ARTIFICIAL INTELLIGENCE (AI) IN MEDICINE AND LAW

By Judge Robert G. Rassp
California Workers’ Compensation Appeals Board
Los Angeles, CA

Position paper presented at CSIMS 2024 by Hon. Robert G. Rassp, Chairman of the Board of Directors, Friends Research Institute (friendsresearch.org)

Disclaimers: The opinions expressed in this article are those of the author and are not those of the State of California Department of Industrial Relations, Division of Workers’ Compensation, or the Workers’ Compensation Appeals Board. The opinions expressed herein are based in part on the “Common Rule” 45 CFR 46 that pertains to the ethical requirements in medical research and the protection of research participants. There is no current legal requirement that 45 CFR 46 applies to injured workers whose claims may involve the use of AI.

INTRODUCTION

Artificial Intelligence, or “AI,” is taking our society by storm. When computers first became in wide use in business applications, advances in the programming language would occur every five years or so with upgrades in software development that would cause users of computers to replace old operating systems and download the latest operating system applicable to either their Mac or IBM based computer. Today, software is being upgraded by software itself by at least six versions of machine language. In fact, computer programmers can download software applications that are bundled so that applications can easily be embedded in sophisticated computer programs. Have you used a kiosk at McDonald’s? Or ordered a coffee from Starbucks lately? Machines are now processing our orders at fast food joints thanks to the sophisticated computer programming. You call a Call Center and you never speak to a human being. You see the “Chat” icons for banks and other services with a web site? Those are run by AI based software. If you want to speak to a human being, you usually have to keep repeating “representative!” multiple times, or hit “0” repeatedly, and you might get lucky and get a live person on the phone or in the chat.

Call centers for some companies are now voice activated and responses are via a computer program upon verbal or numerical prompting by the calling party. Most of these programs are driven by AI. AI is now affecting much of our daily lives even though we may not even realize a response to something is driven by a computer program. Your physician interacts with you by physician-patient portals that may be driven by AI via link to your medical records and the physician’s electronic medical record notes. Did you know that the telehealth appointment you had with your doctor was actually with an avatar while your real doctor was golfing at his favorite course?

So how does AI fit in the context of medicine and law? This article was written by this author as a result of prior notes he utilized for a presentation at the California Society of Industrial Medicine and Surgery Conference that occurred on August 14, 2024, at the Loews Coronado Island Resort. The title of the program was “Artificial Intelligence in Medicine and Workers’ Compensation Law.” The panel consisted of this author (in the capacity of both a workers’ compensation presiding judge and Chairman of the Board of Directors of Friends Research Institute (friendsresearch.org), Dr. Christopher Brigham MD (editor of the AMA Guides To The Evaluation of Permanent Impairment, 6^th Ed. and principal of emedicine.com), Ray Mieszaniec (COO of Evenup – a legal tech company), and defense attorney Negar Matian (who is using AI applications in her workers’ compensation defense law practice).

This author’s presentation focused on the author’s opinion that guardrails need to be placed in the use of AI in the context of medicine and workers’ compensation litigation. While there is no question that AI development companies have emerged to focus on specific industries, including our own in workers’ compensation claims, a discussion of ethical considerations is necessary as these applications are introduced into our everyday lives. This is especially true in the context of workers’ compensation claims and the role of physicians including treating doctors and medical-legal evaluators.

So how does the legal requirements for medical-legal reporting work if a physician utilizes AI software to review medical records, to communicate with the injured worker, to write reports that are admissible at the Workers’ Compensation Appeals Board? Can a defense attorney rely on AI software to write a communication to the employer or claims examiner recommendations for further case handling? Can defense counsel rely on AI to provide an injured worker’s deposition summary or to develop questions to ask a physician at a deposition? Can counsel delegate writing Points and Authorities, a legal brief, or a Petition for Reconsideration to a generative artificial intelligence-based software program? Can a workers’ compensation judge write a decision with the use of an AI program? These questions are all relevant and everyone in the workers compensation system have or will be confronted by the issue of how AI affects the way these cases are handled going forward. What is a legitimate role, if any, of the use of AI in the context of workers’ compensation cases?

CHATGPT

Most of the public’s first exposure to AI occurred in November 2022 with the public launch of ChatGPT which allowed anyone with a computer to seek information from an AI platform. You type in a key word, or words, and the program would produce a litany of information that the user can obtain from the program. Think in terms of a Google Search on steroids. Sometimes the information would be “garbage in and garbage out” but more on that issue below. Commercial use of AI became the goal of the software developers of AI – how can AI be developed and marketed to assist specific industries in their use of computer based intelligent information processing? The goal was and is to monetize the applications of artificial intelligence to the public from how to apply in the logistics and warehouse industry, medicine, transportation, legal, educational, and research. The potential use of AI is endless.

AI IN MEDICINE IN WORKERS’ COMPENSATION CLAIMS

There are two aspects of artificial intelligence that exist in the practice of medicine from an analytical standpoint, not including such things as robotic assisted surgical procedures or other “hands-on” clinical practice. AI in medicine has two forms: (1) predictive analytics and (2) generative AI. Predictive analytics involves such things as AI indicating that a patient has a 75% likelihood of being admitted into an intensive care unit. Generative AI is more prevalent in the context of workers’ compensation related medical practice where, for example, a computer program using AI using a large language model writes an article. This author guarantees to you, the reader, that this article was NOT generated by AI. Generative AI involves relationships between people.

Further examples of Generative AI include patient-portal messages which can use conversational interfaces for patients to learn about their diagnosis, treatment options, or prepare for surgery (based on patient’s literacy level), or for patients to self-diagnose a condition. Can Generative AI be used for a medical-legal physician to “write” a medical-legal report? Can a medical-legal physician rely on a commercially available proprietary generative AI program to review and summarize medical records? There are AI companies who are selling the commercial use of their AI programs that claim, for example, that an accurate summary of 500 pages of prior medical records for an injured worker takes seven minutes for the AI program to generate. You are reminded that medical records review of over 200 pages are billed by the medical-legal physician at $3.00 per page pursuant to the medical-legal fee schedule under Title 8 Cal. Code of Regulations Sections 9793(n) and 9795. Is an AI generated summary of medical records in a litigated workers’ compensation case reliable, accurate, credible, and persuasive evidence of the actual records?

AI IN LAW AND IN WORKERS’ COMPENSATION CLAIMS

The use of AI in a workers’ compensation attorney’s law practice could include such things as researching statutes, regulations, and case law. AI based programs could write a legal brief, a legal article for a legal publication, or establish a best-worst case scenario for the outcome of a claim. AI programs could summarize deposition transcripts of injured workers, witnesses, or physicians. Can an AI application be used by a judge to write a Summary of Evidence, an Opinion on Decision? A Report and Recommendation on a Petition for Reconsideration or Removal?

The use of AI is already embedded in legal search engines that attorneys and judges use every day. When counsel enters a word or phrase into the LexisNexis data base, an AI assisted search engine can and will generate a list of statutes, regulations, and cases that may be pertinent to the search. Are those search engines accurate? Are trial briefs, Points and Authorities, medical or deposition summaries generated by an AI assisted search engine reliable, accurate, credible, and persuasive? Is a judge’s decision or response to a Petition for Reconsideration or Removal reliable, accurate, credible, and persuasive? Can an AI based program write a medical-legal report including providing WPI ratings of an injured worker or write predictive apportionment findings?

Here is an example of an advertisement for a Generative AI subscription that was advertised online:

With the most robust set of capabilities (sic) in the market, “NAME OF AI PROGRAM” helps you:

Review Documents: Ask complex questions about a batch of documents and receive a substantive analysis complete with citations.
Search a Database: Pinpoint relevant documents within a large database of your files.
Draft Correspondence: Draft tailored letters and emails with speed.
Summarize: Condense long, complex documents into succinct summaries.
Extract Contract Data: Obtain precise information about the content of contracts.
Timeline: Automatically assemble chronologies of events described in your documents.
Contract Policy Compliance: Provide a set of policies to identify non-compliant contract language and receive automated redlines to bring the contracts into compliance.
Prepare for a Deposition: Easily identify pertinent topics and questions for investigative projects of all kinds.

Does this generative AI program replace law clerks, staff attorneys, paralegals, secretaries, and first-year attorneys? Do you trust a computer application to guide your legal analysis of what may become a disputed issue? Where are the analytical skills about credibility or issue spotting? Can this program identify legal or factual issues that only a practicing attorney can determine? How do we know that if this generative AI program cannot find a legitimate legal citation that it will invent a fictitious one instead? What is really irritating about this is that speed is not necessarily quality, accuracy or reliability.

A generative AI program cannot replace an attorney’s gut feelings or ability to smell a rat or to simply know what to ask in a deposition while on the fly during a deposition. Sometimes an attorney’s instincts kick in and will establish a strategy just based on those instincts – which generative AI cannot accomplish. Generative AI does not have human intuition, feelings, or empathy.

OVERLAPPING ETHICAL ISSUES

The use of artificial intelligence in the context of workers’ compensation litigation raises significant ethical issues that need to be developed in order to keep pace with the usage of AI. Since no formal ethical code of conduct exists in the use of AI in workers’ compensation litigation, a discussion of some basic premises of ethics in medicine may apply.

The analysis of ethical considerations in the medical-legal context begins with the Belmont Report in 1979 that was adopted by the federal government to apply to any federally funded medical research that involved human participants for new drugs, biologics, or devices. This broad ranging mandate was codified under 45 CFR 26 called the “Common Rule” which applies throughout the United States and has been adopted in our own Health and Safety Code [see Health and Safety Code Sections 24170-24179.5]. While ethical requirements in human subject protections in medical research are mandated by law, no such mandate exists in use of AI in legal or medical-legal applications.

Since there is no law that governs how AI can be used or restricted from use in workers’ compensation litigation, the legal protection of human subjects in medical research community can be analogized to form a framework of protection against abuse of the use of AI in workers’ compensation claims. We are, after all, engaging in a form of social, medical, and legal research just by using artificial intelligence in proposed ways during the course of a workers’ compensation claim. We do not have enough data or experience to draw any conclusions about the short term or long-term effects on a claim or individuals involved in a claim when a party uses AI in the prosecution or defense of a claim. As of today, there are no legal or ethical guardrails in place to limit or regulate the use of AI in litigation. So how do we develop an ethical framework for the use of AI outside of the medical research community? We use medical research guardrails as a guide for the development of ethical usage of artificial intelligence in both medicine and the law.

The Belmont Report and 45 CFR 46 have a tripartite mandate:

1) Respect for Person – treat people individually and account for individual variances, perform research [or in our context – use artificial intelligence] in the best interest of a patient.

2) Beneficence: medical research must provide a benefit to society and improve diagnostics and the treatment of disease [AI should be available to everyone for the benefit to individuals and groups of individuals]

3) Justice: – apply the concept of equality in the selection of research participants [the benefits of artificial intelligence should be distributed equally among populations and individuals].

In addition to the proposed basis for guardrails for the use of AI in medicine and law, there is also the concept in medicine that medical processes follow, FAVES: Fair, Appropriate, Valid, Effective, and Safe. You are reminded that in the context of medical-legal evaluations in workers’ compensation cases in California, Title 8 California Code of Regulations Sections 41 and 41.5 govern the ethical considerations for all physicians who perform medical-legal evaluations. Someday there should be a provision in those sections that indicate that if any part of the medical-legal process is performed with the assistance of an artificial intelligence resource or program, a written disclosure statement shall be part of the physician’s reporting requirements.

POTENTIAL SHORTFALLS OF THE USE OF AI IN WORKERS’ COMPENSATION LITIGATION

There are a number of concerns about the use of artificial intelligence in the context of any form of litigation, especially in workers’ compensation cases. For the use of AI in both law and medicine, the FAVES factors should apply because AI can be misdirected to what is financially favorable to the doctor or claims administrator and not of ultimate benefit to legitimately injured workers. The use of AI by physicians and attorneys should be transparent, explainable, and subject to inspection. Remember, no one can cross-examine a computer or a computer program or application. How do you cross-examine a medical-legal physician who uses AI to (1) establish a diagnosis, (2) causation of injury, (3) determine WPI ratings, or (4) apportionment? An AI program cannot examine the injured worker, can it? Will it some day?

Those of you who are not familiar with the mechanism of artificial intelligence, there are some aspects of it that are very concerning. There are at least six machine languages that have been developed that can allow artificial intelligence programs to write its own codes. Generative AI can have a “hallucination” when it generates a false medical or legal citation. AI programs can deteriorate or drift from when it was first introduced. In addition, AI could invent its own data set that is not based on reality. This phenomenon is called “performance drift” and must be monitored by human-based evaluation and oversight.

At the time of publication of this article, there is an organization called the “Coalition for Health AI” (chai.org) which has developed what is called an “Assurance Standard Guide” that divides oversight into three categories:

(1) AI developer’s Responsibility – evaluate the AI model thoroughly before deployment to ensure it meets safety and performance standards
(2) End-User’s Responsibility – conduct local evaluations to ensure the AI tool fits the specific needs and conditions of the health system
(3) End-User’s Monitoring Responsibility – monitor AI tool performance over time, ensuring it remains effective and adapting to any changes in conditions.

The Coalition for Health AI is a public-private oversight organization involving academia, tech companies, and the federal government to develop a national quality assurance laboratory to evaluate the safety and effectiveness of AI in medicine (covering the concept of beneficence). The idea is to prevent AI from making financial decisions in favor of payers rather than decisions benefitting a patient (sounds like UR, doesn’t it?).

Remember, there is no legal mandate (legislative or regulatory) to require these guardrails in the development or use of AI in medicine or in law. The promotors and supporters of the Coalition include major, credible, medical groups including but not limited to UCLA Health, Mayo Clinic, Google, Johns Hopkins Medicine, Boston’s Children’s Hospital, Kaiser Permanente, UC Irvine, UC Davis, UC San Diego, and others. The Coalition plans on monitoring AI models use in medicine, developing best practice guidance for developing and deploying health AI technologies on a use case by use case basis, and to publish an AI “report card” on an accessible registry that has public access.

Is there a similar “Coalition for Law AI” that will do the same things as Coalition for Health AI? Not yet – the only “oversight” of AI-based programs currently being marketed to medical-legal physicians and attorneys is the market itself. Software developers are beginning to saturate the market to sell AI based programs to medical-legal physicians, claims administrators, and attorneys to help streamline the processing of information that is needed in the prosecution or defense of workers’ compensation claims.

These include programs that summarize deposition testimony, provide predictive case outcomes based on mechanism of injury and parts of body injured, set loss reserves, summarize 500 pages of medical records in seven minutes, analyze a mechanism of injury, develop, and send a client the “attorney’s” recommendations for further case handling, managing a law practice, and answering emails or phone calls from clients.

This raises a serious point: How much inter-rater reliability is there for a summary of medical records that is generated by an artificial intelligence program versus the medical-legal physician actually doing the summary as well? We would like to see a side-by-side comparison of an AI generated medical records summary with one that is actually done by a human QME or AME. Would a 5% variation be acceptable? There are no studies yet on this issue. Further, who does the claims administrator pay the $3.00 per page above 200 pages of records to be reviewed? Doesn’t that alone raise some significant ethical issues for QMEs and AMEs who use artificial intelligence programs to review and summarize medical records?

Artificial intelligence is currently embedded in MS Office (WORD especially) and now in a LexisNexis search. All you have to do is type a word or phrase into the search engine and AI will assist the user to obtain a data base. We already know that some AI based programs have gone awry – a Federal judge in New York received an AI assisted legal brief from an attorney who did not check the legal citations that were generated by the AI program. The judge did check them and discovered that the citations were a figment of the AI program’s imagination – the cited cases never existed. It did not take a computer program to generate sanctions against the attorney who filed the AI generated brief.

Counsel is strongly advised to check their work.

ETHICAL CODE OF CONDUCT?

Artificial intelligence is becoming part of our normal day to day lives. AI is being used even when you do not know it. Artificial intelligence programmers can take the likeness of any person, say Taylor Swift for example, and generate what is known now as a “deepfake” which generates her likeness in an AI generated image and uses her voice to say anything the programmers want that sounds like her real voice. The introduction of our AI seminar at the CSIMS conference in Coronado Island in August 2024 used the likeness of Scarlett Johansson and her voice in a video that was developed using AI. The image and sound were very real, but the actual person and her voice were not.

So how would the Belmont Report of 1979 along with the protections of human research participants apply in the context of the use of predictive analytics and generative AI in medicine and law? Respect for persons: (1) there needs to be transparency on how patient data is being used, (2) clarity of the role AI is being used in decision making, and (3) allowing regulators access to the algorithms. Beneficence: A patient should be able to decline using AI as part of the informed consent process. An injured worker should be told that the utilization review process may be determined by AI but the injured worker will be provided reasonable treatment to cure or relieve the effects of the injury that is based on the medical treatment utilization schedule in ACOEM upon review by a licensed physician and/or a licensed physician through the Utilization Review and Independent Medical Review processes of Labor Code Sections 4610, 4610.5 and 4610.6. Justice: any decision-making process or review of a record by artificial intelligence is subject to scrutiny by the Workers’ Compensation Appeals Board.

Here is an ethical issue: can a treating physician create an avatar who meets with the patient electronically? Is a physician obligated to disclose to a patient that some of the interactions between the patient and the doctor’s office is through an avatar or otherwise from an artificial intelligence-based application? Does a physician have to disclose that the probable outcome of surgery is based on a predictive analytics algorithm from an AI program?

An AI based algorithm has to be “fair” one that provides the same treatment recommendation for all patients with the same clinical features. Can AI undermine physicians’ or attorneys’ professional role as a fiduciary for a patient’s or client’s best interests? Ethical considerations exist in both the medical and legal fields of practice. Attorneys are bound by the Code of Professional Conduct [See Business and Professions Code Sections 6000 et. seq.] and physicians are bound by their own professional standards and ethics. Specifically, Title 8 Cal. Code of Regulations Sections 41 and 41.5 govern the ethical considerations for medical-legal evaluators.

DISCLOSURE-DISCLOSURE-DISCLOSURE!

There is no formal code of conduct in medicine or in law as to the limitations by practitioners of the use of applications programmed with artificial intelligence. There need to be guardrails along the use of both predictive analytics and generative AI in medicine and law. We need to look to the National Institutes of Health, the Centers for Disease Control and Prevention, and the federal Office of Human Research Protections for guidance. Meanwhile, the California Business and Professions Code or the Rules of Professional Conduct do not cover ethical considerations for attorneys’ use of predictive analytics or generative AI in a law practice. There has to be a movement to build public trust in the use of artificial intelligence in medicine and in the courtroom. A lawyer, like a doctor, has a fiduciary duty to their client. There should be a requirement that if a physician, an attorney or a judge writes anything using generative AI, the physician, the attorney or the judge has to disclose its use and to attest to its authenticity and accuracy.

After all, the attorney or physician owns what is written and has to defend its contents. The missing element from written articles or reports that are generated by artificial intelligence is the style or uniqueness of the writer’s prose. There is almost an innate ability to tell when something was written by a machine and not by a person. All of us have a certain style of writing and there is always a human touch to how it reads. This article for example has some clunky word usage to it that are a product of this author’s unique writing style. The tone and emotion of writing is missing from AI generated prose. You can tell it was not written by a human. It just does not pass the smell test. But the AI-based applications will improve over time.

The narrative of the concept of disclosure is not new or foreign in the practice of medicine or in the practice of law. Informed consent is the hallmark of any fiduciary relationship between a patient and their physician or between a client and their attorney. If any part of a workers’ compensation claim has been run through an artificial intelligence application by a physician or injured worker’s attorney, the injured worker should have knowledge of that fact. The metrics that are offered for claims administrators are limited as well – no one can predict the outcome of a claim – not every lumbar spinal fusion surgery has the same outcome. Predictive AI probably has very little use in the legal profession other than to give a claims examiner, risk manager, or defense attorney a “best case” and “worse case” scenario that a good defense attorney could already do just by reading the case file.

I SENSE DANGER, WILL ROBINSON!

Do you remember Robot in the television show, “Lost In Space?” So how far can a medical-legal physician rely on a currently marketed application that is based on generative artificial intelligence to write a medical-legal report? Can a physician utilize a program that uses generative artificial intelligence to write a summary of 500 pages of medical and legal records? What about our anti-ghost-writing statute?

Since this article is written about workers’ compensation claims and the use of predictive analytics and generative AI within the workers’ compensation community, a direct quotation of California Labor Code Section 4628 is appropriate. Labor Code Section 4628 is the “ghost-writing” prohibition that says the medical-legal physician writes and signs the report and must disclose who else contributed to the medical-legal evaluation process and report writing process. Here is Labor Code Section 4628 in its entirety:

4628(a) Except as provided in subdivision (c), no person, other than the physician who signs the medical-legal report, except a nurse performing those functions routinely performed by a nurse, such as taking blood pressure, shall examine the injured employee or participate in the non-clerical preparation of the report, including all of the following:
(1) Taking a complete history.
(2) Reviewing and summarizing prior medical records.
(3) Composing and drafting the conclusions of the report.
(b) The report shall disclose the date when and location where the evaluation was performed; that the physician or physicians signing the report actually performed the evaluation; whether the evaluation performed and the time spent performing the evaluation was in compliance with the guidelines established by the administrative director pursuant to paragraph (5) of subdivision (j) of Section 139.2 or Section 5307.6 and shall disclose the name and qualifications of each person who performed any services in connection with the report, including diagnostic studies, other than its clerical preparation. If the report discloses that the evaluation performed or the time spent performing the evaluation was not in compliance with the guidelines established by the administrative director, the report shall explain, in detail, any variance and the reason or reasons therefor.
(c) If the initial outline of a patient’s history or excerpting of prior medical records is not done by the physician, the physician shall review the excerpts and the entire outline and shall make additional inquiries and examinations as are necessary and appropriate to identify and determine the relevant medical issues.
(d) No amount may be charged in excess of the direct charges for the physician’s professional services and the reasonable costs of laboratory examinations, diagnostic studies, and other medical tests, and reasonable costs of clerical expense necessary to producing the report. Direct charges for the physician’s professional services shall include reasonable overhead expense.
(e) Failure to comply with the requirements of this section shall make the report inadmissible as evidence and shall eliminate any liability for payment of any medical-legal expense incurred in connection with the report.
(f) Knowing failure to comply with the requirements of this section shall subject the physician to a civil penalty of up to one thousand dollars ($1,000) for each violation to be assessed by a workers’ compensation judge or the appeals board. All civil penalties collected under this section shall be deposited in the Workers’ Compensation Administration Revolving Fund.
(g) A physician who is assessed a civil penalty under this section may be terminated, suspended, or placed on probation as a qualified medical evaluator pursuant to subdivisions (k) and (l) of Section 139.2.
(h) Knowing failure to comply with the requirements of this section shall subject the physician to contempt pursuant to the judicial powers vested in the appeals board.
(i) Any person billing for medical-legal evaluations, diagnostic procedures, or diagnostic services performed by persons other than those employed by the reporting physician or physicians, or a medical corporation owned by the reporting physician or physicians shall specify the amount paid or to be paid to those persons for the evaluations, procedures, or services. This subdivision shall not apply to any procedure or service defined or valued pursuant to Section 5307.1.
(j) The report shall contain a declaration by the physician signing the report, under penalty of perjury, stating:
“I declare under penalty of perjury that the information contained in this report and its attachments, if any, is true and correct to the best of my knowledge and belief, except as to information that I have indicated I received from others. As to that information, I declare under penalty of perjury that the information accurately describes the information provided to me and, except as noted herein, that I believe it to be true.”
The foregoing declaration shall be dated and signed by the reporting physician and shall indicate the county wherein it was signed.
(k) The physician shall provide a curriculum vitae upon request by a party and include a statement concerning the percent of the physician’s total practice time that is annually devoted to medical treatment.

CONCLUSION – FOR LAWYERS AND JUDGES

There must be a movement to build public trust in the use of AI in medicine and in the courtroom. A lawyer, like a doctor, has a fiduciary duty to their client. There should be a requirement that if an attorney or a judge writes anything using AI, the attorney or judge has to disclose its use. For goodness sakes, check your work! Double check the citations that are generated by the software and read the actual cases to verify the authority you are citing. No one can cross-examine a computer or its programming.

CONCLUSION – FOR MEDICAL-LEGAL PHYSICIANS

Is Labor Code Section 4628 a full stop for medical-legal physicians to use generative AI in their report writing process? Can a medical-legal physician use AI to summarize medical records? Could a judge disallow payment and deem a medical-legal report inadmissible because the evaluating physician was assisted by AI in the generation of the report? Regulations and case law may be necessary to answer these questions. In the meantime, we can look forward to some ethical considerations within the medical, medical-legal, and legal communities in the use of predictive analytics and generative AI since artificial intelligence in general is rapidly becoming part of our daily lives as human beings.

CONCLUSION – THE ULTIMATE GUARDRAILS FOR INJURED WORKERS

Is there potential civil liability of the owners and developers of proprietary artificial intelligence software that generates a deepfake image of an injured worker, their attorney, or a proprietary generative AI program that has an inaccurate medical record summary or claim analysis that a QME, AME, employer, or claims examiner relies on? The ultimate guardrail against harm by a software company who sells artificial intelligence programs to participants in a workers’ compensation claim is a civil lawsuit against the AI developers in Superior Court for damages in addition to costs, sanctions, and attorney’s fees in the workers’ compensation case at the WCAB against am applicant or defendant who misuses AI.

The ultimate responsibility of anyone who utilizes any form of artificial intelligence in the course of a workers’ compensation case is full disclosure by the person or persons who utilize AI during any step along the claims process. There needs to be regulations, industry standards, or other required ethical considerations that any use of AI by any person involved in a workers compensation case be fully disclosed to any affected participant in that case. Generative and predictive analytics by artificial intelligence does not have a human touch. No one knows what software was written by a human and what was written by a machine.

In addition, there should be required written disclosure that AI was utilized and how it was utilized with some form of assurance that a human being reviewed information that was generated by an AI program before any substantive decision making was made by a human being concerning all aspects of a claim. There is absolutely no room for deception in the course of a workers’ compensation claim since every judge has a duty to decide the rights and obligations of parties based on the evidence admitted at trial. That evidence has to be valid, reliable, accurate, credible, and persuasive. A computer software system that uses artificial intelligence cannot make those determinations for us. There must be a human touch from claim form to claim resolution.

Postscript: The author of this article wants to acknowledge the essay “The Ethics of Relational AI – Expanding and Implementing The Belmont Principles” by Ida Sim M.D. Ph.D. and Christine Cassel MD., New England Journal of Medicine, 391:3, July 18, 2024, pp. 193-196.

NAWCJ