times explored perception disputes through reporters demonstrating

different points of view of the same incident by filming a staged

interaction from different angles.

They compared the footage taken

by two police officers’ body cameras with the footage taken at the

same time by a reporter at a different angle.

Each of the officers’

body cameras made it appear as though they had an altercation,

whereas it was evident from the reporter’s camera angle that the

officers were dancing with each other.

Another illustration of the

complexities of perception was seen in Scott v. Harris, where

dashboard video evidence played a large role in the United States

Supreme Court’s holding that a police officer used unjust force in a car

chase and caused the plaintiff’s injuries.

In contrast to the majority’s

perception of the dashboard video, Justice Stevens and several lower

court judges disagreed and posited a different version of what they

believe occurred.

revealed 58.3% of jurors expect to see some kind of scientific evidence in every

criminal case, 42.1% expect DNA evidence in every criminal case, and 56.5% of

jurors expect to see fingerprint evidence in every criminal case. Id.

See Williams et al., supra note 22 (showing disagreement over what occurred in a

police body camera video). See also Granot et al., supra note 12, at 96 (describing

a study where participant viewers missed crucial details when asked to focus on a

specific portion of a video). In one study, 46% of people watching a scene with

people passing a ball failed to see distractions coming right through the middle of

the video. Id.

See Williams et al., supra note 22 (explaining the different camera angles used in

the study). One angle was filmed from a police officer’s body camera and the other

was filmed from a reporter’s vantage point where both parties could be seen. Id.

See id. (setting forth how the footage of the incident was compared in the study).

See id. (highlighting the bias that comes from how video footage is obtained and

presented).

In this case, the “struggle” appears to be far more involved than it

actually is because the camera is mounted on the officer’s chest,

producing herky-jerky movements that exaggerate what’s going on.

Even if the camera was on the officer’s glasses or hat, the up-close

footage would be confusing — proof that perspective matters.

Id.

See Scott v. Harris, 550 U.S. 372, 375–76 (2007) (providing the facts of the case).

See id. at 378–81 (detailing the Supreme Court’s view of what occurred on the

videotape evidence). The Court rejects the lower court’s interpretation that the

Respondent was driving with care, holding that it clearly appeared that Respondent

was fleeing from the police. Id. See Granot et al., supra note 12, at 94 (emphasizing

______________________________________________________________________________

JOURNAL OF HIGH TECHNOLOGY LAW [Vol. XXI: No. 1

144

B. Disinformation and its Consequences on Public

Perception

Further complicating the impression of events is the alarming

prevalence of disinformation in the era of fake news.

In the United

States, most people are familiar with, or at the very least have heard

of, the impact of false information on the 2016 presidential election.

To illustrate the effect of disinformation, consider the fact that a false

story claiming the Pope endorsed Donald Trump received over two

million Facebook engagements.

In contrast, the top New York Times

news story that year received just over 300,000 online engagements,

highlighting the power of a sensational false headline in grabbing

public attention.

Many news commentators suggest that Donald

Trump won the 2016 presidential election because of the prevalence

the fact that some Americans had widely different interpretations of the video that

the Supreme Court Justices used to make their judgment). A quarter of 1,300

Americans surveyed disagreed with the Supreme Court Justices’ views on what

occurred in the video. Id.

See Nuñez, supra note 1, at 785 (reiterating the difference between misinformation

and disinformation is that disinformation “is deliberately created to mislead and

influence the public”); Barthel et al., supra note 6 (providing statistics on the

prevalence of sharing fake news). 23% of Americans have shared a fake news story,

knowingly or not, and 64% say that fake news stories cause confusion about current

events. Barthel et al., supra. See also Hall, supra note 1, at 55 (sharing an MIT

Twitter study that found a false story spread “six times more rapidly than a true

story”).

See Danielle Kurtzleben, Did Fake News On Facebook Help Elect Trump? Here's

What We Know, NPR (Apr. 11, 2018), archived at https://perma.cc/PTX9-CMC4

(discussing the various impacts of fake news on the 2016 presidential election).

More than a quarter of voters in the 2016 election visited a fake news site. Id.

See Hannah Ritchie, Read all about it: The biggest fake news stories of 2016,

CNBC (Dec. 30, 2016), archived at https://perma.cc/8MTB-7W8U (recapping the

top fake news stories of the 2016 presidential election).

See Hunt Allcott & Matthew Gentzkow, Social Media and Fake News in the 2016

Election, 31 J. ECON. PERSP. 211, 212 (2017) (focusing on the fact that popular fake

news stories were shared more often on Facebook than popular mainstream news

stories); Lamo & Calo, supra note 5, at 999 (asserting that the prevalence of

disinformation is furthered through bots which “can help spread false or misleading

news or else stoke national strife during a crisis or other salient news event.”); Nelson

& Simek, supra note 56 (reporting that there is “no absence of demand for these

images” as there were 134 million views of deepfake videos on the four most popular

deepfake sites).

______________________________________________________________________________

2021] SILENT NO MORE

145

of disinformation, demonstrating its dangerous effects on the public.

Coupled with the disconcerting rise of false news stories is the

expanding presence of deepfake videos, which are projected to

increase in the coming years.

Deepfakes are gaining so much

traction that companies are forming specifically to create doctored

videos for customers.

Like the effect that false news stories had on

the 2016 presidential election, deepfake videos similarly can have a

massive impact on public perception of events.

Just as people are

See Lamo & Calo, supra note 5, at 998 (citing statistics showing that bot activity

spreading stories on election day was “five-to-one pro-Trump to pro-Clinton”);

Allcott & Gentzkow, supra note 95, at 212 (stating commentators suggested that

Donald Trump won the 2016 election because of the prevalence of fake news

stories); Caitlin Dewey, Facebook fake-news writer: ‘I think Donald Trump is in the

White House because of me’, WASH. POST (Nov. 17, 2019), archived at

https://perma.cc/S3HW-Y3NQ (interviewing a fake news reporter who stresses that

many people take stories at face value); Hannah Jane Parkinson, Click and elect: how

fake news helped Donald Trump win a real election, GUARDIAN (Nov. 14, 2016),

archived at https://perma.cc/9Q8Q-9YBF (emphasizing that millions of people

clicked on political fake news stories shared on Facebook); Max Read, Donald

Trump Won Because of Facebook, N.Y. MAG. (Nov. 9, 2016), archived at

https://perma.cc/DGY8-9Z5K (focusing on the alarming fact that many online users

simply read fake news headlines and discuss them with others as though they are

true).

See Metz, supra note 50 (noting the huge increase in the number of deepfake

videos online counted by a study in October 2019). Deeptrace said there was an 84%

increase in deepfake videos online since its last count in December 2018. Id. See

also Harwell, supra note 13 (comparing the number of researchers creating deepfake

detection technology to the number of people creating deepfake videos). “‘We are

outgunned,’ said Hany Farid, a computer-science professor and digital-forensics

expert at the University of California at Berkeley. ‘The number of people working

on the video-synthesis side, as opposed to the detector side, is 100 to 1.’” Id.

See Delfino, supra note 7, at 893 (assessing the impact of “Fake App” by stating

“[b]efore [its] development, the production of realistic doctored videos was an

expensive and arduous process”); Chesney & Citron, supra note 9, at 1763

(recounting how quickly the public began using “Fake App” to manipulate images

once it appeared online); Michael Andor Brodeur, The future of your face on the

Internet, BOS. GLOBE (Feb. 13, 2020), archived at https://perma.cc/55EC-2FP3

(reporting on how “[o]ur face can be peeled off and deepfaked onto other bodies”

through apps Morphin and Doublicat); Nelson & Simek, supra note 56 (discussing

the easy accessibility to deepfake creation technology through avenues such as the

$50 app called DeepNude); Venkataramakrishnan, supra note 56 (providing that

Deepfakes Web charges two dollars an hour to create deepfake videos).

See Amer, supra note 13 (noting that while “one single incident of a deepfake may

not lead to a permanent distortion of facts” multiple instances will cause public

distrust in photos and videos); Huston & Bahm, supra note 6 (stating that “deepfakes

______________________________________________________________________________

JOURNAL OF HIGH TECHNOLOGY LAW [Vol. XXI: No. 1

146

likely to believe what they read online is true, studies have shown over

and over again that people tend to believe what they see, despite

knowing that videos can misrepresent facts.

100

Even judges can

disagree over legal implications of video evidence to the point where

cases are decided differently on appeal because of conflicting

evidentiary interpretations.

101

C. Legislative Action to Curb Deepfakes

In December 2019, the President signed the first bill

concerning deepfakes into law.

102

The law, a part of the National

Defense Authorization Act for Fiscal Year 2020, requires a report on

foreign use of deepfakes, notification to Congress of foreign deepfake

misuse to influence elections, and the creation of a competition to

encourage deepfake research.

103

A number of other bills regarding

pose ‘an extraordinary threat to the sound functioning of government, foundations

of commerce and social fabric.’”); Venkataramakrishnan, supra note 56 (citing

examples of deepfakes used against powerful public figures such as Matteo Renzi,

Italy’s former prime minister, and Speaker of the House, Nancy Pelosi); Fagan, supra

note 56 (discussing how a deepfake video of Barack Obama calling Donald Trump

a “dipshit” quickly became widely circulated).

100

See Granot et al., supra note 12, at 97–98 (warning how powerful video evidence

can be in convincing people that a fake event occurred). In a study conducted by a

bank where no participants illicitly took money, the bank was still able to convince

participants that they stole money after showing them a doctored video in which it

appeared that they stole. Id. After watching the video, while knowing that they did

not steal, participants would confess to taking money from the bank. Id.

101

See id. at 97 (summarizing McDowell v. Sherrer, 374 Fed. Appx. 288 (3d. Cir.

2010) where judges disagreed on what was demonstrated through video evidence).

In McDowell, a district court judge granted summary judgment for a prison guard

after viewing a video of a prison altercation and finding it conflicted with the

prisoner’s claim that the guard used excessive force against him. Id. An appeals

court reversed the grant of summary judgment, holding that the district court judge

assigned too much weight to what she believed she saw in the video and should have

left the question of excessive force to the jury. Id.

102

See Matthew Ferraro et al., First Federal Legislation on Deepfakes Signed Into

Law, WILMERHALE (Dec. 23, 2019), archived at https://perma.cc/C9CV-FQFG

(announcing that Donald Trump signed the first law concerning deepfakes on

December 20, 2019).

103

See id. (summarizing the report that will be delivered to Congress in 2020 on

deepfakes by the Director of National Intelligence). The report to Congress is

required to include (1) the technological capabilities of foreign countries to create

______________________________________________________________________________

2021] SILENT NO MORE

147

deepfake reporting and research are also pending in Congress.

104

The

most significant is H.R. 3230, otherwise known as the Deepfakes

Accountability Act, which mandates transparency when videos are

falsely edited in order to combat the spread of disinformation and

incorrect perception.

105

The Deepfakes Accountability Act calls for

specific ratifications for tampered videos, such as mandatory inclusion

of digital watermarks and express disclosures that describe how the

video was altered.

106

deepfakes; (2) how disinformation from foreign governments could harm the United

States’ elections; (3) what technology the United States can develop to combat

deepfake attacks; (4) current deepfake capabilities of the United States; (5) an

explanation of what is currently being done regarding deepfakes in the United States;

and (6) recommendations for additional needs to combat deepfakes. Id. See National

Defense Authorization Act for Fiscal Year 2020, S. 1790, 116th Cong. (2019)

(enacted) (authorizing the appropriations and policies for the Department of

Defense).

104

See Defending Each and Every Person from False Appearances by Keeping

Exploitation Subject to Accountability Act of 2019, H.R. 3230, 116th Cong. (2019)

(providing guidelines to regulating and marking tampered videos); Deepfake Report

Act of 2019, S. 2065, 116th Cong. (2019) (requiring the Department of Homeland

Security to report on state of “digital content forgery technology” during specified

periods); Identifying Outputs of Generative Adversarial Networks Act, H.R. 4355,

116th Cong. (2019) (asking for federal support in manipulated media research).

105

See H.R. 3230 (naming proposed legislation by Congress that protects the public

from disinformation spread through deepfakes); Dorfman, supra note 11, at 21

(asserting that the Deepfakes Accountability Act is “[t]he most significant bill in

Congress”). See also Chesney & Citron, supra note 9, at 1758 (highlighting the

issues of subject consent that arise with the creation of deepfakes). “Although deep

fakes can be created with the consent of people being featured, more often they will

be created without it.” Id.

106

See H.R. 3230 § 1041 (a)–(e) (specifying ratifications needed for altered videos

under the Texas law). The new regulations include providing a digital watermark on

the altered image; and an audiovisual disclosure; a visual disclosure; or an audio-

disclosure. Id. See also Daniel Lipkowitz, Article, Manipulated Realty, Menaced

Democracy: An Assessment of the Deep Fakes Accountability Act of 2019, 2020

N.Y.U. J. LEGIS. & PUB. POL’Y QUORUM 30, 31 (2020) (critiquing the reforms

outlined in the DeepFakes Accountability Act). Watermarks can be easily removed

and it is difficult to find the creators of false content. Id. However, the legislation

is a step in the right direction towards regulating deepfakes because it (1) draws a

clear line between criminal and non-criminal deepfakes; and (2) current criminal and

tort law does not adequately address harms caused by deepfakes. Id.

______________________________________________________________________________

JOURNAL OF HIGH TECHNOLOGY LAW [Vol. XXI: No. 1

148

However, there is still an alarming lack of state legislation

pertaining to deepfake video use.

107

This is incredibly dangerous

because accurate public perception of events is extremely vulnerable

when powerful individuals and organizations endorse deepfake

videos.

108

Despite the shortcomings of most states, others realized the

imminent threat of deepfakes and took action.

109

Virginia amended its

law banning the use of images “with the intent to coerce, harass, or

intimidate” another person to encompass falsely created videos in early

2019.

110

Texas then became the first state to amend its Election Code

to explicitly criminalize the creation and distribution of deepfakes

107

See Kenneth Artz, Texas Outlaws 'Deepfakes'—but the Legal System May Not Be

Able to Stop Them, LAW.COM (Oct. 11, 2019), archived at https://perma.cc/VDM3-

T444 (noting the lack of action in the legal system to criminalize deepfake videos);

Dorfman, supra note 11, at 21 (foreshadowing that more states will begin to enact

deepfake legislation when they realize deepfakes can “manipulate financial markets,

slander professional and personal rivals, incite violence and blackmail people falsely

depicted as engaging in unethical conduct.”).

108

See Venkataramakrishnan, supra note 56 (cautioning on the immense power

deepfakes have when shared by public officials). Donald Trump shared an altered

video of Nancy Pelosi on Twitter which received 30,000 retweets and 90,000 likes.

Id. See Lauren Feiner, Facebook says the doctored Nancy Pelosi video used to

question her mental state and viewed millions of times will stay up, CNBC (May 24,

2019), archived at https://perma.cc/ZGZ2-G6S4 (emphasizing the impact of the

deepfake video of Nancy Pelosi on public perception). Rudy Giuliani, Donald

Trump’s attorney at the time, shared the false video and stated “[w]hat is wrong with

Nancy Pelosi? Her speech pattern is bizarre.” Id. The video was slowed down by

75% to make it appear that Pelosi was slurring her words. Id.

109

See Lucas Ropek, Handful of States Begin Legislating “Deepfake” Videos,

GOV’T TECH. (Apr. 30, 2019), archived at https://perma.cc/WFG4-9WGZ (reporting

on some states’ efforts to create legislation criminalizing deepfakes). The Pentagon

announced an effort to research ways to combat false videos and several states began

considering federal legislation. Id. These efforts are largely based out of concern

for false information influencing elections. Id. See also Dorfman, supra note 11, at

21 (predicting that states banning deep fakes altogether rather than labeling them

may face constitutionality issues in the future).

110

See VA. CODE ANN. § 18.2–386.2 (2020) (criminalizing falsely created

pornographic images under Virginia law); H.B. 2678, 2019–2020 Leg., Reg. Sess.

(Va. 2019) (providing the amended language to Virginia’s original law criminalizing

the malicious distribution of pornographic images without the subject’s consent);

Robert Volker & Henry Ajder, Analyzing the Commodization of Deepfakes, 2020

N.Y.U. J. LEGIS. & PUB. POL’Y QUORUM 22, 27 (2020) (crediting Virginia as the first

state to criminalize “nonconsensual, ‘falsely created,’ explicit images and videos . .

. a Class 1 misdemeanor.”).

______________________________________________________________________________

2021] SILENT NO MORE

149

intended to harm a political candidate in September 2019.

111

Under

the Texas law, it is a criminal offense to knowingly post a manipulated

video of a political candidate within thirty days prior to an election.

112

Shortly after Texas, California passed a similar election bill in October

2019.

113

The California legislation makes it illegal for anyone to

knowingly post a deepfake video relating to a political candidate

within sixty days prior to an election.

114

California also amended its

Penal Code in January 2020 to specifically define deepfakes and

outline criminal penalties for the creation of malicious deepfake

videos.

115

111

See Matthew Ferraro, Texas Law Could Signal More State, Federal Deepfake

Bans, LAW360 (Sept. 6, 2019) [hereinafter Texas Deepfake Law], archived at

https://perma.cc/6N2D-T6HY (observing that Texas is first state to enact deepfake

legislation criminalizing deepfake-related conduct). Texas is the first state to enact

legislation banning the creation of deepfake videos and the second state to enact

criminal penalties for the distribution of deepfake videos. Id. See also Volker &

Ajder, supra note 110, at 27 (noting that Virginia was the first state to criminalize

“‘falsely’ created explicit images and videos,” although the legislation did not

explicitly call them deepfakes).

112

See 15 TEX. ELEC. CODE § 255.004(e) (2019) (defining the term deepfake in the

Texas law). “[A] video created” with artificial intelligence that, with the intent to

deceive, appears to depict a real person performing an action that did not occur in

reality.” Id.

113

See Will Fischer, California’s governor signed new deepfake laws for politics and

porn, but experts say they threaten free speech, BUS. INSIDER (Oct. 10. 2019),

archived at https://perma.cc/D2QQ-HF5F (describing the California deepfake

legislation on elections and reactions to it). Critics of the deepfake legislation say it

may hurt free speech principles under the First Amendment because “[t]he law is

overbroad, vague, and subjective.” Id. Assemblyman Bernman countered this idea

by stating, “‘[w]hile the First Amendment gives you the right to say whatever you

want, it does not give you the right to put your words into my mouth, or to use AI

technology to take my body and make it look like I did something I never did.’” Id.

114

See Assemb. B. 730, 2019–2020 Leg., Reg. Sess. (Cal. 2019) (setting forth how

both California’s Code of Civil Procedure and Elections Code were amended

through the legislation). The bill’s protections are active until January 1, 2023. Id.

115

See Assemb. B. 1903, 2019–2020 Leg., Reg. Sess. (Cal. 2020) (defining

deepfakes and acts using them which are criminalized under California law).

[A deepfake is] a recording that has been created or altered in a

manner that it would falsely appear to a reasonable observer to be

an authentic record of the actual speech or conduct of the

individual depicted in the recording . . . [the] bill would also

criminally prohibit a person from preparing, producing, or

developing, without the depicted individual’s consent, a deepfake

depicting sexual conduct.

______________________________________________________________________________

JOURNAL OF HIGH TECHNOLOGY LAW [Vol. XXI: No. 1

150

D. Deepfake Evidence in the Courtroom

In response to growing public awareness and newly introduced

deepfake legislation, legal scholars are finally considering how to

handle deepfake videos in the courtroom.

116

Due to the nature of

deepfakes and their modernity, there is hardly any case law suggesting

how to confront them in court.

117

However, one case—United States

v. Chapman—provides helpful language from an expert witness

authenticating video evidence in court which may be useful to consider

in the context of deepfakes.

118

In Chapman, an undercover police

officer filmed the defendant engaging in drug activity, and at trial the

defense offered two witnesses to establish whether any video

Id. See also Van Lier, supra note 46, at 21 (praising the California legislature for

enacting a criminal deepfake legislation in January 2020).

116

See Mraunac, supra note 11 (explaining the need for the legal industry to respond

to fake video evidence) (alteration in original); Huston & Bahm, supra note 6

(warning that “[c]riminals could use deep fakes to defraud victims, manipulate

markets, and submit false evidence to courts.”). “The legal industry, along with other

sectors, will have to take serious action in order to avoid litigating and living in a

world where audio and video recordings are presumed untrustworthy.” Mraunac,

supra. See also Catherine Stupp, Fraudsters Used AI to Mimic CEO’s Voice in

Unusual Cybercrime Case, WALL ST. J. (Aug. 30, 2019), archived at

https://perma.cc/5CM9-YD92 (providing a high-profile criminal example of a

deepfake scam). A CEO fell victim to believing he spoke with his boss about a

money transfer and mistakenly sent funds to a scammer using deepfake technology

to sound like his boss. Id. This is one example of how even those who work closely

to an individual can fall victim to deepfakes, which draws doubt on how those

removed from a situation may be able to identify a fake video or audio recording.

Id.

117

See Mraunac, supra note 11 (noting a lack of formal guidance on confronting

deepfakes in case law). Although courts are obviously concerned with the reliability

of video evidence, there is little case law actually dealing with manipulated videos

and whether they should be admitted into evidence because of their relative novelty

in the last few years. Id.

118

See U.S. v. Chapman, 804 F.3d 895, 897–98 (7th Cir. 2015) (providing the factual

background of the case). An officer acted as an undercover wearing a Hawk

recording device while engaging with the defendant in several drug transactions. Id.

at 897. Each time the officer began recording, he stated his name and the time. Id.

at 897–98. If there were any gaps in the recording, it was shown in the device’s time

stamp. Id. at 898. After the device was deactivated, the agent returned it to the FBI

office where the data was downloaded onto a DVD through an FBI software

program. Id.

______________________________________________________________________________

2021] SILENT NO MORE

151

tampering occurred.

119

This expert testimony regarding the

verification of the video evidence would fall under the pictorial

evidence theory, discussed in Part II of this Note, which requires that

a witness testify as to the video’s authenticity.

120

The expert witnesses

verified several factors, including that the video was consistent with

FBI procedures, that the data collection time was verified, and that the

individual files were verified.

121

This application of the pictorial

evidence theory allowed the jury to reasonably believe there was little

to no chance that anyone wrongfully tampered with the video

evidence.

122

With minimal case law to guide authentication of video

evidence, legal scholars are also recommending that electronic

discovery professionals and expert witnesses must become

increasingly knowledgeable of video verification to ensure proper data

119

See Chapman, 804 F.3d at 900 (describing the backgrounds of the expert

witnesses). One of the expert witnesses, Dew, had over 10 years’ experience with

video evidence in all forms and owned a video production company. Id. The second

expert witness, Dickey, was an expert in “forensic evaluation and/or authentication

of acoustical/visual media, including the analysis of elemental acoustics and video

images contained therein.” Id.

120

See supra, History, Section A, The Pictorial Evidence Theory.

121

See Mraunac, supra note 11 (quoting language from Chapman where an expert

witness verified the authenticity of video surveillance evidence).

[The expert] noted that the data appeared “consistent with

surveillance recordings commonly

associated with federal law enforcement agencies.” [The expert’s]

procedures included verifying the frame rate of the visual

recording, examining the quality of the imaging, and examining the

audio embedded from the Hawk recording device. In addition,

“[an] overview of HBI/VBI, color scheme, vector/waveform,

embedded data, transitions and other parameters were also

performed.” [The expert’s] report was as follows: “Data integrity

checks verified the files as individually and collectively

continuous. Data creation and download time/date information was

also verified. All creation time and dates are sequentially uniform.”

Therefore, [the expert] concluded: “[the November 2, 2010,

recording] does not contain any anomaly which would question its

authenticity as a continuous and reliable record of the events

existing therein.”

Id.

122

See id. (inferring that the expert’s description in Chapman is an appropriate way

to demonstrate video evidence is authentic).

______________________________________________________________________________

JOURNAL OF HIGH TECHNOLOGY LAW [Vol. XXI: No. 1

152

collection and video analysis.

123

This includes becoming familiar with

certain technical terms used in electronic discovery, such as

metadata.

124

It is also vital for lawyers to educate themselves on how

to confront potential deepfake evidence in the early stages of court

proceedings in order to prevent the evidence from being admitted.

125

Experts recommend that attorneys take steps to verify the video’s

chain of custody, whether similar videos exist, and whether there are

witnesses who can account for what actually happened if video

tampering is suspected.

126

It is also crucial for lawyers to consider the change in public

perception as deepfake knowledge continues to grow, and how this

may impact jurors’ trust in video evidence.

127

Currently, there is

strong favorability towards admitting video evidence, with studies

demonstrating that jurors who hear oral testimony along with video

testimony are 650% more likely to retain information.

128

However,

123

See id. (emphasizing that it will be essential for e-discovery professionals and

expert witnesses to understand deepfakes). There is the potential for additional

authentication requirements to develop as deepfakes become more complex as

deepfake creators use more advanced technology, and clients demand law firms to

become knowledgeable in this area. Id.

124

See Kathryn S. Lehman et al., 5 Ways To Confront Potential Deepfake Evidence

in Court, LAW360 (July 26, 2019), archived at https://perma.cc/GNY8-95BU

(asserting various areas in discovery requests lawyers should be proficient in). See

generally Electronic Discovery Reference Model, EDRM Model, EDRM (2020)

[hereinafter EDRM Model], archived at https://perma.cc/FHF5-4HUH

(demonstrating the various stages in the e-discovery lifecycle and defining important

terms).

125

See Lehman et al., supra note 124 (indicating that early intervention before

evidence reaches the jury is key). If video evidence is admitted, it can be difficult to

attack the evidence’s credibility in court without getting negative reactions from the

jury. Id.

126

See id. (suggesting steps attorneys should take if they suspect that video evidence

is tampered with).

127

See Mraunac, supra note 11 (inferring various challenges that may arise when

jurors understand ways in which video evidence can be tampered with).

128

See Karen Martin Campbell, Roll Tape—Admissibility of Videotape Evidence in

the Courtroom, 26 UNIV. MEMPHIS L. REV. 1445, 1447 (1996) (providing statistics

on how jurors retain videotaped information at trial). Jurors who received visual

testimony were 100% more likely to retain information than jurors who received

only oral testimony. Id. See also Parry, supra note 27, at 185 (citing statistics on

the impact of visual evidence on jurors). “Jurors often are bored, confused, and

frustrated when attorneys or witnesses try to explain technical or complex material”

______________________________________________________________________________

2021] SILENT NO MORE

153

considering the recent increase of deepfake videos and the growing

concern of deepfake influence on the 2020 presidential election and

beyond, public trust in video footage may start to decline.

129

This in

turn may lead to plummeting juror confidence in video evidence absent

a sponsoring witness, even if it is authenticated by a judge prior to it

reaching the jury.

130

IV. Analysis

This section provides three scenarios and recommends how to

confront deepfake video evidence in light of the issues that these

scenarios pose.

131

and having visual aids can help them retain information much better. Id. at 184.

Jurors can retain up to 85% of information visually and in contrast only retain about

10% of what they hear. Id. at 185.

129

See Hall, supra note 1, at 58 (warning that “[t]hese manufactured videos have the

potential to create doubts about every recently released film.”); Huston & Bahm,

supra note 6 (stating that “the mere idea that [deepfakes] could be used to manipulate

public opinion is already causing some to start questioning the validity of real

events”); Thomas, supra note 57 (noting that deepfake technology will inevitably

become “more widely commodified and accessible”); Deepfakes in 2020 elections,

supra note 13 (cautioning how deepfakes may influence voters in the 2020 election);

Demystifying Deepfakes, supra note 72 (emphasizing that growing concern about

deepfake influence in elections draws researchers to develop applications against

deepfakes); Deepfake Detector Wins PennApps XX, UNIV. PA. (Sept. 10, 2019)

[hereinafter PennApps XX], archived at https://perma.cc/4BNT-8NEE (celebrating

the first place winner at hackathon for building a deepfake detector). The University

of Pennsylvania students who built the deepfake detector application did so to guard

against deepfake influence on public perception in elections and to ensure that videos

the public views reflect reality. PennApps XX, supra.

130

See Westerlund, supra note 68, at 42–43 (describing how the public may begin to

distrust authorities deemed reliable in the past because of deepfakes); Mirra, supra

note 45, at 3 (cautioning that courts must be prepared for how “new technology may

threaten existing and well established forms of evidence.”); Hall, supra note 1, at 58

(contending that video may lose its value because “[t]he same accountability video

that brings action can now be abused in a number of ways.”); Harwell, supra note 13

(warning how the public may begin to generally distrust video footage because “[i]t’s

too much effort to figure out what’s real and what’s not”).

131

See infra, IV Sections A, B, & C (illustrating three different scenarios in which

deepfake evidence will impact courtroom procedures).

______________________________________________________________________________

JOURNAL OF HIGH TECHNOLOGY LAW [Vol. XXI: No. 1

154

A. Continuing the Silent Witness Theory if Deepfake

Regulation Legislation is Passed

If Congress enacts federal legislation requiring states to

enforce specific methods of video verification, courts may be able to

continue using the silent witness theory without a dangerously high

risk of admitting unverified false videos into evidence.

132

However,

courts should only allow this application of the silent witness theory if

they can determine with certainty that video manipulation occurred

and that jurors will recognize indicators of tampering such as

watermarks.

133

Courts continuing to apply the silent witness theory

132

See H.R. 3230 § 1041 (a)–(e) (listing ways that federal legislation would require

deepfake video disclosures). The federal legislation would require a watermark on

any portion of the video that was digitally altered, as well as include some form of

disclosure that the video is altered. Id. The audiovisual disclosure would require

that clear text be displayed on the bottom of the video or some verbal statement

detailing that the video is altered; the visual disclosure would require some clear text

description on what was altered; and the audio disclosure would have at minimum

one clear audio statement on what portions are altered in the video. H.R. 3230 §

1041 (c)–(e). See also Madison, supra note 17, at 711 (inferring that courts may be

reluctant to abandon the silent witness theory altogether). Courts tend to abstain

from barring the admission of video or photographic evidence because there is no

sponsoring witness, which is why the silent witness theory was originally adopted.

Id.

133

See Lipkowitz, supra note 106, at 31 (cautioning that the reforms outlined in the

federal deepfake legislation may not be enough to safeguard victims of deepfakes).

“Watermarks are easily removable, and it is extremely difficult to track down the

creators of harmful false content.” Id. See Madison, supra note 17, at 709

(acknowledging that courts are reluctant to admit tampered video evidence). Courts

will only admit video evidence that has been altered “when the jury can understand

the changes in appearance that occurred between the relevant time and the time the

photograph was taken.” Id. See Farrell, supra note 14, at 2 (articulating why the

silent witness theory can be applied in certain circumstances). It is only appropriate

to apply the silent witness theory when the photographic or video evidence can be

verified independent of a sponsoring witness and corroborated by other pieces of

evidence. Id. See also Kalinski, supra note 11, at 815–16 (describing basic ways in

which video evidence can be misleading). Even beyond deepfakes, video can easily

be manipulated to make circumstances appear different than they actually are, such

as angling a camera to make it seem like a head wound is worse than what the victim

actually endured. Id. See also Harwell, supra note 13 (criticizing the current lack

of research in identifying deepfakes); Venkataramakrishnan, supra note 56 (inferring

that deepfakes are becoming more complex and will become more difficult to

identify with the naked eye or ear). With little money to be made in researching how

______________________________________________________________________________

2021] SILENT NO MORE

155

should use more stringent requirements like the seven-prong standard

because of cases’ varying facts and types of visual evidence.

134

This

standard provides the most rigorous verification process and offers

more cohesive court guidance on the implementation of the silent

witness theory.

135

If deepfake evidence frequently comes before a

court, it may become necessary to amend silent witness theory protocol

to require an additional qualifying witness with knowledge on

deepfakes—rather than just one knowledgeable witness on the chain

of custody of that piece of evidence—in order to verify that the video

is an accurate representation of events to the judge.

136

It will also be

to identify deepfakes and considering the serious threat that they pose, much of the

research is being funded by a government program run through the Pentagon.

Harwell, supra.

134

See Dorfman, supra note 11, at 20 (arguing that courts must consider adding new

standards in court to ensure audio and video recordings are authentic because of

deepfakes); Videotape Evidence, supra note 25, § 62 (conceding that there may be a

“revival of certain foundational requirements” like the seven-prong standard because

of the growing ability of the general public to edit videos). See also Lehman et al.,

supra note 124 (outlining an example of an interrogatory to verify the chain of

custody that a lawyer may send to the opposing party).

For the video previously produced, please provide (1) the time,

place, and date of the recording was made; (2) the name and address

of any individual depicted in or present at the time of the recording;

(3) the name and address of any individual under whose direction

and upon whose behold the recording was created; (4) the name and

address of any other individual involved with the creation of the

recording; (5) the steps undertaken by the in deified individuals to

create the recording; and (6) the name and address of any individual

who has had possession or control of the recording (either the

original or a copy) since it was created.

Id.

135

See Farrell, supra note 14 (listing the wide variety of jurisdictional practices in

applying the silent witness theory); Madison, supra note 17, at 713 (explaining how

complications can arise in verifying the authenticity of evidence when the origins of

photographs cannot be verified). Madison suggests that courts should especially

apply a more rigorous standard when photos of videos are taken on less sophisticated

equipment such as a personal camera. Madison, supra.

136

See Harwell, supra note 13 (quoting a leading deepfake researcher stating that

“people do need to understand that video may not be an accurate representation of

what happened”); Williams et al., supra note 22 (inferring the importance of a

qualifying witness). Even with untampered video evidence there can be multiple

interpretations, which demonstrates the importance of having a qualifying witness

testify at trial. Williams et al., supra. But see Madison, supra note 17, at 714–15

(arguing that witnesses have imperfect memories). Even under the pictorial evidence

______________________________________________________________________________

JOURNAL OF HIGH TECHNOLOGY LAW [Vol. XXI: No. 1

156

essential to implement training for judges to ensure that they

understand the implications of technology and indicators of evidence

tampering.

137

Courts will be able to measure the threat of deepfake

evidence by considering the number of states adopting deepfake

legislation, identifying major companies that are enacting formal

policies on deepfakes, and tracking the number of individuals who

were subjected to criminal penalties outlined in deepfake statutes.

138

However, given the complexities of deepfake technology and the

potential for jurors to become inherently skeptical of photographic and

video evidence in general, it is preferable for courts to abandon the

silent witness theory and instead adopt the pictorial evidence theory.

139

theory, the verifying witness may not have a completely accurate representation of

events if significant time has elapsed between the event and trial. Id.

137

See Whitney, supra note 42 (declaring a need to train judges about technology to

ensure they properly admit evidence); Tech Competence for Judges, supra note 42

(providing statistics that two-thirds of judges say they need more training on e-

discovery); Dixon, supra note 42 (illustrating a scenario where the authenticity of

cell phone video evidence could be questioned by parties in litigation).

(1) [A] party offers an exhibit of a cell phone video disclosed during

discovery that supports the offering party’s position of an

agreement reached by the two parties, (2) the offering party will

testify affirmatively concerning the authenticity and accuracy of the

video, and (3) the opposing party will testify that he never said the

words portrayed in the video.

Dixon, supra.

138

See H.R. 3230 § 1041 (f)(1)–(f)(2) (naming the criminal and civil penalties for

violating the deepfake law). Criminal penalties to disclose tampering with a video

could result in 5 years imprisonment, a fine, or both. Id. § 1041 (f)(1). Civil penalties

include a fine up to $150,000. Id. § 1041 (f)(2). See Texas Deepfake Law, supra

note 111 (inferring that more states will create legislation banning or criminalizing

deepfakes). The Texas law also provides an avenue for private individuals to seek

injunctive relief if they were the victims of a deepfake video, which will infiltrate

the courts with more deepfake legal controversies. Id. See Ingram & Ward, supra

note 79 (inferring that more social media companies may create regulations

surrounding deepfakes). Two senators wrote a letter to social media companies

emphasizing that they must take a strong policy stance against deepfake videos. Id.

139

See Dorfman, supra note 11, at 21 (providing potential situations of juries

becoming skeptical of police body cameras and confessions in criminal cases

because of deepfakes); Huston & Bahm, supra note 6 (warning that deepfakes are

already causing people to question “the validity of real events and un-doctored

video.”); Amer, supra note 13 (cautioning that “[w]e’re at a moment when we can

literally no longer believe our own eyes — seeing is not necessarily believing.”);

LaFrance, supra note 79 (quoting an expert who warns that “people will not believe

______________________________________________________________________________

2021] SILENT NO MORE

157

B. Attacking the Silent Witness Theory if No Deepfake

Regulation Legislation is Passed

If no deepfake regulation legislation is passed and courts

continue to apply the silent witness theory, lawyers will need to

become zealous advocates and ensure they convince a judge that there

is no fathomable way video evidence should be admitted absent a

qualifying witness testifying.

140

In order to accomplish this, lawyers

will need to gain a complex understanding of how images can be

potentially altered through deepfake technology.

141

If a lawyer

videos, just like how we do not believe photos once we’re aware that tools like

Photoshop exist”); Harwell, supra note 13 (emphasizing that deepfakes may lead to

public denial of legitimate video). For example, Donald Trump said in 2016 that a

leaked video of him boasting about assaulting women was doctored, leading many

members of the public to falsely believe that was true. Harwell, supra. See Farrell,

supra note 14 (defining the pictorial evidence theory). The pictorial evidence theory

requires that a witness confirm the evidence is a “fair and accurate portrayal” of what

they saw. Id. See Madison, supra note 17, at 707 (explaining that proponents of

evidence authenticate it “by showing that the evidence accurately represents its

subject.”). Authentication can be easily verified by a sponsoring witness who is

familiar with the subject matter of the evidence. Id. at 707–08.

140

See Lehman et al., supra note 124 (describing how lawyers must attack the chain

of custody of opposing counsel’s video evidence if they suspect that it is a deepfake).

Lawyers will need to demonstrate that there is evidence of tampering in any way

possible, including identifying and locating possible witnesses who could add

helpful information about the chain of custody of a video. Id. An example of this is

locating a witness who notes issues with the metadata or a recording. Id. See

Mraunac, supra note 11 (highlighting the importance of a qualifying witness

testifying that the image is an accurate representation of what occurred). A Reuters

reporter edited smoke after an airstrike in Lebanon to make the damage appear worse

than it actually was. Id. The airstrike editing demonstrates the importance of having

a qualifying witness to verify the true extent of events. Id. Even a deepfake

researcher at the University of Wyoming has commented on the fact that at times, he

distrusts videos that are real, and believes that videos people send him are fake

because the video quality is so high. Id. See Williams et al., supra note 22

(demonstrating the importance of a qualifying witness with video evidence).

Individuals would have a misunderstanding of what occurred without the journalists

posting two different video angles showing that the police officers were actually

dancing and not fighting. Id. See also Thomas, supra note 57 (reporting how in

2018 Sao Paulo’s married governor asserted that a video of him having an orgy was

a deepfake to diminish its value). The allegations could neither be confirmed nor

denied. Id.

141

See Dorfman, supra note 11, at 23 (asserting that detection technology cannot be

the only solution to confronting deepfakes in court, and that lawyers must understand

______________________________________________________________________________

JOURNAL OF HIGH TECHNOLOGY LAW [Vol. XXI: No. 1

158

representing a client is unable to undertake this on their own, they will

need to hire an expert witness to testify as to why the evidence should

not be admitted.

142

The downside to this approach will be higher costs

to the client because expert witnesses can be extremely expensive and

retaining them requires more attorney time that the client is billed

for.

143

However, this practice may become necessary in cases that

hinge on questionable video evidence and its admission to the jury.

144

Further, lawyers cannot assume that all judges fully understand the

technology behind deepfake imagery, so they must use all resources

how audio and video recordings can be altered); Spivak, supra note 50, at 351

(inferring how complex understanding deepfake technology can be and describing

technologies that run parallel to deepfake video technology); Lehman et al., supra

note 124 (describing how video evidence is admitted under the silent witness theory

after a “strong showing of authenticity and competency, including proof that the

evidence was not altered.”).

[A] skilled litigator should rest assured that there will be some

opportunity to attack a deepfake, even if presented pursuant to the

silent witness theory. A knowledgeable witness who has identified

an issue with, for example, the metadata of a recording could prove

especially fruitful here, helping you to convince the judge that the

other side needs to call someone to lay the chain of custody.

Lehman et al., supra. See also EDRM Model, supra note 112 (defining various e-

discovery terms that lawyers will need to become familiar with in order to challenge

potentially inauthentic video and photographic evidence).

142

See Mraunac, supra note 11 (asserting that “expert testimony becomes relevant”

when the authenticity of video evidence is questionable); Lehman et al., supra note

124 (articulating that a lawyer has a duty to seek expert opinion if they “have reason

to suspect the video was fabricated but lack the technical knowledge to reach a

conclusion”); What Deepfakes Are, supra note 49 (inferring the difficulty in staying

abreast of deepfake technology if one is not a deepfake-specific researcher).

143

See Dorfman, supra note 11, at 20 (warning of the “steep costs” of litigation that

may occur because of deepfakes); Lehman et al., supra note 124 (inferring that

experts are expensive to hire and the cost should be carefully evaluated in light of

the value of the case).

144

See Scott v. Harris, 550 U.S. 372, 378–81 (2007) (illustrating the huge impact one

piece of video evidence can have in the outcome of a case); Parry, supra note 27, at

185 (warning how “there is absolutely no way the average juror could tell the

difference between a doctored and a pure photo.”); Granot et al., supra note 12 at 94

(inferring the immense power visual evidence has in court cases through jurors

evaluating video with “naïve realism”). See also Lehman et al., supra note 124

(providing advice on what lawyers should do if questionable video evidence is

admitted). Lawyers should aggressively attack the chain of custody through cross-

examination if questionable video evidence is admitted. Id. Once jurors realize the

authenticity of the video is in question, they will consider whether or not they should

assign serious weight to it in deliberations. Id.

______________________________________________________________________________

2021] SILENT NO MORE

159

available to communicate the technology’s potentially grave

implications on a case.

145

The lawyer will need to question the

authenticity of the opposing party’s video evidence and convince the

judge that, at a bare minimum, the opposing party must present a

witness at trial for cross-examination about the process used to collect

the admitted evidence.

146

Some may argue that this is too excessive of

a process for seemingly valid evidence, but given the expansion of

deepfake videos into originally unimagined areas, it is essential for

lawyers to be vigilant for their clients when encountering visual

evidence.

147

145

See Tech Competence for Judges, supra note 42 (quoting statistics demonstrating

that two-thirds of judges feel they need more education on e-discovery); Harris,

supra note 7, at 110 (noting how “[d]espite the transformative nature” of deepfakes

judges are relatively inexperienced in dealing with this type of technology);

Pfefferkorn, supra note 42 (cautioning that deepfakes could infiltrate trial courts).

“Points where deepfakes could infect a court case run the gamut from clients who

fabricate evidence in order to win, to fake videos ending up in archives that have

historically been considered trustworthy.” Pfefferkorn, supra.

146

See Lehman et al., supra note 124 (recommending that litigators should request

an authentication witness from the opposing party that may be cross-examined). The

main idea behind the cross-examination is to call the jury’s attention to the fact that

there is a disagreement over the evidence’s authenticity. Id. See also Delfino, supra

note 7, at 895 (stating that “[t]o protect victims of deepfakes and to prevent the

negative societal consequences they cause, the laws need to keep pace with this

technology.”); Huston & Bahm, supra note 6 (listing ways in which deepfake

technology can be abused to manipulate public perception); Harwell, supra note 13

(indicating that researchers “remain vastly overwhelmed” as deepfake technology

advances); Goggin, supra note 8 (declaring the widespread use of deepfakes and that

researchers cannot keep up with new technologies used to create them); Stankiewicz,

supra note 10 (emphasizing how difficult it will become to identify deepfake videos

in just a few months).

147

See Chesney & Citron, supra note 9, at 1763 (cautioning how accessible photo

and video manipulation technology is to the general public); Harris, supra note 7, at

123–24 (warning about the dangers of hyper-realistic deepfakes and the need for the

legal system to take action); Stupp, supra note 116 (providing an alarming case

where a CEO was fooled into wiring millions of dollars to a scammer using deepfake

voice recordings); USPTO, supra note 7 (articulating the growing problem of fake

patent and trademark applications); Jee, supra note 48 (reporting on the use of

deepfake technology by an Indian political party to influence voters in an election).

“Deepfakes will become so life-like that they will be indistinguishable from actual

videos. This is an inevitable consequence of artificial intelligence and machine

learning technologies. The law should be equipped to handle this impending

problem.” Harris, supra. See also Ingram & Ward, supra note 79 (commenting on

the scale of the deepfake analysis problem by noting that even Facebook is

outsourcing research to combat fake videos on its site).

______________________________________________________________________________

JOURNAL OF HIGH TECHNOLOGY LAW [Vol. XXI: No. 1

160

C. Adopting the Pictorial Evidence Theory if No

Deepfake Regulation Legislation is Passed

Absent significant deepfake legislation, courts should adopt the

pictorial evidence theory to combat heightened public skepticism of

photographic and video evidence.

148

There have already been

instances of high-profile criminal activity involving the use of

deepfakes, such as the hackers who used the technology to impersonate

a parent company executive and convince one of its subsidiaries to

transfer nearly $250,000 into a bank account controlled by the

perpetrators.

149

The Chief Executive Officer who transferred the funds

was under the genuine belief that he was speaking with the proper

executive and participating in an authorized transaction.

150

This

example demonstrates just one of the potential threats posed by

deepfakes to public trust in audio and video recordings.

151

Because

growing media coverage about the dangers of deepfakes will likely

lead the public—which includes jurors—to view videos with

skepticism, courts should consider adopting the pictorial evidence

148

See Harwell, supra note 13 (stressing that hysteria and fake videos will change

how the public considers videos and photos); Goggin, supra note 8 (noting how the

deepfake movement has grown substantially since its initial rise in 2017);

Stankiewicz, supra note 10 (emphasizing that everyday people will have access to

“perfectly-real” deepfake videos in six months to one year). See also Delfino, supra

note 7, at 890 (highlighting the alarming ease at which deepfake content can be

created). “Deepfake technology has evolved so quickly that an app designed to

create deepfakes is now widely available. The app lowers the technical threshold

required to create such images and videos, which will likely make . . . [them] . . .

more prevalent.” Id. See also USPTO, supra note 7 (providing details about a new

reporter program created by the United States Patent and Trademark Office).

Artificially generated images are becoming so frequent with patent applications that

the USPTO created a new program for the public to report any images they identify

as fake. Id.

149

See Stupp, supra note 116 (describing deepfake criminal activity that led to

money being unknowingly transferred to a criminal).

150

See id. (emphasizing that the employee transferred the money in good faith). The

audio in this case was so convincing that it fooled even someone who worked closely

with the person that the deepfake was imitating. Id.

151

See Mirra, supra note 45, at 16 (warning about the impact voice alteration

software can have on verdicts in criminal cases). “The entire course of a defendant's

life could be altered with some quick changes made on [voice editing software] to a

voice recording that provides a pivotal piece of evidence at trial.” Id. See also Eckert

Seamans, supra note 45 (detailing a scam where a lawyer father almost wired $9,000

to a scammer believing that it was his son).

______________________________________________________________________________

2021] SILENT NO MORE

161

standard for all video evidence to ensure that jurors properly consider

the weight of evidence.

152

The pictorial evidence theory requires a

witness to testify before the jury that the photographic or video

evidence is a fair and accurate representation of what occurred.

153

Having the witness on the stand to verify that the visual evidence is

accurate will allow an initially skeptical jury to view the evidence as a

true image and not question whether it was artificially generated.

154

At a minimum, courts should always require a witness to

authenticate the process under which the video was obtained before

152

See Westerlund, supra note 68, at 42–43 (cautioning that the public will become

skeptical of normally trustworthy authorities because of deepfakes); Amer, supra

note 13 (noting the role that the media plays in directing public attention to

deepfakes); LaFrance, supra note 79 (alleging that deepfakes are dangerous because

“[p]eople are already fooled by doctored photos, impostor accounts on social media,

and other sorts of digital mimicry all the time”); Engler, supra note 10 (asserting that

deepfakes are dangerous to public trust in digital content). “Deepfakes pose a

significant problem for public knowledge. Their development is not a watershed

moment—altered images, audio, and video have pervaded the internet for a long

time—but they will significantly contribute to the continued erosion of faith in digital

content.” Engler, supra. See also FED. R. EVID. 403 (providing circumstances under

which a judge may decide to exclude relevant evidence). “The court may exclude

relevant evidence if its probative value is substantially outweighed by a danger of

one or more of the following: unfair prejudice, confusing the issues, misleading the

jury, undue delay, wasting time, or needlessly presenting cumulative evidence.” Id.

See also Tanford, supra note 14 (illustrating reasons why evidence has probative

value and reasons probative value can be substantially outweighed by unfair

prejudice).

153

See Farrell, supra note 14, at 2 (explaining how witness testimony is used to

authenticate events illustrated in photographic and video evidence under the pictorial

evidence theory).

154

See U.S. v. Chapman, 804 F.3d 895, 900–01 (7th Cir. 2015) (providing an

example of how courts are already willing to allow multiple experts if there is a

showing that evidence was tampered with). Chapman introduced two experts to

testify as to the authenticity of the video expert and wished to call a third. Id. at 900.

The lower court denied the motion, but “suggested that it might have considered the

appointment if Chapman had indicated any flaw in the software or system the

government used to transfer the recording from the Hawk recording device onto the

DVD.” Id. at 901. See Engler, supra note 10 (warning of a “threefold” effect on

public perception of information because of deepfakes). As a result of deepfake

technology, the public will: (1) have a “visceral reaction” to fake information and

spread it more quickly; (2) have more difficulty in ascertaining what stories are true;

and (3) accuse true sources of being fake because of disagreeable content. Id. See

also USPTO, supra note 7 (describing how the USPTO is becoming skeptical of

trademark applications). The USPTO is relying on the public to draw attention to

patent or trademark applications that it believes are artificially generated because

there has been such an influx of fake applications. Id.

______________________________________________________________________________

JOURNAL OF HIGH TECHNOLOGY LAW [Vol. XXI: No. 1

162

the jury, even in typically routine circumstances such as surveillance

videos.

155

Critics may argue that always applying the pictorial

evidence theory is an unnecessarily high standard because most

deepfakes are pornographic.

156

However, with an over 80% increase

in deepfakes in just one year and recent uses of deepfakes to influence

business and politics, it is dangerous to assume deepfakes will not

become a legal problem, especially in high-profile cases.

157

People are

so influenced by video evidence that they can be convinced that they

committed an act that they did not do.

158

One study conducted by a

bank had participants view doctored videos of themselves stealing

money.

159

After watching the doctored video, the participants would

confess to taking money, despite knowing that they had not stolen

155

See Parry, supra note 27, at 186 (calling upon courts to issue stricter evidence

admission standards because of the ease at which digital evidence can be altered);

Spivak, supra note 50, at 342 (emphasizing that Photoshop technology makes the

public skeptical of visual evidence). See also Chapman, 804 F.3d at 900 (using

helpful language in considering the value of expert witness testimony in

authenticating the process by which evidence is collected).

156

See Metz, supra note 50 (stating that 96% of the deepfakes that Deeptrace found

online were pornography). But see Attestiv, Inc., supra note 80 (announcing how

the company is launching an application to detect fake media “[i]n a world rampant

with digital fraud and rapidly emerging deepfake technology.”).

157

See Hall, supra note 1, at 75 (forewarning that “[d]eepfake videos will only

become more of a problem.”); Henderson, supra note 44, at 1148–49 (contending

that courts already confront doctored photographic evidence in the context of child

pornography cases); Huston & Bahm, supra note 6 (emphasizing that confronting

deepfakes will require “a whole of society approach” where multiple organizations

and entities proactively consider how to handle the technology); Stupp, supra note

116 (discussing a case where a CEO wired millions to a scammer using deepfake

voice recordings to pose as his supervisor); LaFrance, supra note 79 (cautioning on

the danger of deepfakes influencing the public by stating “[i]magine the confusion

that might surround a convincing video of the president . . . say[ing] something he

never actually said.”); Goggin, supra note 8 (listing incidents of deepfake videos

targeting politicians Nancy Pelosi, Barack Obama, and Alexandria Ocasio-Cortez);

Metz, supra note 50 (providing statistics from the Deeptrace study stating that there

was an 84% increase in deepfake videos since December 2018); Fagan, supra note

56 (reporting on how a deepfake video of Barack Obama calling Donald Trump a

“dipshit” quickly reached millions of people); Nelson & Simek, supra note 56

(warning about the easy access the public has to deepfake creation technology

through apps like DeepNude).

158

See Granot et al., supra note 12, at 97–98 (emphasizing the impact video evidence

had in convincing participants they committed a crime in a study conducted by a

bank).

159

See id. (illustrating the background of the bank study).

______________________________________________________________________________

2021] SILENT NO MORE

163

anything.

160

If deepfake videos can change peoples’ minds about what

they did themselves, it only becomes more dangerous when used by

jurors to evaluate situations that they were not a part of.

161

Due to First

Amendment concerns, it is unlikely that deepfakes will ever be banned

entirely; therefore, it is of the utmost importance that courts adapt to

this technological threat before it infiltrates the legal system and

creates widespread juror doubt in video evidence.

162

V. Conclusion

Misuse of deepfake technology is an imminent threat to

society, and there is no doubt it will eventually impact the courtroom.

The number of researchers creating tools to identify doctored videos is

severely outpaced by the number of people creating them. Deepfake

videos are projected to quickly become so convincing that they will be

unidentifiable with the naked eye. It is of the utmost importance that

lawyers and the legal system as a whole consider how to handle this

threat before it undermines the justice system. While Congress is

attempting to take some action, the legal system should not put faith in

legislation alone. More jurors will become aware of deepfakes and

without proactive judicial safeguards in place, they will resultantly be

more skeptical of visual evidence in court proceedings. Allowing the

continuous inconsistent application of the silent witness theory across

jurisdictions will only fuel the growing mistrust of visual evidence.

Courts should adopt the pictorial evidence theory and abolish use of

160

See id. (stating the results of the bank video study).

161

See id. (inferring how dangerous visual evidence can be in changing peoples’

perceptions of events).

162

See Harris, supra note 7, at 128 (emphasizing the need for the legal system to be

proactive in its dealings with deepfakes because “technology is only going to

improve”); Smith-Roberts, supra note 5, at 123 (articulating how difficult it is to

restrict “even false speech” in the United States because of First Amendment

protections); Hall, supra note 1, at 62 (explaining how “[m]any deepfake videos

would be protected by the First Amendment as free expression under the defense of

parody or satire . . . .”); Spivak, supra note 50, at 356–64 (discussing First

Amendment concerns that deepfake legislation raises). First Amendment challenges

to deepfakes will likely prevail, unless the deepfake falls under the category of

obscenity or child pornography. Spivak, supra. See Feiner, supra note 108

(contrasting the tension between free speech and the threat of deepfakes). Facebook

stated the following on fake videos being posted on its site: “‘[t]here’s a tension here:

we work hard to find the right balance between encouraging free expression and

promoting a safe and authentic community . . . . ’” Id.

______________________________________________________________________________

JOURNAL OF HIGH TECHNOLOGY LAW [Vol. XXI: No. 1

164

the silent witness theory in order to ensure that admitted video

evidence is authentic and that jurors place appropriate weight in video

evidence. Deepfakes threaten the foundation that our justice system is

built upon—that evidence leads to the truth and a just outcome.