I recently attended a 6 hours webinar on risk management for psychologists, presented by Leisl Bryant, PhD (1). In the segment of her presentation on AI she introduced the concept of “AI drift,” which struck me as extremely interesting and salient. In the broadest terms, AI drift refers to a slow process, the drift, in which we use AI to assist us with tasks and over time results in our offloading more and more of our thinking to AI. This can occur in a variety of ways. Dr. Bryant focused on how this risk exists for therapists (and I assume the same would apply to other professions, e.g., physicians) when we start to rely on AI notetakers to write our progress note because this will start to define how we conceptualize our work. Before proceeding, it is important to note that a number of social commentators and writers have addressed the idea that AI will influence how we think and approach the world. Ezra Klein, (2) the New York Times reporter and podcaster has delved into this concept and has also noted that the idea of how we receive information influences us as much if not more than the information itself. The importance of the “medium,” the way in which we receive information, goes back to Marshall McLuhan, and his seminal work, “The Medium is the Message,” (3) which focused on the idea that any new technological innovation changes how we relate to the world and to those around us. While McLuhan wrote before AI, and before the internet, his argument that technological innovation changes us remains as, if not more, relevant than ever.
Dr. Bryant identifies “AI drift” as the subtle tendency to become more and more reliant on AI as we use it more. She argues that we will likely begin by using AI as an adjunctive tool but that over time AI will end up as the authority. A simple example may help: many of us (myself included) use google maps or a similar AI based map to assist us with directions. I have noticed that over time I have become more and more reliant on google maps, almost automatically checking it before I start on any drive, even a short one. Now I can rationalize that this is a way for me to avoid road closures or delays, which it is, but it also results in my slowly and almost imperceptively finding myself questioning where a street that I have turned on my times is. Thus, a helpful tool, an adjunct to going somewhere, has now become more of an authority. Obviously, there are times when AI is right and our memory of how to get somewhere is wrong. I remember my mother arguing with me that google maps was wrong, as we went out to dinner, and that there was no road where I was going (there was a road, and google maps was right and my mother was wrong). While this is one of many poignant memories of my mom, it also serves to illustrate the benefits of AI (yes, google maps and other similar maps are AI).
For therapists notetaking programs, which are being marketed to us at a frightening rate, are just one example of where we are at significant risk, overtime, of becoming more reliant on AI. If you look at Berries (4) website the issue/risk of AI drift is glaringly apparent. Note: I am not singling out Berries as a bad actor compared to other notetakers. In fact other sites have a very similar menu or list of benefits, see Blueprint, Upheaval, Mentalyc, Abridge and others, as well as the notetakers embedded in electronic record programs. Serendipity resulted in my clicking on the Berries website. There are dozens of notetakers, some of which are part of Electronic Health Systems, and others which are freestanding programs. Again, I am using Berries to illustrate the drift issue, but basically all AI notetakers pose the same challenge. For example, Simple Practice is marketing “Care Aide,” which provides virtually the same services offered by the Berries site, and by other AI notetakers. The Berries site states that their note taker provides the following: a clinical note which the site states is “written perfectly in seconds,” diagnostic suggestions, treatment plans, summaries to prepare you for your next session, and an AI chat assistant that “helps you clarify your thinking.” This list of benefits is likely quite attractive to many therapists who may feel burdened by the demands of writing progress notes and updating treatment plans. However, this list also illustrates the risk of AI drift, where overtime therapists will likely be increasingly vulnerable to turning over more and more aspects of their thinking to the notetaking program.
To return to the concept of AI drift, one of the ways in which we are at risk in turning over more authority and thinking to AI is that AI sounds authoritative and as Dr. Bryan notes, AI will even provide citations and quotes. Again, the risk of AI drift is not limited to therapists. A number of news stories have highlighted how attorneys have submitted briefs that included made up cases (5). In a recent case in Mississippi (6) attorneys for both the defense and the plaintiff were cited for submitting briefs that included cases that were made up by AI and which never occurred (6). There have been examples about this occurring in medical records and concerns about this were recently highlighted in the television show, The Pitt.
I am not willing to attribute bad motives to the professionals in these cases. Rather, my benign hypothesis is that they were rushed/have very demanding schedules and many responsibilities, AI was available as a tool, and as Dr. Bryant and others have noted, AI sounds or appears very authoritative. This risk is compounded by the fact that much of what AI generates may be correct, and the hallucination, or made up material, occurs only occasionally. If AI was faulty a significant amount of the time we would all likely avoid it. Rather, it is the infrequent and unpredictable errors that make AI dangerous, particularly if we fall into a pattern of overreliance. It is also relevant to note that recent reports suggest that more advanced AI models may be prone to even more errors, i.e., hallucinate more often (7). Studies of notetaker use by physicians have also suggested the risk of errors (8).
The concern about AI doing our thinking for us has been highlighted in a variety of different ways. A similar concept to AI drift is what has been termed “cognitive offloading.” Cognitive offloading is best thought of as a process or scenario where we allow others, or software, to manage a task for us. This could be a scheduling or billing task. Reminders that we send and receive about appointments is a simple example of this process. Cognitive offloading has benefits. It may reduce stress related to having to follow up on and keep track of more mundane tasks. Thus, we are saved time and energy. However, cognitive offloading has its dangers. It may result in our becoming overly reliant on tools (e.g., AI). This overreliance will place us at increasing risk of thinking less about what we are doing and becoming less thoughtful. Chirayath, Premamalini and Joseph detail the benefits of and risks of cognitive offloading in a thoughtful article summarizing the thinking and research in this area (9). Clearly offloading and drift are highlighting the same benefits and risks regarding the use of AI.
Cognitive surrender is an additional term that has been used to describe the processes and concerns addressed in the discussions of AI drift and cognitive offloading (10). In many respects all three concepts are similar, focusing on how the use of AI in various ways may result in our becoming overly reliant on AI to do our thinking for us. However, the concept of cognitive surrender is a more extreme version of drift and offloading, as it is defined as a giving over of our thinking to AI without reviewing or questioning the information provided by AI. At its essence cognitive surrender involves an abdication of our critical thinking and a reliance on AI to do our thinking for us. In his article on cognitive surrender Kyle Orland explains the process of cognitive surrender and notes research done at the University of Penn. which found evidence to support the risk of cognitive surrender (10). A more acerbic piece, by the BBC, provocatively titled: “AI chatbots could be making you stupider” (11) makes the same argument in a more pointed and sharply written article. However, this article also cites a number of different pieces of research illustrating examples of how we are prone to cognitive surrender when using AI. Moreover, several of these studies show reduced brain activity associated with using a Chatbot, versus not using one, for a writing task, while other studies have shown people tend to remember less when they rely on AI to assist them with or to manage a task.
In sum, my concern remains that the risks of drift or offloading, or even cognitive surrender are subtle and pernicious. Over time we may become less thoughtful and more reliant on AI. While this will be beneficial regarding mundane tasks, and may reduce burnout and stress, it poses risks that may not be fully appreciated as we seek relief from the growing list of demands that many of us face in our professional and personal lives. While many experts and certainly all notetaking programs (designed for professionals) stress that we should proofread our notes and reports, the same forces that led us to seek out these services likely place us at risk of not consistently monitoring AI and offloading or surrendering our critical thinking.
References:
- The recent presentation by Leisl Bryant, PhD. “Sequence XVI: Risk, Relationship, and Responsibility: Ethics and Risk Management in the Web of Modern Times,” June 5th, 2026.
- Article by Ezra Klein, I Didn’t Want It to Be True, but the Medium Really Is the Message, on how AI can influence and shape us. https://docs.google.com/document/d/1_NrmjpaDSetbehTINTBScuZ_pTRuj1otZaRqSv33mQI/edit?tab=t.0 Note: Klein has talked about this issue in subsequent articles and podcasts.
- The seminal work, “The Medium is the Message,” by Marshall McLuhan. There are multiple summaries and podcasts about this book that can be easily found online. MIT press has excerpts of key pieces of the book, available for free at: https://web.mit.edu/allanmc/www/mcluhan.mediummessage.pdf
- Berries, an AI notetaker. The link to their features page: https://heyberries.com/features. Simple Practice is an EHR that also offers an AI powered notetaker.
- This article notes how a number of lawyers, including lawyers from prestigious law firms, have entered briefs that include made up cases. https://www.nytimes.com/2026/04/21/nyregion/sullivan-cromwell-ai-hallucination.html
- Mississippi attorneys fined and reprimanded for submitting briefs that included made up cases. https://www.nytimes.com/2026/06/09/us/ai-lawyers-sanctioned-mississippi.html
- The increased risk of hallucinations in newer and more powerful AI models. https://www.nytimes.com/2025/05/05/technology/ai-hallucinations-chatgpt-google.html
- Errors by AI notetakers. See a study of errors in AI notetakers used by physicians, https://www.sciencedirect.com/science/article/pii/S2949761225000999 and a more general discussion of this issue, in NYT article, https://www.nytimes.com/2026/04/28/well/doctors-using-ai.html?unlocked_article_code=1.e1A.Y2qF.USs1so4tXibA&smid=url-share&utm_source=Newsletter_Daily-Rundown-Member&utm_medium=WBEZEmail&utm_campaign=Daily_Newsletter_Daily-Rundown_Sponsored_20260430&utm_content=4/30/2026&DE=.
- An interesting article summarizing the risks of cognitive offloading. Ginto Chirayath, K Premamalini, and Jeena Josep, Cognitive offloading or cognitive overload? How AI alters the mental architecture of coping. https://pmc.ncbi.nlm.nih.gov/articles/PMC12678390/
- Cognitive Surrender and AI. An interesting article summarizing the concerns with links to and discussion of research on this issue. https://arstechnica.com/ai/2026/04/research-finds-ai-users-scarily-willing-to-surrender-their-cognition-to-llms/
- A BBC article on the deleterious effects of AI on our thinking, “Chatbots could be making us stupider?” by Melissa Hogenboom. https://www.bbc.com/future/article/20260417-ai-chatbots-could-be-making-you-stupider.