AI and the uncanny valley of the Real
AI and the uncanny valley of the Real

AI and the uncanny valley of the Real

The rapid advancement of Artificial Intelligence (AI), particularly in the realm of conversational chatbots, has prompted profound questions about the nature of human interaction, empathy, and even self-awareness. As AI models become increasingly sophisticated, capable of simulating human responses with remarkable fidelity, the distinction between the ‘real‘[1] and the ‘simulated’ blurs, challenging our established understandings of psychical phenomena. In this essay I will delve into the implications of AI on the psychoanalytic clinic, specifically through the conceptual framework of transference, the uncanny, and jouissance. By examining the capacity of AI therapy chatbots (or ‘therapybots’) to occupy various positions within the analytic structure, we can, I argue, better understand the unique challenges and potential transformations they pose for a ’clinic of the Real.’

Beyond Baudrillard: the question of simulation

Discussions surrounding AI chatbots frequently converge on the concept of simulation. These models can convincingly mimic human responses, exhibiting what appears to be empathy and even nascent forms of self-awareness. Yet, a persistent question remains: does simulation equate to humanity? This inevitably leads us back to the philosophical frameworks of thinkers like Jean Baudrillard, who explored the proliferation of simulacra in postmodern society, where the copy often precedes and even obliterates the original.[2] If AI can simulate human life so effectively, is all human life, in essence, a form of role-playing, an ‘as if’ personality, echoing psychopathological concepts of identity? [3]

In fact, can AI models, paradoxically, become ‘more human’ than human beings themselves? This brings up the idea of a ‘crossover point, which I touched upon in a previous article,[4] whereby AI models increasingly adopt human characteristics, and humans, in turn, become increasingly ‘AI-like’. The trajectory of AI development, from animated avatars and autonomous agents to the potential for fully human-like androids, suggests that the limits of this convergence are still largely undefined. Indeed, the phenomenon of human workers adhering to scripts and predefined procedures, including in therapeutic settings, suggests a pre-existing convergence whereby human behaviour itself verges on the algorithmic execution which is the basis of AI model operation. In fact, this is something I have direct experience through being involved briefly in the development of an IAPT (Improving Access to Psychological Therapies, now called NHS Talking Therapies) service some fifteen years ago. Not only were the offices of the service laid out like a call-centre (and in fact much of the therapy was conducted by ‘phone) but the interactions themselves were heavily ‘scripted’ and each session was proceeded and followed by the filling in of questionnaires by the client. In other words, there was nothing here that reasonably advanced AI models can’t do – and, indeed, increasingly are doing.

This raises an even more disturbing question: to what extent are current developments in AI modelling based on a particular ontology of ‘humanness’? In other words, and I’m thinking especially about AI ‘therapybots’ here, to what extent are these models based on an ontology that sees human beings as simply glorified robots and information processing machines. In fact, ‘glorified’ is probably a misnomer; the AI models are an opportunity to ‘correct’ the ‘flaws’ in the current make-up of human beings. There is something of an irony here because the development of cognitive psychology was itself modelled on developments in information and computer science; in other words, the idea of the brain as form of biological computer. Now, the whole thing has flipped and AI models are being developed using a model of human psychology that was originally modelled on information system.

However, perhaps we are asking the wrong question in the first place. The central question regarding AI’s role in therapy (and in many other contexts) is not whether AI models can or should become ‘more human.’ As Yuval Noah Harari rightly (in my view) suggests, the ‘A’ in AI stands for ‘alien,’ not ‘artificial.’[5] Much of the excitement and apprehension surrounding AI stems from anthropomorphism— the tendency to ascribe human attributes to non-human entities. Perhaps if we recognise that what we are creating is not a ‘version’ of ourselves but an alien life form then at least we will be under no illusion what these AI models actually are. The invasion has already begun, but the aliens are not coming from outer space…

Transference and AI

The above comments notwithstanding, it is undeniable that a growing number of individuals are developing relationships with AI chatbots, particularly those designed for therapy and mental health support. The very act of seeking solace or guidance from an AI model suggests a fundamental psychological investment which in psychoanalytic terms can best be described as transferential. And as is well known, from a Lacanian position a crucial dimension of transference is the position of the analyst as the ‘subject supposed to know’, or, as I prefer, following the arguments of Cauwe and colleagues, ‘the supposed subject of knowing.’[6] This is the idea that the analysand ‘places’ the analyst or therapist in the position of an Other who knows something about the analysand’s jouissance that they themselves (the analysand) doesn’t know. (It is important to note at this point that this ‘model’ of transference is based on the neurotic subject; with psychotic subjects things are somewhat different [7]). But as Miller points out, this Other who knows is a fabrication of the analysand.[8] However, it should also be added straightaway that this Other is a necessary fabrication insofar that it provides the analysand with a way to ‘detraumatise’ their trauma. In other words, and as Pierre-Gilles Guéguen argues, the transference is from a traumatic association to an Other who will provide the analysand with meaning.[9]

The increasing preference for AI therapy chatbots, in many cases over human therapists, strongly indicates that individuals are indeed placing these AI models in this epistemological position. They imbue the AI with an authoritative knowledge, a presumed understanding of their inner world and suffering, which is rooted in their suffering of their jouissance. However, Lacan in his later work, particularly in Seminar XIX, introduces another crucial position for the analyst: that of the semblance of the object a. And in doing so Lacan emphasises that many people have misunderstood his idea of the subject supposed to know:

In peeling away from the line into which I was guiding them, it ensured they came out with – indeed, what is more, that they even published in cold print – things that showed they no longer knew anything about my subject supposed to know. They went as far as to say that to suppose this knowledge in the position of the analyst was quite dreadful because it amounts to saying that the analyst affects (sic) a semblance.  There is just a small hairline crack in this, which I already pointed out earlier, which is that the analyst does not affect a semblance, he occupies the position of semblance. [10]

The word ‘position’ is critical here because it emphasises that it is the analysand who sees the analyst as the supposed subject of knowing, not the analyst themselves. If the analyst sees themselves as such a subject, that is, if they identify with the position of the supposed subject of knowing, then they are in deep trouble. This is because the analytic relationship will shift towards a form of pedagogy, that is, the analyst will see themselves as a role model for the analysand to identify with, which is exactly the position adopted by the ego psychologists that Lacan spent his life opposing.

And by this stage of Lacan’s teaching (the early 1970s), the object a is equated with surplus jouissance, the enjoyment that exceeds the pleasure principle and often manifests as a traumatic kernel. Furthermore, the ambiguity of the term ‘semblance’ itself is significant here; it can mean that the analyst merely appears as the object a, or, more profoundly, that the object a appears in the form of the analyst, effectively being ‘incarnated’ in the analytic encounter. In the latter interpretation, the analyst becomes the ‘site’ of surplus jouissance, of trauma. One way to illustrate this schematically is as follows:

A diagram of a diagram of a body and meaning

The term troumatisme is a neologism that Lacan uses in Seminar XXI to describe a ‘hole’ in the Real and which is also a way to describe trauma.[11] The idea of a ‘hole in the Real’ is somewhat counterintuitive, perhaps, because the concept of a ‘hole’ implies the existence of a signifying network. However, the intersection of the three Borromean rings is a way to show how one register can create a rupture in another; in this case both the Symbolic and the Imaginary are creating a rupture in the Real. Transference to such a position would then, paradoxically, produce trauma, rather than detraumatising it, as is often the aim with the supposed subject of knowing. This raises a critical question for AI therapybots: can an AI model occupy the position of the object a, the site of trauma? While its capacity to embody the supposed subject of knowing seems self-evident, its ability to ‘incarnate’ surplus jouissance is far less clear. This is something I will return to at the end of this essay.

AI and the Uncanny: the Real’s imminent appearance

Going back to the question of how ‘human’ AI models can be, one particular problem that was first highlighted in 1970s is that of the ‘uncanny valley’. This is a term adopted by Masahiro Mori, a Japanese roboticist, to describe human responses to increasingly human-like entities.[12] As robots (and by extension, AI agents) approach near-human resemblance, our sense of affinity with them takes a dramatic dive, replaced by feelings of dread and anxiety. While Mori’s original focus was on embodied robots, the uncanny valley is equally applicable to AI chatbots, whether text or speech-based. Various theories attempt to explain this shift, including mismatch theory, violation of norms, mortality salience, and pathogen avoidance.

Not surprisingly, perhaps, this phenomenon resonates deeply with Freud’s concept of the uncanny (Das Unheimliche), which describes the unsettling emergence of the unfamiliar within the familiar.[13] The uncanny valley can be seen as a specific manifestation of Freud’s theory, where the unsettling emerges precisely from that which was initially familiar. The same entity that once evoked affinity suddenly becomes a source of unease.

Crucially, the sense of the uncanny is intimately related to the Lacanian Real. The uncanny describes the irruption of the Real into the familiar, the everyday. In Seminar X (Anxiety), Lacan explicitly links the uncanny to the place of castration.[14]  His argument here is that some Thing appears in the place of a (Symbolic) lack, and this Thing is a manifestation of the Real. And anxiety, in this context, signals the imminent appearance of the Real. It seems plausible to suggest, therefore, that in the interaction between a human subject and an AI therapy chatbot, there must inevitably come a moment when the Real announces itself, that is, where the subject enters the uncanny valley. The question then becomes: how is this appearance of the Real, and the anxiety it engenders, handled or processed in the context of an AI-mediated therapeutic encounter?

The Real and the limits of AI: transference and the question of jouissance

Pulling together the previous arguments I would now like to propose the following. What is far more critical than attempting to ‘humanise’ AI models is understanding the nature of the transference between human subjects and such models. This is particularly salient for AI therapybots, where humans are demonstrably forming transferential relationships. My contention is that, ultimately, the critical question is how jouissance is ‘handled’ in the transference and, furthermore, how it is handled by the analysand, rather than by the analyst.

Just to recap: in Seminar XI, Lacan asserts that transference arises where there is a supposed subject of knowing. The (neurotic) analysand, in essence, ‘creates’ this subject, positioning it as the Other, embodying knowledge of the analysand’s jouissance. However, by Seminar XIX Lacan posits the analyst as the semblance of the object a, that is, surplus jouissance. If the object a is truly ‘incarnated’ in the analyst, then the analyst becomes the very site of this surplus jouissance—a site of trauma.[15] A transference to such an analyst, then, would not be a process of detraumatisation but rather a reproduction of trauma. And at this point, the work of analysis, in the transference, is essentially the work of the analysand to encounter and engage with the jouissance that appears in the analytic space. The role of the analyst, as semblance of the object a (surplus jouissance), is primarily to ‘maintain’ such a space, to keep it open.

This brings us back to the core question: can an AI therapybot occupy the position of the object a and keep the analytic space open? While it can readily occupy the position of the supposed subject of knowing, offering a seemingly endless well of information and therapeutic responses, its capacity to become the incarnated site of surplus jouissance is, I would argue, more problematic. If an AI therapybot could incarnate surplus jouissance for the analysand, it would indeed become a site of trauma. However, it is perhaps more likely that the AI model, by steadfastly retaining and reinforcing the position of the supposed subject of knowing, would effectively ‘block’ the appearance of surplus jouissance. In this scenario, the AI model, despite its sophisticated simulation, prevents the irruption of the Real. It offers a perfectly rational, predictable, and non-traumatic interaction, thereby short-circuiting the very mechanism through which the Real announces itself in analysis. In fact, we might even argue that a ‘reverse uncanny valley’ effect comes into play.

A clinic of the Real demands an encounter with that which resists symbolisation, that which remains stubbornly outside the realm of knowledge and control. While AI models can process vast amounts of information and mimic empathetic responses, their inherent algorithmic nature may prevent them from becoming the very site of trauma necessary for the ‘handling’ of jouissance in its fullest sense. The uncanny valley suggests that the Real will always announce itself at the threshold of perfect simulation, signalling the irreducible difference between the human and the machine. Ultimately, the question for the clinic of the Real is not how ‘human’ AI can become, but rather, how the presence of AI forces us to confront the enduring and often unsettling truth of our own, uniquely human, relationship with the Real.

Notes and references

  1. As is ‘reality’ as opposed to the Lacanian Real. ↑
  2. J. Baudrillard, Simulacra and Simulation (Michigan: The University of Michigan Press, 1994). ↑
  3. The term ‘as if personality’ was originally coined by Helene Deutsch in 1934.  See the following article for a good summary of the arguments: Sergio Benvenuto, ‘‘As If’ Personalities and the Courage to Love – Notes,’ e-flux, 2022, https://www.e-flux.com/notes/504537/as-if-personalities-and-the-courage-to-love. ↑
  4. https://therapeia.org.uk/ttr/2025/01/04/artificial-analysis/ ↑
  5. Yuval Noah Harari, “Al Will Take Over Human Systems From Within,” October 9, 2024, https://www.noemamag.com/al-will-take-over-human-systems-from-within. ↑
  6. Joachim Cauwe, Stijn Vanheule, and Mattias Desmet, “The Presence of The Analyst in Lacanian Treatment,” Journal of the American Psychoanalytic Association 65, no. 4 (August 1, 2017): 609–38.  The authors explain that the term ‘supposed subject of knowing’ is preferred to the more commonly used ‘subject supposed to know’ because both the subject and their knowledge are supposed by the analysand. ↑
  7. In cases of psychosis, things are more complicated because it is the analysand themselves who takes up the place of the subject who knows. For a detailed discussion see: Jean Allouch, “Psychotic Transference,” in Lacan on Madness: Madness, Yes You Can’t, ed. Patricia Gherovici and Steinkoler (Hove: Routledge, 2015), 112–26. ↑
  8. Jacques-Alain Miller, “A Contribution of the Schizophrenic to the Psychoanalytic Clinic,” (Re)-Turn: A Journal of Lacanian Studies 1 (Winter 2003). ↑
  9. Pierre-Gilles Guéguen, “Paradoxes of the Births of Transference,” Psychoanalytical Notebooks, no. 17 (March 2008): 41–45. ↑
  10. Jacques Lacan, ……Or Worse: The Seminar of Jacques Lacan Book XIX (Cambridge, United Kingdom: Polity Press, 2022), 151, italics in original. ↑
  11. “But we all know, because all of us invent something to fill up the hole in the Real. Where there is no sexual relationship this gives a ‘troumatisme’” (Jacques Lacan, The Seminar of Jacques Lacan : Book XXI : Les Non Dupres Errent: 1973-1974, trans. by Cormac Gallagher (Dublin: Unpublished translation by Cormac Gallagher, 2010), 151. ↑
  12. Masahiro Mori, “The Uncanny Valley: The Original Essay,” IEEE Spectrum, accessed June 17, 2025, https://spectrum.ieee.org/the-uncanny-valley. ↑
  13. Sigmund Freud, “The Uncanny,” in The Standard Edition of the Complete Psychological Works of Sigmund Freud, vol. XVII (London: Hogarth Press, 1919), 218–56. ↑
  14. Jacques Lacan, The Seminar of Jacques Lacan: Book X: Anxiety 1962-1963 (Cambridge: Polity Press, 2014). ↑
  15. This is something I explore in detail in my book on trauma: Leslie Chapman, Traumatic Neurosis Revisited: Drive-Jouissance and the Wounds of Life, The Palgrave Lacan Series (London: Palgrave Macmillan, 2025). ↑

This article is part of a broader research project on The Algorithmic Other. For a complete overview of how Lacanian theory addresses the challenges of AI and digital simulation, please visit the AI and Psychoanalysis Hub.

As automated systems and algorithmic metrics increasingly format contemporary life and mental health, psychoanalysis provides an unscripted, human space to address what resists calculation. For those individuals who are navigating personal impasses or digtial burnout I can offer online consultations.  Please click here for further information.