You are receiving this article because you are currently subscribed to S.P.I.R.I.T. newsletter. If the IRB Coordinator’s Diary doesn’t serve you, you are more than welcome to unsubscribe.
The IRB Coordinator’s Diary | The Raw Side of Research Ethics and Compliance.
Whether you’re a seasoned professional or an IRB novice, everyone can learn something from The IRB Coordinator’s Diary.
DISCLAIMER: Consider topics in this newsletter as food for thought. This is meant to be an educational resource only. This is NOT meant to provide legal advice or replace guidance provided by your institution. You should always consult your institution for specific practices related to research compliance or any legal subject matter presented in this newsletter.
Dear diary (and everyone else reading my diary),
Movies like M3GAN have shown us the dangers if AI was an entity that lived among us like other humans. But…what if we aren’t far from this becoming reality? What if AI entities and humans had to coexist? And if we were to coexist, what does this mean for the future of research within the regulatory landscape?
This article will explore a hypothetical nightmare for IRBs if AI entities lived among us. Specifically, we will explore the if AI was a:
- Principal investigator conducting human subjects research,
- Participant in a human subjects research study, or a
- Reviewer of human subjects research.
Before we get started, I encourage you to please share this newsletter with your network and subscribe.
Table of Contents
AI as a Principal Investigator (PI)
Imagine an AI serving as a PI on a human subjects research study. At first glance, this might sound appealing. An AI PI (admit it, you laughed too) could theoretically:
- Process vast amounts of literature in seconds,
- Generate research protocols almost instantly, and
- Never miss a deadline.
However, the regulatory landscape becomes considerably more complicated when the individual responsible for protecting human subjects is not actually human.
Current regulations place responsibility and accountability squarely on the shoulders of a human PI. When protocol deviations or adverse events occur, a human is held accountable. But what happens when the PI is not a human?
- Can an AI PI truly exercise ethical judgment when unexpected situations arise?
- Can an AI PI weigh competing interests, recognize subtle signs of participant distress, or make decisions that prioritize participant welfare over study outcomes?
- Who bears responsibility when things go wrong: the institution, the developer of the AI PI, or the AI PI?
Perhaps the greatest challenge for IRBs would be determining whether an AI can satisfy the fundamental expectation that investigators respect and protect human subjects.
Regulations imply that PIs possess not only expertise but also a moral compass that guides them in conducting responsible human subjects research.
While AI may be capable of following rules, it remains unclear whether it can genuinely understand the ethical principles underlying them (let alone human emotions such as distress). The nightmare scenario for IRBs is not an AI that intentionally acts unethically. It is an AI that follows every instruction perfectly while missing the human element that ethical research demands.
(Back to the Top)(Back to the Contents)
AI as a Study Participant
Now consider the opposite scenario: an AI entity enrolled as a participant in a human subjects research study.
You would think the idea of AI serving as a research participant sounds like science fiction. Yet researchers are already encountering versions of this reality. In a 2024 qualitative research study, investigators unexpectedly discovered AI-generated responses within their dataset while conducting online research. The authors described the challenge of identifying “AI-as-participant” responses and raised concerns about how generative AI may blur the line between genuine human participation and artificial responses. As AI becomes increasingly capable of producing human-like narratives, researchers may find it difficult to determine whether they are collecting data from people, machines, or potentially both.
At the same time, some scholars are exploring whether AI can intentionally serve as a participant substitute. A recent study proposed using trained AI systems as experimental participants in behavioral and economic research, allowing investigators to create simulated participant groups that serve as rational, bias-free, or otherwise controlled comparison populations. It’s argued that AI participants could reduce recruitment costs, accelerate studies, and help researchers test hypotheses under conditions that would be difficult or impossible to create with human subjects alone.
For IRBs, this creates a fascinating regulatory dilemma.
- If AI-generated responses are mixed with human data, what obligations do investigators have to disclose this to participants?
- If researchers intentionally use AI participants, is the resulting activity still human subjects research?
The nightmare scenario is not necessarily an AI entity demanding informed consent or asserting legal rights.
It’s a future where human participants and artificial participants become blurred.
IRBs may find themselves reviewing protocols where the central question is no longer how to protect human subjects, but how to preserve the scientific validity, transparency, and integrity of research where participants may not always be human.
(Back to the Top)(Back to the Contents)
AI as an IRB Reviewer
AI serving as an IRB reviewer may be the closest scenario in which AI entities truly become reality. Regulatory agencies are already experimenting with AI-assisted workflows. FDA’s Elsa 4.0 was developed to help agency staff analyze, summarize, and navigate large volumes of information. Likewise, recent research such as Chat-IRB for LMICs: an opportunity for ethics review capacity-building and protection against ethics dumping, IRB shopping, and other exploitative research practice has proposed supporting IRB operations. Rather than replacing reviewers, these systems would provide assistance.
Anyone who has served on an IRB knows that committees routinely face increasing workloads, complex protocols, and pressure to provide timely reviews. An AI reviewer could instantly identify missing consent elements, compare protocols against institutional requirements, flag inconsistencies across submissions, and surface relevant regulations or precedent decisions. In an interview discussing the Chat-IRB concept, the author emphasized that properly designed systems could help standardize reviews, reduce administrative burden, and allow reviewers to focus their attention on the most challenging ethical questions rather than routine compliance checks.
But this is where the hypothetical nightmare begins.
Imagine a future in which an AI reviewer becomes so accurate and efficient that committees increasingly defer to it by default. Over time, reviewers may not introduce varying perspectives. The danger isn’t that the AI reviewer would ignore regulations.
The danger is that IRB review is not solely about regulatory compliance.
It requires judgment about fairness, community values, participant vulnerability, and whether a study respects the individuals it seeks to enroll. Ironically, the more reliable AI becomes, the easier it may be for humans to disengage from the very ethical deliberation that makes IRBs valuable. The future challenge for IRBs may not be deciding whether to coexist with these AI entities but determining how much ethical authority should ever be delegated to it.
(Back to the Top)(Back to the Contents)
Leave a Reply