Emerging Ethical and Legal Issues Related to the Use of Artificial

Emerging Ethical and Legal Issues Related to the

Use of Artificial Intelligence (AI), Automatic

Speech Recognition (ASR), Voice Cloning, and

Digital Audio Recording of Legal Proceedings

National Court Reporters Association (NCRA)

November 2023

STRONG Committee, 2020-2023

TABLE OF CONTENTS

Page

Background: A Brief Overview of Court Reporting ………………………………. 3

I. Introduction to the Risks and Harms of Artificial Intelligence (AI) and Automatic

Speech Recognition (ASR) in the Court Reporting Process …………….……… 5

II. Consequences of Choosing Digital-Only Methods ......................................... 6

III. Readiness of the Legal Justice System for Automatic Speech Recognition

(ASR), Artificial Voice Recognition (AVR), and Voice Cloning Technology …. 10

IV. ASR and AI and Racial, Gender, and Age Bias ………………………..……… 12

V. Statutes or Court Rules: Which Supersedes? ……………………………….... 13

VI. Conclusion: Protecting Judicial Integrity and Public Faith in America’s

Legal System: The Necessity of Court Reporters vs. AI and ASR

Technology …………………………………………………….……………………….. 15

VII. Endnotes ………………………………………………………………………….… 18

Acknowledgments

Background: A Brief Overview of Court Reporting

The National Court Reporters Association (NCRA) is recognized worldwide as the leading

authority on capturing and transcribing the spoken word into writing. The organization’s mission

is to apply the knowledge and experience of verbatim stenographic reporters working in

cooperation with the courts and bar toward improving the criminal and civil justice system to

best serve the public good.

Court reporters are highly trained professionals who share a unique ability to convert the spoken

word into information that can be read, searched, and archived. These experts also are known

as “guardians of the record” because of their impartiality and role within the judicial process

primarily capturing the words spoken by everyone during a court or deposition proceeding.

Court reporters then prepare verbatim transcripts of proceedings.

Stenographic court reporters are impartial officers of the court who must comply with state laws

and federal and state court rules regulating their practice.

The official record or transcript they generate helps safeguard the legal process. By combining

their skills with the latest technology, some court reporters provide realtime access to what is

said during a trial or deposition for the benefit of all involved parties. A court reporter providing

realtime — the only proven method for immediate voice-to-text translation — allows attorneys

and judges to immediately access a transcript while also providing a way for people who are

deaf or partially deaf to participate in the judicial process.

Artificial Intelligence (AI) and its subsets, such as machine learning, are only among the latest in

more than a century of technological advances that have disrupted and ultimately advanced the

professions of court reporting and captioning. AI is defined as “a machine-based system that

can, for a given set of human-defined objectives, make predictions, recommendations, or

decisions influencing real or virtual environments,” according to the National Artificial

Intelligence Act of 2020.

As highly trained, tech-savvy professionals, America’s court reporters and captioners have long

mastered cutting-edge innovations to bring the spoken word to text accurately in real time.

Indeed, today’s court reporters can accurately capture in writing 225 or more spoken words per

minute in real time.

To ensure the highest professional skills possible, including mastery of relevant technologies,

NCRA is the primary leader in setting national certification standards and assisting states with

their own certification or licensing requirements. To that end, the Association has administered a

nationally recognized certification program for court reporters since 1937. In addition, many

states currently accept or use the NCRA’s Registered Professional Reporter (RPR) certification

in place of state certification or a licensing exam.

Clearly, court reporters do not fear or avoid technology innovations when they know that such

changes are reliable, trustworthy, and strengthen transcript protection, accuracy, access, and

speed. However, AI has flaws and is developing so rapidly that even the world’s leading AI

developers and users are urging caution and greater control by governments. Indeed, more

than 1,000 CEOs and AI leaders sent a joint letter in early 2023 to policymakers urging federal

and state governments to establish guardrails and standards for the application and creation of

AI systems.

These leaders emphasized that AI and the use of machine learning or AI technologies to

process human speech into text (known as Automatic Speech Recognition or ASR) are not

unimpeachable in their work and outcomes.

From the perspective of NCRA, such technologies have already proven to be untrustworthy and

risky for the court reporting profession, especially when compared to the tremendous human

expertise, integrity, and experience of court reporters.

Other leaders and practitioners in the judicial system also have expressed concern and

warnings about the growing use, ethics, and influence of AI and ASR in the nation’s legal

system. In August 2019, for instance, the American Bar Association (ABA) House of Delegates

adopted Resolution 112 urging courts and lawyers to address the emerging ethical and legal

issues related to the usage of AI in the practice of law.

NCRA adopted a resolution urging stenographic court reporters, stenographic captioners,

affiliate associations, lawyers, bar associations, courts, and federal communication regulatory

agencies to address the emerging ethical and legal issues related to the use of AI, ASR, voice

cloning, and digital audio recording of legal proceedings without a stenographic court reporter

present to verify the chain of custody of the official record.

With both resolutions in mind, NCRA launched a three-year study exploring how these emerging

technologies could impact the capture of speech-to-text modalities within the legal justice

system when creating the record for courts and other testimonial proceedings. NCRA has

developed long-standing principles, opinions, and guidelines that are essential to the justice

system.

Ensuring AI is developed and used in accordance with established standards and guidelines as

well as in conjunction with state and federal rules and laws is critical when the use of such

technology could erode the public’s confidence, trust, and faith in the fairness and legitimacy of

the judicial system.

I. Introduction to the Risks and Harms of Artificial

Intelligence (AI) and Automatic Speech

Recognition (ASR) in the Court Reporting Process

The federal justice system and the justice systems of every state rely on the integrity and

accuracy of the trial court record for appeals. Without a proper record of what occurred in the

trial court, an appellate court may be left to decide matters of law based on the best available

means, including the appellant’s recollection.

Courts unable to attract and retain stenographic court reporters, coupled with always-evolving

technology, may push decision-makers and the public to ask why court proceedings are not

digitally recorded and automatically transcribed by computers or manually transcribed by

humans later.

In this white paper, NCRA examines the possible risks and harm to a legal record when

introducing the use of AI and digital audio recording of legal proceedings without a stenographic

reporter present to verify the chain of custody of the official record. This paper also explains how

the use of such technologies may be inimical to the public’s faith in the fairness of the judicial

system.

Most people agree that digital technology has brought important benefits to society. However,

the authentication of digital multimedia is an emerging challenge since it has become

increasingly easy to manipulate recorded audio contents using a growing number of free and

generally accessible software tools.

Today, when authenticity examinations are conducted on digital audio and video files, the

purpose of these complex analyses is fourfold to determine if a file

(1) is an original or clone (i.e., a bit-for-bit copy) or a re-encoded or transcoded copy,

(2) contains any alterations,

(3) has any discontinuities due to stop or start events, and

(4) matches the characteristics of a specified recording system, if known.

These authenticity examinations, probably more than any other forensic laboratory analysis,

require a more conceptual rather than purely “cookbook” protocol since every case differs from

prior ones. This variance is due to dissimilarities in the audio or video material, metadata,

compression effects, and diverse forms of possible duplication and alteration. For example, the

"loudness of the sound environment will produce prominent peaks, which may be too loud for

the mix. However, [courts] can lose the quieter moments when turning the overall gain down. In

this case, the solo speaker may get lost."

“The role of an audio recording is very important for criminal investigation agencies and for a

court of law [since] it is admissible evidence,” according to authors D. P. Gangwar and Anju

Pathania in their July 2020 article “Authentication of Digital Audio Recording Using File’s

Signature and Metadata Properties” in the International Journal of Engineering Applied

Sciences and Technology.

Unauthentic and forged multimedia can influence the decisions of courts. It also is fact that

audio recording evidence remains as useless evidence until it is proven that a recording is

authentic or free from any kind of tampering or editing. The detection of editing in an audio

recording is a challenging task to forensic scientists, one that requires greater attention in this

field.

The ease of altering digitally recorded audio files poses a major problem for the legal justice

system. A person with no training and minimal experience can manipulate the audio recording in

an effort to change testimony nearly seamlessly. Given the availability of free audio editors and

universally accessible tutorials on how to alter audio, the public, courts, lawyers, litigants, and

justice protection organizations must examine the trustworthiness of an audio- or video-only

record more thoroughly than ever.

A digital recording is the result of laying down a track of numbers, zeros and ones, in various

sequences into an audio file. Analyzing whether a digital recording has been altered and

whether the ones and zeroes follow each other in a seamless stream that shows no interruption

where an edit has occurred is difficult without highly trained forensic experts.

As Section 7.1.1 of the Scientific Working Group on Digital Evidence’s (SWGDE) Best Practices

for Forensic Audio advised in 2022, “Transcoding could affect the audio content (aliasing,

compression).”

For the same reason, a digital stream, once placed into a computer, can be extracted in any

sequence required by the operator. This ease of manipulation, coupled with the difficulty of

detecting such manipulation, creates an opportunity to tamper with important audio evidence.

In stark contrast to traditional alteration is the possibility that audio may be completely

fabricated. With the emergence of voice cloning, also known as deepfakes, participants in court

proceedings, including depositions, may be targeted for impersonation. Using tools available to

anyone online, a person’s voice or image may be manipulated into a deepfake.

The first publicized instance of how effective deepfakes can fool even the most educated people

was a 2019 case in which the CEO of an energy firm transferred hundreds of thousands of

dollars to a scammer after the leader thought he was speaking to his boss based on the caller’s

“melody” and slight accent, according to Forbes.

The technology for voice cloning has certainly

advanced since 2019, increasing concerns of more sophisticated malfeasance including acts

that could harm court proceedings and decision-making.

II. Consequences of Choosing Digital-Only Methods

Close inspection of the transcription protocols for courts also is required to maintain the

American public’s trust in their judicial system. Stenographic court reporters provide an

unbiased certification to their firsthand knowledge as witnesses to the proceedings for which

they are charged with making an accurate record, guarantee a chain of custody of that record,

and are backed by regulatory and licensing oversight. Court reporters ensure an unbiased

certification that is easily verifiable through the embedded stenographic watermark within

personal shorthand notes contained with their file.

In addition, human transcribers living in the United States are subject to subpoena, if required.

Outside of its jurisdiction, a subpoena may become no more than a “clumsily worded wish list,”

as Aaron Lukken explains in a 2016 post on the Hague Law Blog.

Additionally, any errors in

transcription, particularly with offshore transcription, can have serious consequences.

Quality Control

The quality of an audio recording can lead to serious gaps in testimony. Transcripts have been

made from audio recordings with portions missing because testimony was inaudible or garbled,

according to Joseph Darius Jaafari and Nicole Lewis in their 2019 article “In Court, Where Are

Siri and Alexa?” for The Marshall Project.

Any single case has great value for the parties

involved.

The risk of a transcript being incomplete, inaccurate, or misleading cannot be overstated. In the

context of AI, individual and enterprise accountability and human authority, oversight, and

control are required, and it is not appropriate to shift legal responsibility to a computer or an

algorithm rather than to responsible people and other legal entities, as stated the American Bar

Association Cybersecurity Legal Task Force Antitrust Law Section in 2023.

One startling example of the shortcomings of relying on digital recording methods to capture an

accurate and official record appears in a 2022 article, “Make Sure Your Court Reporter Is Really

a Court Reporter,” in the Los Angeles and San Francisco Daily Journal. Family law attorney

Melissa Buchman and coauthor Mary Pierce recount how Buchman lost a domestic violence

case because 55 pages of testimony were missing — all uncaptured by the digital recording

method used during a remote deposition.

She is not alone in her experience.

Additionally, transcripts prepared by anyone other than the person who digitally recorded the

proceedings make chain of custody difficult to track and do not comply with Federal Rules of

Civil Procedure (FRCP) 80. FRCP 80 declares, “If stenographically reported testimony at a

hearing or trial is admissible in evidence at a later trial, the testimony may be proved by a

transcript certified by the person who reported it.” Stenographic court reporters can provide that

certified proof while AI or digital recordings alone cannot.

How does human error in transcription factor into this equation? One might assume that the

answer to this question is to automate the process. Automatic speech recognition is often touted

as an incredible tool available for use by the bench, bar, and other court users and participants.

Unfortunately, when ASR or AI misunderstands a word, there is a good chance it will be omitted

or substituted with something that is completely irrelevant. Omissions can drastically change the

meaning of what has actually been stated, resulting in inaccurate transcripts. As research has

shown, the accuracy ratings of ASR leave much to be desired and also demonstrate the high

risk of bias.

For example, an objective review of ASR by Stanford University researchers studying a

potential “Race Gap in Speech Recognition Technology” showed an error rate of about 20

percent for white male speakers, 40 percent for Black male speakers, and 35 percent and 19

percent for white female and Black female speaker respectively, and an even higher error rate

when speakers spoke in the African American English dialect (AAE).

Of particular concern is the study’s conclusion that all of the most common speech recognition

systems performed “particularly poorly for Black men, with more than 40 errors for every 100

words.”

This is especially concerning since disproportionately large numbers of Black males

are engaged in the legal system in positions ranging from judges to the accused and thus are

regularly recorded in court.

“The leading speech recognition tools misunderstand Black speakers twice as often as white

speakers,” wrote the research team. “To close the gap, we must create more linguistically

diverse and inclusive datasets.” That takes time and investment, and in the meantime, ASR and

AI continue to be applied widely throughout nearly all aspects of society.

Concerns also have been raised regarding the use of AI translation in immigration cases.

“Machine translations of Pashto and Dari, in particular, are riddled with errors that have

introduced confusion into already complex immigration processes, and led to the rejected

asylum claim of at least one Afghan refugee,” stated an April 19, 2023, article, “AI Translation Is

Jeopardizing Afghan Asylum Claims,” in Rest of World.

In a justice system dedicated to equality, automated transcription has no place because it

continues to yield transcripts that contain many “inaudible” parentheticals as well as higher error

rates than trained human court reporters.

Costs

The cost of digital recording also can soar above a stenographic record.

Depending on the

accounting methods, personnel, and court setup, comprehensive pricing can vary wildly.

In late 2018 the Pierce County Superior Court, state of Washington, was asked if cost savings

could be achieved by use of electronic recording and creating a small pool of court reporters for

specific trials. This question triggered a statewide staffing study by the court administrator to

determine the staffing and cost breakdown for other courts in Washington.

Based on the research, the study showed “actual cost savings would be minimal at best, with

the court performance suffering greatly from the lack of realtime reporting. There is no court

recording equipment which has the performance level of a court reporter.”

Without training and staff dedicated to the supervision of audio systems, these systems cannot

be expected to perform well.

To make an audio- or video-only record, a court must — at a

minimum — have signage that notifies participants when proceedings are being recorded, have

judges remind participants to speak slowly and clearly for each case, position microphones

and/or cameras to fully capture the audio, and have a dedicated courtroom audio monitor to

assist the judge in conducting court proceedings.

Transcript Turnaround Times

It is clear stenography remains the most reliable and cost-effective method of capturing the legal

record, not only due to stenographers’ crucial roles as impartial observers and verbatim

notetakers, but also because transcription from an audio source is far less efficient, resulting in

delayed transcript turnaround times.

The average person can transcribe an hour of audio in four hours.

Seven hours of audio could

take more than 28 hours to transcribe.

By comparison, stenographic court reporters are relied on to produce “daily” transcripts,

meaning the full day of testimony would be delivered to end users shortly after conclusion of the

proceedings. High-quality rough drafts and realtime services are available for immediate review

of testimony during the pendency of the proceedings.

With stenographic reporters’ superior efficiency, reliability, expertise, similar cost, and built-in

protection against emerging deepfake and voice cloning technologies, courts and the general

public would be wise to continue to entrust stenographic court reporters with guarding the legal

record.

Archiving Requirements

Another concern raised about AI and ASR relates to the safety of digitally stored media. Storage

of digital media on traditional hard drives leaves much to be desired. Analysis of hard drive

lifespans concludes that only 80 percent of hard drives will reach their fourth anniversary without

malfunctioning.

A minimum of three backups, including one offsite, is recommended for

traditional digital data storage.

Courts and attorneys that seek to keep their own personal storage libraries but do not invest in

personnel to organize, supervise, and maintain the equipment face a serious chance that data

will be lost.

In addition, storing data through cloud computing solutions remains an optional practice rife with

cybersecurity threats, including cryptojacking, hijacking of accounts, data breaches, and denial-

of-service attacks. Cloud computing is considered a responsibility shared by the customer and

the service provider.

The hijacking of accounts is among the greatest risks for administrators of high-volume courts.

More cases mean more personnel require access to the cloud accounts. More personnel with

access means more potential openings for successful phishing scams.

III. Readiness of the Legal Justice System for

Automatic Speech Recognition (ASR), Artificial

Voice Recognition (AVR), and Voice Cloning

Technology

The global speech and voice recognition industries are worth tens of billions of dollars built on

the collection of users’ biometrics without meaningful protections for the public in terms of data

storage, retention, security, and privacy concerns.

Biometrics are defined by Merriam-Webster

as “the measurement and analysis of unique physical or behavioral characteristics, especially

as a means of verifying personal identity.” Common examples of biometric data are fingerprints,

voiceprints, faceprints, and iris scans.

Video and audio cloning have benevolent uses in film, audiobooks, podcasts, and even medical

settings where throat cancer or laryngectomy patients can clone their own voice to be used later

through an implanted box, allowing them to regain the ability to speak in their own voice.

However, with the advent of the ethical and beneficial uses of these technologies comes the

higher likelihood of nefarious and unethical uses that could occur using the same voice cloning

concepts. AI and biometrics are certainly welcome advancements in the ethical uses of voice

cloning, but there is growing concern for courts and litigators who now must further protect the

safety, reliability, and security of court records.

Gone is the time when a transcript or evidence derived from an audiotape or videotape

introduced at trial was assumed to be authentic with few questions raised. Courts must now be

forward-thinking and extremely cautious in identifying such authenticity and ensuring a strict,

documented chain of custody in all audio and video files presented.

Courts also must maintain the same level of security with their own audio and video files which

they may transfer to transcribers. If a particular audio or video file were in the hands of a

transcriber or multiple transcribers, the possibility exists that the file can be altered to change

what is actually spoken on the media file with creative editing or voice cloning technology.

Consider this scenario: An audio or video file could be altered, whether purposefully or

unwittingly, and then sent to a transcriber. How then would a transcriber be able to certify that

what was received and listened to is indeed the authentic recording? The same can be said

when the transcriber sends the transcribed file back to the court or other agency. Who verifies

that the transcript received matches the audio that was provided to the transcriber?

In the case of the 55 pages of missing testimony cited earlier, it appears there was a lack of

comparison with the audio file. The bottom line is that court files have many touchpoints, all with

the potential of a breach and all with the potential to influence court and legal outcomes.

Unfortunately, legislation and court rules have lagged behind emerging AI, ASR, and deepfake

technology. Few federal and state laws currently exist that address deepfakes. Although AI has

existed in some form since the 1950s, the first federal law to address deepfakes was enacted in

late 2019.

Recently, governmental agencies and other scientific organizations have proposed

AI governance best practices and principles, attempting to mitigate or minimize risks associated

with the technology. On October 30, 2023, President Joseph R. Biden, Jr. issued the Executive

Order on Safe, Secure, and Trustworthy Artificial Intelligence, charging multiple federal

agencies with producing guidelines and taking other actions to advance the safe, secure, and

trustworthy development and use of Artificial Intelligence.

On January 28, 2020, the Federal Trade Commission held a workshop with world-renowned

experts titled, “You Don’t Say: An FTC Workshop on Voice Cloning Technologies.” The

assembled experts and regulators explored the dangers and risks that this technology poses to

society. Their takeaway was heightened concern about how easily this technology can be

acquired, deployed, and seemingly go undetected without the use of a forensic expert.

On April 26, 2023, Sen. Mark Warner (D-VA), Chair of the Senate Select Committee on

Intelligence, sent letters to CEOs of leading AI companies to express concerns about the risks

of these technologies.

“’[W]ith the increasing use of AI across large swaths of our economy, and the possibility for

large language models to be steadily integrated into a range of existing systems, from

healthcare to finance sectors, I see an urgent need to underscore the importance of putting

security at the forefront of your work,” Sen. Warner wrote. “Beyond industry commitments,

however, it is also clear that some level of regulation is necessary in this field.”

Sen. Warner highlighted several security risks associated with AI, including data supply chain

security and data poisoning attacks. He also voiced apprehension about algorithmic bias,

trustworthiness, and potential misuse or malicious use of AI systems.”

The introduction of deepfake evidence in the courtroom raises new, profound issues for the

administration of justice in both civil and criminal proceedings.

The use of AI in America’s legal

system is certainly a high risk that should be of particular concern and subjected to a higher

degree of scrutiny. Security and accuracy of confidential and sensitive information contained in

exhibits and transcripts are essential, and transparency in how these records are handled is

paramount in maintaining public trust in the nation’s justice system. Members of the public

should be advised if AI and/or ASR is being used in litigation and should be cautioned regarding

those risks.

Courts should expect deepfake audios and videos to begin to find their way into court transcripts

and files. Currently, transcripts prepared by digital reporters and/or ASR transcription

companies include cleverly worded or misleading “certificate” pages that are designed to give

legitimacy to a document that, in essence, has no basis for certification if one person who

administered the oath and witnessed the proceedings records the audio file and another person

transcribes the audio file without comparing the resultant transcript. Proper certification and

authentication are further complicated if there are multiple undisclosed transcribers involved.

How can courts protect against these dangers?

First, courts can take stock of their own internal security practices. If audio or video is being

used to capture the record in courtrooms, it is vital that control measures be implemented that

govern who has access to these files and how file access to others outside the organization is

granted.

Washington, after a legislative effort, is an excellent example of a state establishing safeguards

for its courts. Its new court rule requires chain of custody for audio file in addition to a court

requirement to provide authorized transcriptionist standards and oversight.

Another safeguard would be if a court maintains an “approved” transcriber list and prudently

takes the additional measure of knowing each person within the transcriber’s organization who

has access to the files and how they are transferred. Does each transcriber maintain adequate

virus protection on their computer, for instance?

While the latter may seem like a mundane question, the public may be surprised at the answers,

and the court may want to ensure that safe practices are being followed by its authorized

providers. It only takes one data breach for a catastrophic event to occur, so protocols and

standards should be developed, defined, and adhered to rigorously.

Another dilemma the courts must wrestle with is evidence introduced by counsel at trial that

involves transcripts introduced after having been transcribed from audio or video sources

created outside of the court on privately owned audio or video recording devices. Strong

safeguards should be put into place regarding the chain of custody of all audio or video media

and/or transcripts generated by a method using only audio or video for transcript production.

These pose significant risks of manipulation if strict protocols are not followed — another reason

close attention should be given to the “certificate” page of all transcripts. Such scrutiny would

determine when one has been recorded and transcribed after the fact, especially when doing so

conflicts with any state laws. “Accountability and human authority, oversight, and control are

closely interrelated legal concepts.”

IV. ASR, AI, and Racial, Gender, and Age Bias

AI lacks the capacity for contextualization that humans possess. It works by combining large

data sets with intuitive processing algorithms that can be manipulated by learning behavior

patterns within one or more data sets. Most of these sets are created by predominantly white

male voices, commonly referred to as “pale male data.”

Issues of fairness have arisen because systems do not perform equally well for all subgroups of

the population. This is where bias research shows the true risk of harm that can result from the

fallibility of AI and ASR.

According to the Stanford University study

referenced previously, “Error rates for Black

speakers are nearly double those for white speakers. We found that all five ASR systems

exhibited substantial racial disparities, with an average word error rate (WER) of 0.35 for Black

speakers compared with 0.19 for white speakers.” What is clear across multiple studies is that

training data has led machines to learn more about white men’s speech patterns and less about

those of women and people of color.

One case in point is that Stanford researchers found that Google’s speech recognition is 13

percent more accurate for men than it is for women. ASR or AI platforms frequently use

feminine voices as digital assistants such as Siri, Alexa, and other voice-activated chatbots.

However, virtual assistants are more likely to understand male users than female users.

In a different study shared in the article “Bias in Automatic Speech Recognition: The Case of

African American Language” in Applied Linguistics, “researchers Joshua Martin and Kelly

Elizabeth Wright list numerous cases of systemic bias in their evaluation of various speech

activation technologies and its often poor interpretation and transcription of speech by African

Americans.”

In another example of AI and ASR bias, researcher Lauren Werner concludes in a 2019 article,

“Automated Speech Recognition Systems and Older Adults: A Literature Review and

Synthesis,” that “age-related physical changes may alter speech production and limit the

effectiveness of ASR systems for older individuals. Evaluation of several automated speech

recognition systems has confirmed previous research that suggested those systems have more

difficulty recognizing the speech of older adults.”

Efficient, effective, and equitable solutions will be found when decision-makers include the

knowledge and expertise of stenographers in future planning of the best means to capture

speech-to-text that depend on humans to hear and conceptualize speech rather than depend on

predictive speech models and algorithms.

V. Statutes or Court Rules: Which Supersedes?

While all states have court rules related to the people who may administer oaths and function as

a deposition officer, some states have laws that prohibit notaries from providing court reporting

services.

Over half the states have statutes regulating all facets of court reporting where legislatures have

found it necessary to regulate the practice of court reporting with licensure or certification

requirements to protect the public safety and well-being.

However, for many years the litigation arena has seen analog and digital recording equipment in

courtroom proceedings. For the same number of years, transcriptions of those recordings have

been littered with inaudible and/or unintelligible parentheticals in addition to unidentified

speakers.

Despite the problems, many court systems have slowly and tragically adopted a standard of

“adequate” rather than “accurate” for these all-important records that must stand the test of

reliable review over years or decades.

Although courtroom recording equipment is frequently monitored by court personnel who

operate under predetermined court standards, this is certainly not the universal practice. The

recordings themselves are known to be largely transcribed by remote employees or contractors

or authorized transcriptionists who have no knowledge of the speakers, have no standard

guidelines on transcript preparation, and have no requirements in place for security, chain of

custody, or preservation of the actual transcription.

The process of digital recording does not transfer well from the courtroom setting to the regular

deposition setting in that there is no universal standard or set of requirements related to the

equipment used to create an audio recording.

Unfortunately, recording and transcription standards that do exist are widely diverse and largely

unregulated. The range of digital audio and video files used by multiple companies is large and

ever-changing, as are the practices for distribution and transcription of media. Some digital and

video companies are incorporating AI and ASR into transcription of files even though most court

rules do not contain specific language allowing either technology.

The first significant challenge of deepfakes is proving that a piece of digital image or audio

evidence is genuine.

The Federal Rules of Evidence, 901(a), states, “In General. To satisfy

the requirement of authenticating or identifying an item of evidence, the proponent must

produce evidence sufficient to support a finding that the item is what the proponent claims it is.”

K. Chasse, in the article “The Admissibility of Electronic Business Records,” suggests that the

authentication rule at times appears inadequate because it cannot be established that an

electronic record is the same as its first instantiation simply by looking at the record itself.

public knowledge of deepfakes continues to grow and people become increasingly skeptical

about the credibility of audiovisual images, jurors will be primed to doubt the authenticity of even

real audio and video content.

In R v Nardi 2012 BCPC 0318, the court held that in order to support a finding that electronic

records are authentic and the “best evidence” of the information proffered, the party seeking to

admit the evidence “cannot simply look to the documents themselves.” Further, the court said

this is especially so when considering information generated from a novel “system.”

The court reporting industry is facing an existential challenge in the provision of its vital services.

Some video and court reporting companies are encouraging attorneys to stipulate to “digital

court reporters” or “digital video reporters” rather than the skilled and highly trained stenographic

reporters on whom they historically rely.

Frequently these companies are encouraging this practice with full knowledge that it does not

comply with existing court rules and laws in many jurisdictions. These measures often disregard

prior agreement of counsel to use a method other than stenography, which is required in most

existing state court rules.

In many instances counsel is not even aware until after the digital reporter or videographer

arrives at the deposition, administers the oath, and the deposition is set to begin. When parties

believe they are contracting a certified court reporter but instead are ambushed with a non-

certified digital reporter or digital videographer, it not only causes strife among litigants, but it

also may lead to unwitting violations of laws and court rules. A transcript that is impartially

attended and accurately transcribed and certified by the person who witnessed the proceedings

imports a verity that is vital to the integrity of legal proceedings.

VI. Conclusion: Protecting Judicial Integrity and

Public Faith in America’s Legal System: The

Necessity of Court Reporters vs. AI and ASR

America’s justice system is founded on the premise of providing fair and equal access and

equitable treatment to all, and in many circumstances, both in the legal system and without,

stenographic court reporters and captioners have always embraced the concept of fair and

equitable access.

Stenographic court reporters are committed to the rule of law and their role in it. In its simplest

form, the rule of law means that every person in the United States is subject to clearly defined

and publicly promulgated, well-accepted legal standards and principles that are equally

enforced rather than subjected to the personal whims of powerful corporations, individuals,

governments, or other entities.

The concept also embraces two other foundational principles deeply rooted in American

jurisprudence: (1) American laws apply equally to all people at all times, and (2) no one is above

the law.

However, a September 2022 Gallup survey found that just 47 percent of adult respondents

“expressed even some trust in the judicial branch of the federal government, a stunning 20-point

drop over the last two years and a 7-point drop from last year.”

Reasons vary, but it is safe to say that few within the public understand that transcript integrity

and accuracy are foundational to fair, equitable court decision-making. Individuals charged with

determining the future of people’s lives — from juries to judges — must have total faith that

transcripts reflect the actual stated words and/or testimony of those involved in or presenting a

case.

AI and its subset technologies have no loyalty to the rule of law and no ability to perceive and

interpret nuance or emotional expression. Although exciting in their abilities to achieve tasks

such as processing mass data, none has the ability (as yet) to provide risk-free and near-total

accuracy in capturing, transcribing, and protecting the words of all people regardless of gender,

race, national origin, or age.

In recognizing AI’s serious risks and fallibility, the ABA and many of its specialty sections have

studied and passed numerous related policies that acknowledge innovative opportunities but

strongly caution users and developers about the real or possible harmful impacts on insurance,

privacy, and cybersecurity, to name a few sectors.

In February 2023, the ABA adopted Resolution 604 calling on organizations that design,

develop, deploy, and use AI to follow three key guidelines: (1) ensure their products are subject

to human authority, oversight, and control; (2) ensure accountability measures if developers

have not taken reasonable steps to mitigate harm or injury; and (3) provide transparency and

traceability for their products. Resolution 604 also encourages federal and state lawmakers and

policymakers to follow the same standards in AI policymaking.

A universal thread throughout these statements, resolutions, and reports is an urgent call to

action to decision-makers that “responsible individuals and organizations should be accountable

for the consequences caused by their use of AI products, services, systems, and capabilities,

including any legally cognizable injury or harm caused by their actions or use of AI systems or

capabilities, unless they have taken reasonable measures to mitigate against that harm or

injury.”

NCRA continues to direct attention to the extreme legal risks associated with the use of AI,

ASR, and digital recording in today’s courts and deposition rooms. However, misinformation

about speed, safety, accuracy, accountability, and cost savings continues to flourish.

• Predictions that digital recording and AI can increase efficiency at lower cost are simply

not supported by the facts. Digital recording systems require multiple individuals to

monitor and maintain the systems, incur additional storage costs, experience undetected

malfunction, and need teams of far less efficient transcriptionists to produce transcripts

than skilled court reporters.

• Traceability is considered a key element for trustworthy AI. Traceability relates to the

need to maintain a complete account of the provenance of data, process, and artifacts

involved in the production of transcripts that incorporate an AI model. Without

traceability, AI should not be used in the production of certified court or deposition

transcripts.

• Inequalities in automated voice recognition translation affect access to justice for already

marginalized communities such as immigrants and people of color.

In May 2020, NCRA appointed the STRONG Committee to undertake a comprehensive study

that included review of scholarly articles and thousands of pages dedicated to the science and

technologies of AI and ASR.

After reviewing the work of this committee, NCRA adopted the following resolution:

NOW, THEREFORE, BE IT RESOLVED that the National Court Reporters Association

urges that lawyers, bar associations, courts, and federal communication regulatory

agencies should ensure that digital court reporter or transcriptionist products, services,

systems, and capabilities are subject to human authority, oversight, and control to verify

the chain of custody of the official record when use of such technology may fail to protect

the privacy of litigants and could erode the public’s confidence, trust, and faith in the

fairness and legitimacy of the judicial system, and that

1) Organizations that design, develop, deploy, and use artificial intelligence and

automatic speech recognition systems and capabilities must be subject to human

authority, oversight, and control.

2) Responsible individuals and organizations should be accountable for the

consequences caused by their use of AI products, services, systems, and capabilities.

3) All individuals participating in legal proceedings should be duly advised if AI or

ASR will be utilized in the production of transcripts. Appropriate cautions of the risks and

dangers the use of such technology poses to biometric and private information should be

disclosed. Each individual involved in legal proceedings should be allowed to decide

whether they wish to opt in or out of being subjected to its use.

BE IT FURTHER RESOLVED, that the National Court Reporters Association urges

Congress, federal executive agencies, and state legislatures and regulators, courts,

lawyers, court reporting firms, consortiums, and associations, together with law firms and

bar associations and broadcasting companies, to follow these guidelines in legislation

and standards pertaining to the use of AI and ASR in court and legal environments.

Success cannot be measured by short-term budgetary considerations but instead should

be measured by honest, equal, and fair treatment for all parties. Increased costs

combined with lack of oversight, security flaws, poorly trained personnel, and equipment

failures cannot equal or exceed the performance of a human stenographic reporter or

captioner.

The November 2023 publication of this white paper is the first of what is expected to be

a series of ongoing updates and documents about emerging ethical and legal issues

related to the use of AI, ASR, voice cloning, and digital audio recording of legal

proceedings. Due to the fast-changing pace of development, use, and potential harms

related to these issues, NCRA acknowledges that this is an inherently dynamic paper

subject to future changes

VII. Endnotes

Federal Rules of Appellate Procedure Rule 10. Cornell Law School, Legal Information Institute.

Last accessed Sept. 29, 2023: https://www.law.cornell.edu/rules/frap/rule_10

Sones, Z. “What is Audio Compression?” Aug. 18, 2022. The Beat. Last accessed Oct. 24,

2023: https://www.premiumbeat.com/blog/what-is-audio-compression

Gangwar, D. P., and Pathania, A. “Authentication of Digital Audio Recording Using File’s

Signature and Metadata Properties” (2020). International Journal of Engineering Applied

Sciences and Technology 5(3): 162-165. Last accessed Oct. 26, 2023:

https://www.ijeast.com/papers/162-165,Tesma503,IJEAST.pdf

”The Case Against Digitally Recorded Evidence.” June 20, 2016. FindLaw. Last accessed

Sept. 28, 2023: https://www.findlaw.com/legal/technology/modern-law-practice/the-case-

against-digitally-recorded-evidence.html

Scientific Working Group on Digital and Multimedia Evidence (SWGDE). Best Practices for

Forensic Audio (2022). Vol. V 2.5. Last accessed Sept. 26, 2023:

https://drive.google.com/file/d/1zF0KzO5Xe5jAO2CgLxISIm8fLSFJ46ax/view

Damiani, J. “A Voice Deepfake Was Used to Scam A CEO Out of $243,000.” September 3,

2019, Forbes. Last accessed Sept. 28, 2023:

https://www.forbes.com/sites/jessedamiani/2019/09/03/a-voice-deepfake-was-used-to-scam-a-

ceo-out-of-243000/?sh=79c590f12241

Lukken, A. “Serving a Subpoena Abroad? Not So Fast, Counsel.” April 18, 2016. Hague Law

Blog. Last accessed Oct. 24, 2023: http://www.vikinglaw.us/2016/04/18/serving-a-subpoena-

abroad-not-so-fast-

counsel/#:~:text=A%20subpoena%E2%80%99s%20coercive%20effect%20is%20revived%20on

ly%20by,crosses%20an%C2%A0internationalboundary%2C%20it%20cannot%20regain%20its

%20coercive%20effect.%2A

Jaafari, J. D., and Lewis, N. “In Court, Where Are Siri and Alexa?” Feb. 14, 2019. The Marshall

Project. Last accessed Sept. 28, 2023: https://www.themarshallproject.org/2019/02/14/in-court-

where-are-siri-and-alexa

American Bar Association Cybersecurity Legal Task Force Antitrust Law Section, 2023.

Buchman, M., and Pierce, M. “Make Sure Your Court Reporter Is Really a Court Reporter,”

April 13, 2022. Los Angeles & San Francisco Daily Journal. Last accessed Sept. 28, 2023:

https://www.thejcr.com/2022/04/19/make-sure-your-court-reporter-is-really-a-court-reporter

The Race Gap in Speech Recognition Technology.” Fair Speech. Stanford University. No

date. Last accessed Sept. 29, 2023: https://fairspeech.stanford.edu

NAACP, Criminal Justice Fast Sheet. No date. Last accessed Oct. 23, 2023:

https://naacp.org/resources/criminal-justice-fact-sheet

“AI Translation Is Jeopardizing Afghan Asylum Claims,” April 19, 2023. Rest of World. Last

accessed Sept. 27, 2023: https://restofworld.org/2023/ai-translation-errors-afghan-refugees-

asylum

Crawford, C. “A Study of Court Reporting and Digital Recording (DR) in the California and

Florida Courts.” Dec. 3, 2009. Justice Served. California Court Reporters Association. Last

accessed Sept. 29, 2023: chrome-

extension://efaidnbmnnnibpcajpcglclefindmkaj/https://www.cal-

ccra.org/assets/documents/Justice_Served-CCRA_Dec09.pdf

Pierce County Superior Court, November 8, 2018. Department Budget Presentation.

https://online.co.pierce.wa.us/cfapps/council/model/otDocDownload.cfm

Greenwood, J. M., Horney, J., Jacoubovitch, M. D., Lowenstein, F.B., and Wheeler, R. R. “A

Comparative Evaluation of Stenographic and Audiotape Methods for United States District Court

Reporting.” 1983. Federal Judicial Center, p. 80. Last accessed Sept. 27, 2023:

https://www.ojp.gov/ncjrs/virtual-library/abstracts/comparative-evaluation-stenographic-and-

audiotape-methods-united

Suskin, L., McMillan, J., and Hal, D. “Making the Record: Utilizing Digital Electronic

Recording.” September 2013. National Center for State Courts. Last accessed Oct. 23, 2023:

chrome-

extension://efaidnbmnnnibpcajpcglclefindmkaj/https://www.ncsc.org/__data/assets/pdf_file/0021

/17814/09012013-making-the-digital-record.pdf

“How Long Does It Take to Transcribe One Hour of Audio or Video?” No date. Rev. Last

accessed Sept. 28, 2023: https://www.rev.com/blog/resources/how-long-does-it-take-to-

transcribe-audio-video

New York, Part 108 Format of Court Transcripts and Rates of Payment Thereof,

108.2(b)(2)(iii). Last accessed Sept. 29, 2023:

https://ww2.nycourts.gov/rules/chiefadmin/108.shtml

Ibid. See Suskin, McMillan, and Hal.

Samani, R. “Is Cloud Security an Exaggerated Concern?” April 22, 2016. McAfee. Last

accessed Oct. 24, 2023: https://www.mcafee.com/blogs/other-blogs/executive-

perspectives/cloud-security-exaggerated-concern

Veltman, C. “Send in the Clones: Using Artificial Intelligence to Digitally Replicate Human

Voices.” Jan. 17, 2022. NPR. Last accessed Sept. 28, 2023:

https://www.npr.org/2022/01/17/1073031858/artificial-intelligence-voice-cloning

Thompson, K. “AI Enables Laryngectomy Patients to Get Their Voice Back.” April 13, 2022.

Create Engineering Ideas into Reality. Last accessed Sept. 27, 2023:

https://createdigital.org.au/ai-enables-laryngectomy-patients-to-get-their-voice-back

Deepfake Act of 2019 (S. 2065). 116

Congress (2019-2020), passed Senate Oct. 24, 2019.

Congress.gov, Library of Congress. Last accessed Sept. 29, 2023:

https://www.congress.gov/bill/116

-congress/senate-bill/2065/all-info An identical bill, introduced

in the U.S. House of Representatives as H.B. 3600, was referred to the House Energy and

Commerce Committee’s Subcommittee on Consumer Protection and Commerce Oct. 29, 2019.

Shepardson, D. “U.S. Senator Urges AI Company CEOs to Take Steps to Address Risks.”

April 26, 2023. Reuters. Last accessed Sept. 28, 2023: https://www.reuters.com/technology/us-

senator-urges-ai-company-ceos-take-steps-address-risks-2023-04-26

Lehman, K., Edson, S., and Smith, V. “5 Ways To Confront Potential Deepfake Evidence In

Court.” Law360. Subscription required: https://www.law360.com/articles/1181306/5-ways-to-

confront-potential-deepfake-evidence-in-court

Washington GR 35 Adopted effective Sept. 1, 2015. Amended effective Sept. 1, 2018.

American Bar Association House of Delegates, Resolution 604. Published Feb. 6, 2023. Last

accessed Oct. 24, 2023: chrome

extension://efaidnbmnnnibpcajpcglclefindmkaj/https://www.txmca.org/fileman/Uploads/Symposi

um%20Materials/2023/Susan%20Guthrie%20AI/ABA_604_midyear_2023_Resolution_on_AI.pd

Ibid. See Stanford University.

Martin, J., and Wright, K. E. “Bias in Automatic Speech Recognition: The Case of African

American Language,” August 2023 (Published Dec. 14, 2022). Applied Linguistics, Volume 44,

Issue 4, pp. 613-630. Last accessed Sept. 27, 2023:

https://academic.oup.com/applij/article/44/4/613/6901317 or

https://doi.org/10.1093/applin/amac066 and

http://www.ethicsinnlp.org/workshop/pdf/EthNLP06.pdf

Werner, L., Huang, G., and Pitts, B. “Automated Speech Recognition Systems and Older

Adults: A Literature Review and Synthesis.” Proceedings of the Human Factors and Ergonomics

Society 2019 Annual Meeting. Purdue University. Last accessed Sept. 26, 2023: chrome-

extension://efaidnbmnnnibpcajpcglclefindmkaj/https://journals.sagepub.com/doi/pdf/10.1177/107

1181319631121 or https://journals.sagepubs/werner-et-al-2019-automat_ture-review-and-

synthesis.pdf

RCW 42.45.230€. Washington state government. Adopted effective Oct. 1, 2020. Page list

last accessed Sept. 29, 2023: https://app.leg.wa.gov/rcw/default.aspx?cite=42.45

National Court Reporters Association, State Certification Requirements. Last accessed Sept.

26, 2023: https://www.ncra.org/home/professionals_resources/ncra-information-and-resource-

center/state-resources/State-Certification-Requirements

Delfina, R. A. “Deepfakes on Trial: A Call to Expand the Trial Judge’s Gatekeeping Role to

Protect Legal Proceedings from Technological Fakery” (2023). 74 Hastings L.J. 293. Last

accessed Sept. 28, 2023: https://repository.uchastings.edu/hastings_law_journal/vol74/iss2/3

Chasse, K. “The Admissibility of Electronic Business Records.” (2010) 8:2 CJLT. Last

accessed Sept. 29, 2023:

https://digitalcommons.schulichlaw.dal.ca/cjlt/vol8/iss2/1/#:~:text=Ken%20Chasse%2C%20%22

The%20Admissibility%20of%20Electronic%20Business%20Records%22%20(2010)%208%3A2

%20CJLT. Chasse also refers to the nine points of proof specified in the National Standard of

Canada, Electronic Records as Documentary Evidence CAN/CGSB-72.34-2005, Section 5.5.

Mason, S., and Stanfield, A. Authenticating Electronic Evidence, 2017. Stanfield, School of

Advanced Study, University of London Institute of Advanced Legal Studies, p.p. 193-260. Last

accessed Sept. 29, 2023: https://www.jstor.org/stable/j.ctv512x65.14

Jones, J. “Supreme Court Trust, Jobs Approval at Historical Lows.” (2022, Sept. 29). Gallop.

https://news.gallup.com/poll/402044/supreme-court-trust-job-approval-historical-lows.aspx

American Bar Association Cybersecurity Legal Task Force Antitrust Law Section, 2023.

Acknowledgments

The National Court Reporters Association expresses its sincere gratitude to the NCRA

STRONG Committee for contributing to the research and creation of this white paper. Without

the committee’s invaluable efforts and support, this work would not have been possible.

First and foremost, NCRA deeply appreciates work by the following individuals whose

dedication and expertise shaped the content and insights contained herein, and whose

combined experience in the stenographic court reporting profession encompasses 525 years

and spans 11 states:

Sue A. Terry, FAPR, RPR, CRR, CRC (Ohio), Chair, NCRA STRONG (48 years in the profession)

Stefanie R. Allison, RPR (Nebraska) (25 years)

Douglas Bettis (Ohio) (25 years)

Lisa Migliore Black, CCR (Kentucky), Vice-Chair, NCRA STRONG (26 years)

Christopher Day (New York) (13 years)

Debbie Dibble, RDR, CRR, CRC (Utah) (33 years)

Kimberly Falgiani, RDR, CRR, CRC, CSR (HI) (Ohio) (43 years)

Lillian M. Freiler, FAPR, RMR, CMRS (Pennsylvania) (48 years)

Elizabeth A. Harvey, FAPR, RPR (Washington) (27 years)

Jo Ann Holmgren, CCR (Texas) (38 years)

Andrea Kreutz, CLVS (Iowa) (8 years)

Phyllis Craver Lykken, FAPR, RPR, WA CCR, OR CSR (Washington) (43 years)

Mary E. Pierce, CSR (California) (40 years)

Stacey E. Raikes, RMR, CRR (Florida) (26 years)

Lin D. Riffle, RDR, CRR, CRC (Ohio) (36 years)

Lindsay Stoker, RMR, CRR, CRC (California) (16 years)

Dineen Squillante, RPR (Vermont) (30 years)

Their commitment to excellence and tireless efforts in researching, writing, and revising this

document have been instrumental in producing a high-quality resource.

Keywords: legal transcription, automatic speech recognition, bias, courts, transcription, transcription

errors, court record, privacy, stenographer, transcript security, voice recognition, justice protection, justice

reform, courts and artificial intelligence, audio transcription, civil liberties, ASR, biometric privacy, audio

biometrics, digital audio, discrimination in courts, electronic recording, electronic recording errors, FTR,

JAVS, BIS, court electronic vendor, court electronic recording vendor, transcript quality control, ASR

perturbations, AI security, AI ethics, AI accuracy, speech-to-text accuracy, transcription bias, transcription

errors in court transcripts, courtroom technology