Threats to the Beginning and Ending of an Analysis in the Times of Technology and AI

I complained to my analyst after finishing my analytic training that nobody comes to me asking for an analysis. I had recently completed my analytic training which included three analytic cases under supervision which continue ongoingly, and I have been eager to work analytically with more of my patients. While I have a steady stream of referrals, and a full practice, many only want to meet once a week for psychotherapy, hoping for a practical solution to whatever they think ails them. He reassured me, uncharacteristically, with, “Nobody really comes to an analyst anymore asking for an analysis. You have to offer an analysis, when it is fitting, anyway.” Leaving my personal hesitancy and resistance to offering analysis to the side for a moment, I have heard from other colleagues, including more seasoned ones whom I admire, that it is more challenging these days to initiate an analysis with new patients, and also once an analysis is underway, it can be equally difficult for patients to terminate an analysis. In what follows, I will suggest that our current technologically-mediated environment including ever-increasing use of AI, has shaped our collective thinking in a way that makes both beginning analysis and terminations more challenging for both the analyst and patients or prospective patients.

In Freud’s times, psychoanalysis was sought out and venerated. This was perhaps largely because there really were no other competing treatments that were compelling or offered hope of a cure or improvement for various forms of mental illnesses. Concurrently, there was a general acceptance of the existence of limits to what one might hope for in life regardless of wealth or status. The ordinary developmental experience of infantile omnipotence as elaborated by Winnoctt (1953), was more likely to be limited by the environment including the mothering environment, but also the larger social environment. These limits to omnipotence were experienced collectively in the form of frequent deaths in childbirth (for mother, child or both), lack of reliable birth control as well as no access to IVF, childhood illnesses that we have largely since irradicated (although many seem to be rearing themselves again in the context of our politics around vaccines and relative individual focus), wars and their aftereffects, economic disasters and instability, lack of understanding of how to address common disabilities, not to mention the effects of anti-semitism, homophobia, racism, and subjugation of women. While we still have a number of these issues and we may be on the precipice of new forms of these problems or even newer large scale destructive forces that will provide significant limits in the current times, in the age of the smart phone and AI, we have access to information at our fingertips and this provides us with a felt sense of omnipotent control. This information may not be accurate and certainly has economic and political interests baked-into it, not to mention other biases, but we now have always-on information at our fingertips, nevertheless.  We don’t always know how to interpret fact from fiction within this source of information or how to use the power of AI to do tasks with the information, for “good aims” and further, we don’t all agree on what is “good” to do. Regardless, spending time searching for information or using AI fuels our fantasies of omnipotent control in a way that reinforces itself, particularly because we can’t actually tell yet what the limits of AI are and are not and therefore what our limits are to our omnipotence. Studies abound on how this is impacting us (Ahmad et. al, 2023; Dergaa, 2024; Stein, 2025; Tubali, 2024), but the most concerning impact for psychoanalysis is technology and AI’s reinforcement of our omnipotent fantasies, which I believe makes it more difficult to start and end an analysis, the very thing that might protect our freedom to think. I am thinking of the case where a very wealthy tech entrepreneur is asking for help from his analyst and says he wants a real transformation. He continues to ask AI to help him hack his life but doesn’t discuss these queries with his analyst. He also has an executive coach, whom he pays well, to provide advice and does not disclose these to the analyst. When the transference is beginning to develop with the analyst, he finds this very uncomfortable, cites what his coach has said and AI suggests and leaves, never really allowing the analysis to begin or allowing for a proper termination to the not-yet-begun analysis.

At the same time, each day we are being told that AI is closer to super general intelligence and that it is soon to be smarter than humans (e.g., has surpassed experts whom one may have sought out for consult, such as psychoanalysts). Trillions are being poured into AI companies to develop these very powerful technologies so that we can supposedly improve any number of things that we don’t all agree are necessary, such as environmental threats to public health. But there are also others using AI to try to live forever, upload consciousness into AI to advise us in the future, create bot replicas of dead loved ones and heroes, supplant human relationships and embodied partner sex, and more questionable or dangerous aims.

My clinical practice, which is located in the San Francisco, Bay Area, where the largest concentration of AI companies in the world is located, is filled with patients who are software engineers, AI executives, people married to these people or dating them and people working to serve them. We have been told that soon AI will do much of the manual labor and many boring jobs and we will all be able to enjoy significantly more leisure and interesting work. It will be a major upgrade in health and happiness. More recently, I have been hearing from these same tech executives that they too worry about our future and also foresee not only boring work being supplanted but their own work too. We are told that slowing down to consider if this makes us ill or AI ethics for the public good including its biases and unequal access to its power is not an option because development will be too slow to compete with those of other nations. AI poses a national security threat and could create deep economic issues for many therefore it is an arms race of sorts. Most users I know are busy fiddling around with ChatGPT, marveling at the game of prompts and its responses like a silly party trick. Sophisticated technologically literate users of AI are asking it to write code or grab code that is already written to accomplish this or that quickly, making lower-level engineers obsolete, garnering tremendous profits and developing creative products and services or public use projects at rapid speeds. Even more sophisticated strategic users, some of whom have nefarious aims, are using these tools to replace entire governmental functions and social institutions without the rest of us noticing much, voting or understanding fully what is happening. Regardless, all of us seem to be feeling that technology development cannot be influenced and further that we cannot take any time, human time, to ponder or wonder or fantasize or critique any of our experiences.  We are becoming more like robots. Here I am thinking of the patient who told me once that his mother is the Internet, refusing to consider any associations to his flesh and blood family members and historic past. He felt that the Internet was more available and immediately soothing—why begin an analysis? These are the patients who feel that psychoanalysis is simply too slow, not guaranteed to produce any particular outcome and too time consuming to be beneficial. There is also the opposite issue, where patients sometimes want to remain in our office multiple times per week until one of us dies.

When prospective patients come to us for psychotherapy and do not ask for an analysis, as they often do not, they tend to use technology to find us. Even if they are referred by a trusted person, they still often turn to technology to find out a bit about our practice or our person before calling us. Several recent referrals to me stated something like the following in their initial outreach to me. “When looking for a therapist, I googled and found you.” Or “My friend so and so recommended you and I googled to find your office to see if it would be in a convenient location and so I am calling to set up a consultation.” This is all relatively normal to my mind. We have become accustomed to a form of online always-on brochure-ware that people can use to understand basic facts about our clinical practices. We expect they might google us and we are prepared to include anything they might find out about us as material for the transference to be handled in the treatment.

Recently, however, something different occurred.  Someone called me and announced in that call, “I asked Claude (Anthropic’s AI assistant) to tell me what therapist in my area would be best for me.”, by way of describing how it is that he came to reach out to me.  I asked him if he had any ideas about what he thought Claude might be “thinking” to recommend me to him.  He didn’t, but he was interested in following the recommendation because ostensibly Claude “knows” something more than he does and he was happy to benefit from this capacity in service of “getting the answer right” and “optimizing his selection of therapists”.  I asked if he had any fantasies about how the work with me would unfold.  He didn’t, he claimed.  And we began with scheduling our initial session.  He asked the usual questions that patients often ask at the beginning of either psychotherapy or psychoanalysis, such as, “Do you think you can help me with (fill in the blank thing he was concerned about)?” and “How long do you think it will take?”.  These were questions that Freud (1913) also noted in the early beginnings of psychoanalysis were commonly asked.  To begin an analysis in earnest, however, is neither to start a relationship with the goal of necessarily addressing a particular concern (although many of Freud’s first patients had specific hysterical symptoms that were to be addressed), nor is it to claim that success is certain or even likely (we actually don’t know most of the time what is possible in the work even if we hope that all kinds of wonderful transformations happen for our patients and for many this is the case), and certainly it is not to assert any of this might be accomplished by a particular time frame.  We must communicate our conviction in the analytic process while abstaining from answering these concrete questions.  An analysis is precious because it lives in a place of open-ended timelessness, albeit scheduled in 45-or 50-minute (and sometimes other lengths of time) intervals until it is complete.  We help our patients feel the childlike wonder of timelessness and to face the issues associated with not knowing.  These can be transformative in ways that any other process may not be.  The patient in the first (and all subsequent sessions) is to be “….left to do the talking and free to choose at what point he will begin.” (Freud, 1913, p. 134) and the analyst provides evenly hovering attention and analyzes and the transference develops and deepens. The fundamental rule of psychoanalytic technique must be imparted to the patient at the very beginning, Freud notes (1913) and most analysts also assume some form of these understandings for the beginning of an analysis.

The paradox of this is that most of Freud’s patients were not in open ended analyses (Gabbard, 2008) because many of them travelled to visit him for a set period of time.  In fact, these treatments were often shorter than current analytic treatments on average, and largely with wealthy hysterical patients.  Now analysts work with a broader range of types of patients and tend toward both less frequent meetings and longer treatments for a host of reasons including practical ones, and that there seem to be greater resistances to beginning analytic treatment and to ending treatments.   An entire issue of the Psychoanalytic Quarterly was just published about terminations discussing how many prominent analysts often stop taking new patients as they consider retirement and then terminate with some of their patients but continue until they die with others, with whom a termination does not seem possible.  More and more analysts are acknowledging that there are more and more patients who once they begin an analysis, they are not sure how or if a termination is possible.  Current analyses are simply longer when they do begin and some of them never terminate until one of the dyad dies because they are based on maintaining the transference neurosis or dependencies that are not resolved.  This has always occurred and is a matter of technique, but is happening more and more now because once in an analysis, against the backdrop of always-on technology, we collectively seem to believe that we are entitled to be comfortable and soothed.  What is missing in many of these analyses is the gift of termination, a transformation of the original transferences that releases vitality that is available to the analysand after the treatment ends.  This is what we strive toward and this is so much harder to accomplish in the current AI-mediated context where limitations are not universally understood and grandiosity and omnipotence are reinforced in the ways discussed above.

In the age of AI, our prospective analytic patient also often encounters other forms of treatment in advertising or elsewhere using language like “evidence-based” or “gold standard treatment”. (BTW, psychoanalysis is evidence-based but we don’t often state that in our public communications;  we are more focused on the transference matrix than assertions of our evidence base.) They have access to Talkspace or Betterhelp which promise access to a licensed therapist on demand who you can cancel and find a new one anytime all of which is significantly cheaper and faster than an analysis, so they claim. And they have access to ChatGPT or any number of AI outfits which they can use to create a customized bot that can be their lover, therapist, or friend or any other fantasied transference-based personas in precisely the way that they wish. We tell them that we cannot say how long it will take to help them with their concerns and we ask them to tell us about themselves and we tell them something of the fundamental rule of psychoanalysis and we listen allowing the transference to develop. This prospective patient usually begins to worry at some point (unless they are in the field or have had an analysis already) usually when the transference is difficult, that this is a sham because we are not promising anything particular and often are not saying much at all, or they worry they would be better off meeting once a week like their friend does and why is it that CBT is only 10 sessions or Betterhelp is cheaper?  They might even turn to Claude again for advice when the transference is difficult, or they experience significant resistances. If we are lucky, they have a positive transference to us and this deepens but then we have to worry about being idealized too much, limitations being ignored (ours and the patient’s) and the termination becoming difficult. Claude after all is not only always on but will be available forever!  Why terminate with a person if you can have AI and it will never leave?!

In following the fundamental rule, Freud notes, “However genuinely the patient may believe in his excellent intentions, the resistance will play its part in his deliberate method of preparation and will see to it that that most valuable material escapes communication. One will soon find that that the patient devises yet other means by which what is required may be withheld from the treatment. He may talk over the treatment every day with some intimate friend and bring into this discussion all the thoughts which should come forward in the presence with the doctor. The treatment thus has a leak which lets through precisely what is most valuable” (Freud, 1913, pg. 136). But there are other new leaks in the age of AI.   Some of my patients have reported periodically asking ChatGPT to help them think about their developing transference to me. “My therapist said (fill in the blank interpretation).  What might that mean about our relationship?” Some have recorded sessions without divulging this in the moment, only later to realize that this not only breaches their confidentiality but creates a new kind of leak in the treatment, an intolerance of loss and limitation. These are new kinds of leaks are interesting.  For one, asking or telling AIs about what happens in one’s therapy or analysis teaches the machine more about human experience which is then used more broadly rather than in personal and individually meaningful ways and we cannot yet be sure what it means to share such private material with AI or what these companies do with this material. What about confidentiality and what about the idiosyncratic importance of the singular relationship? But more importantly for our patient it leaks from the transference and weakens the analytic relationship. It generalizes and ignores the particular and drains the transference heat. It creates a bunch of omnipotent fantasies—of the analyst, of the self, or the ChatGPT or all of the above.

Freud continues, “Often enough the transference is able to remove the symptoms of the disease by itself, but only for awhile— only for as long as it itself lasts. In this case the treatment is a treatment by suggestion and not a psychoanalysis at all. It only deserves the latter name if the intensity of the transference has been utilized for the overcoming of resistances. Only then has being ill become impossible, even when the transference has once more been dissolved, which is its destined end” (Freud, 1913, p. 143). We begin an analysis with the idea that we will hopefully work through whatever transferences there are toward termination—and termination, Blum notes (1989, p. 283) “…began the day analysis commenced.” If we are forever either idealized or AI is idealized in the transference as the source of the cure of symptoms, we will not be able to help our patients confront our limitations, their limitations and therefore their strengths and source of vitality. We also may not be able to help them put technology and AI in its rightful place, as a useful but imperfect tool to be used from time to time.

This is even more complicated because “the analyst’s own analysis is the model he or she uses to conduct analyses…” (Blum, 1989 pg. 275) with patients and yet, the context is entirely different even for those of us like me, who began analytic training just 6 years ago! The goals of analysis were, according to Freud, to produce freedom in the patient to work and love. But what is work in the age of AI when we are told that many jobs will be obsolete, or that working altogether will be a thing of the past in order that we might have infinite leisure? And what is love in the age of love and sex bots and dating apps, Ashley Madison, Field.com and so on, or chat bots that can be taught to be our perfect lovers, where you can have love and sex your own way at the click of a button or two? Blum (1989) notes,

I am beginning to understand that the apparent endlessness of the treatment is something that occurs regularly and is connected with the transference…the prolongation of which is a compromise between illness and health that patients themselves desire, and the physician must therefore not accede to it” (p. 279)

The current turn toward longer and longer analyses, avoidance of interpretation of resistance and instead providing relational perspectives may serve to avoid limitations, foster dependencies and avoid terminations in some cases. “Our aim will not be to rub off every peculiarity of human character for the sake of a schematic ‘normality’, nor yet to demand that the person who has been ‘thoroughly analysed’ shall feel no passions and develop no internal conflicts. The business of the analysis is to secure the best possible psychological conditions for the functions of the ego; with that it has discharged its task” (Freud, 1931, p. 249). Knowing what are the “best possible psychological conditions…” can be difficult to ascertain if we haven’t confronted our own limitations and omnipotence sufficiently or if we are worried that an AI will soon replace us and we need our patients as much as they need us to pay our bills.

Blum goes on to note, “There are inherent real limitations of the patient, analyst and the analytic process” (1989, p. 286). “Timeless infantile wishes require renunciation or sublimated transformation” (Blum, 1989, p. 286-287). But what are limitations and sublimation in the age of AI where all of us who use technology are being sold each day a version of omnimpotence, some of which may actually be on offer. For example, who is to say that the great AI revolution won’t produce miraculous gains in public health, help us avoid environmental disaster as well as provide new forms of work that will be beneficial to many if not all of us? We simply don’t know yet. And we might need some of the grandiosity of our AI technocrats to get us out of many of the predicaments we face environmentally, socially and politically. We do know that whatever wonderous good will be available it will come with a serious cost to our minds and perhaps much more.

Shane & Shane (1984) note, that in the termination process, the stability of the analytic achievements is tested and sometimes unsuspected vulnerabilities, structural faults and primordial transference may emerge in this phase. Initial diagnosis and analyzability may be evaluated again in ending, along with the decision to terminate. Termination is a trial analysis of the patient’s progress and a trial of the analysis itself. And if in these doubtful moments a patient asks Claude if it is time to terminate, or if their analysis was good in retrospect, what will it say?

Prospective patients come to us often not requesting an analysis and if they do, they have a fantasy about what that analysis will be like and what it will accomplish for them. Often it is not a fantasy of a good enough analysis (Gabbard, 1989) or one where they will feel a great deal of longing, early developmental pain and suffering, and have to work through cherished and somewhat useful for now defensive structures while reconciling their desires and limitations and yet this is precisely what might be most helpful for the analysand’s highest good. To freely associate is often to say things that are surprising and maybe shameful or difficult. To experience the heat of the transference can be very challenging and to sign up has always required a willingness to relinquish omnipotent wishes; in the time of AI when we are continually promised a correct answer at our fingertips is fairly vulnerable and risky. Why on earth would one voluntarily sign up to recognize one’s limitations and work through pain for an indeterminate amount of time with no guarantee or promise of cure, when one can hang onto omnipotent fantasies, fantasies long familiar to psychoanalysis and now easily mobilized in relation to AI, one of which is that the analyst will provide a magic and complete cure. Yes, patients need to idealize analysts. Yes, analysis can be magical at times, but if it is carried out fully to its conclusion it will strip both analysands and analysts of our omnipotence and confront us with our humanness and our limitations. At the same time AI appears to offer a horizon of limitlessness, or at the very least reveals that we don’t really know what the limits are. These are God-like powers. They may in fact save us from unspeakable horrors in the environment and so on. But they also keep us from mourning or confronting ourselves in significant ways, something philosopher and psychoanalyst Jonathan Lear contends is what makes us human and what makes life meaningful (Lear, 2022) and which I contend makes it very difficult to begin and end analysis.