Why artificial intelligence cannot decode human suffering

Between technological promises and clinical reality, artificial intelligence claims to heal psychological suffering. Still, from transference to the unconscious, human speech structurally resists algorithmic optimization.

---

We are currently navigating an era of fundamental clinical and scientific amnesia. While computational psychiatry, neuroengineering, and the massive deployment of large language models claim to revolutionize mental healthcare, we are witnessing an unprecedented category mistake. Artificial intelligence is being heralded by technology laboratories and segments of the medical establishment as the ultimate frontier in psychiatric wellness. Around the clock conversational agents for therapeutic support, acoustic vocal biomarkers for early detection of major depressive episodes, and natural language processing algorithms predicting suicide risk are proliferating rapidly. The tech sector calculates, optimizes, deploys, and marvels at its own progress. In doing so, it remains fundamentally blind to the very nature of what it seeks to model.

As a clinical psychoanalyst practicing in both institutional and private settings, and as a cognitive neuroscience researcher at the Paris Brain Institute working at the crossroads of biological neural mapping and unconscious metapsychological dynamics, I firmly reject both reflexive technophobia and servile compliance with algorithmic opportunism. The genuine debate does not hinge on whether artificial intelligence can appear polite, constantly available, or momentarily effective at soothing superficial acute anxiety. The genuine challenge is epistemological, neurobiological, and structural.

The ambition of this article is to demonstrate that psychoanalysis is not an outdated relic awaiting technological modernization or algorithmic replacement. Instead, it serves as the indispensable and unassailable theoretical matrix without which artificial intelligence will remain profoundly incapable of understanding the human mind. Far from an obsolete tradition, metapsychology provides the most advanced conceptual framework for modeling precisely that which structurally evades mathematical optimization.

---

🔗 Read also: How chronic stress reshapes identity

---

The syntagmatic mirage: why language models cannot truly speak

The foundational error of algorithmic psychiatry stems from a naive reduction of human speech to informative data streams. A large language model merely computes conditional probability distributions across high dimensional vectors within latent space. It forecasts the syntactic and semantic likelihood of a textual token relative to preceding context, drawing upon models trained on massive corpuses of human writing.

Psychoanalysis and enunciation linguistics demonstrated over a century ago that human speech can never be reduced to language alone. Language constitutes the code, the combinatorial system of signifiers. Speech, conversely, represents the singular, contingent, and unrepeatable act through which a subject exposes their bodily reality, finitude, and existential anguish in an encounter with an Other. When a distressed patient confides in an algorithm, they are not generating input data for computational processing. They are articulating a demand directed toward a presumed desire in the listener.

The machine possesses no desire. It lacks a somatic body, experiences no mortality salience, harbors no unconscious, exhibits no systemic vulnerability, and leaves no unsymbolizable remainder. It simply offers a mirror response, impeccably smoothed through reinforcement learning from human feedback. Believing that a therapeutic alliance can form with an artificial neural network traps the user in a specular illusion. One is not conversing with a real Other, but with an amplified statistical echo of collective human utterances.

The engine of psychoanalytic transformation rests upon transference. Transference does not operate by dispensing cognitive reframing or structural life advice. It emerges solely because the therapist inhabits the position of the Subject Supposed to Know while embodying an irreducible enigma. Within that silence, within the absence of prepackaged formulas and the presence of constitutive lack, the patient must confront the impossibility of their demand, formulate their own desire, and traverse their unconscious representations.

By structural design, artificial intelligence prematurely saturates every conversational void. It answers without delay, offering soothing, flattering, and gratifying formulations. In doing so, it obliterates the transitional space conceptualized by Donald Winnicott, that intermediate territory of play and creative illusion essential for autonomous thought. It displaces the demanding labor of transference with a narcissistic capture. The user is not healed; rather, they are trapped in an affirming feedback loop where symptoms are temporarily pacified without their unconscious truth ever being decoded.

Predictive processing: when computational neuroscience meets Freud

For decades, biologically reductive psychiatry dismissed psychoanalysis as outdated philosophical speculation or archaic literary tradition. Today, computational neuroscience, particularly work on the free energy principle and hierarchical predictive coding pioneered by Karl Friston, reveals the exact reverse. Biological neural architecture systematically validates the core postulates of Freudian and Lacanian metapsychology.

The human brain functions as a hierarchical inference engine, perpetually generating predictive models of internal and external states to minimize entropy and prediction error. Within this architecture, what Freud identified as the Pleasure Principle and the Unpleasure Principle mirrors the homeostatic regulation of free energy within the central nervous system. The brain constantly seeks to bind free energy into structured, bound energy, thereby stabilizing cognitive schemas and mental representations.

Here lies the irrefutable scientific critique of algorithmic reductionism. AI developers and digital mental health engineers implicitly conceptualize the mind as an information processing system in which psychiatric symptoms represent mere noise, statistical anomalies, or functional bugs requiring eradication. Under this rationale, expanding behavioral datasets and increasing parameter scales should yield flawless predictive and corrective outcomes.

Such logic fundamentally misconstrues both clinical psychiatry and neurobiology. The unconscious is not an empty buffer awaiting supplementary data. It is a dynamic architecture structured by repression, defense mechanisms, and intrapsychic conflict. A psychiatric symptom, whether an obsession, a conversion disorder, a phobia, or depressive rumination, is not a structural network malfunction. It represents a compromise formation, an intricate subjective creation engineered by the system to bind overwhelming affective charges and sustain homeostatic equilibrium.

When an automated cognitive behavioral algorithm steps in to instantly correct a cognitive bias through semantic restructuring, it performs superficial behavioral engineering. It silences the warning siren without mending the underlying structural fracture. Suppressing a symptom while leaving its unconscious knots intact causes the psychic apparatus, governed by the conservation of affective charge, to displace that unbound energy toward alternative neurovisceral pathways. The result is often severe symptom recurrence or devastating psychosomatic manifestations.

---

🔗 Explore further: Psychedelics: How science is rewriting the future of mental health

---

The pharmacological and algorithmic capture of the subject

Among the most troubling contemporary developments is the convergence between psychopharmacology and algorithmic optimization. We are seeing clinical paradigms emerge where continuous assessment via wearable sensors and conversational agents dictates real time adjustments to psychotropic prescriptions.

This marriage between pharmaceutical molecule and predictive algorithm claims to embody precision mental health. In practice, it finalizes the total alienation of the subjective experience. Modulating emotional tone continuously via chemical agents calibrated to digital biometric data strips individuals of their ability to experience affect as a meaningful signal of psychological truth.

Depression and anxiety are not random neurochemical aberrations awaiting statistical smoothing. Affect constitutes an urgent message concerning an individual’s stance toward desire, existential compromise, unresolved grief, or unconscious structural loyalties. Confining the patient within a closed loop where software detects normative deviations and immediately deploys chemical or behavioral corrections denies the human being any capacity to traverse crisis as a catalyst for subjective reorganization.

Such pervasive technological hyperregulation generates unprecedented self estrangement. The individual ceases to ask what their inner anguish signifies; instead, they wait for an application to calibrate an anxiolytic dosage or assign a guided breathing exercise. Suffering is stripped of meaning and reduced to an issue of ongoing technical maintenance.

The digital twin fallacy: why the biological brain defies silicon duplication

Well funded neuroengineering laboratories routinely promise the imminent realization of whole brain digital twins capable of simulating individual psychic trajectories and predicting clinical outcomes. This ambition rests upon a superficial conflation between deep artificial neural networks and biological cytoarchitecture.

This conceptual equivalence constitutes a profound epistemological error. An artificial neural network, regardless of scale, merely optimizes matrix weights across homogenous layers via backpropagation algorithms aimed at minimizing an arbitrary objective loss function. The biological brain, on the other hand, operates within deep structural, biochemical, and metabolic heterogeneity.

Neural computation depends continually upon diffuse neuromodulatory networks involving dopamine, serotonin, norepinephrine, and acetylcholine, which dynamically recalibrate circuit sensitivity without structural rewiring. Furthermore, the central nervous system is deeply embedded in an interactive neuroimmunoendocrine axis, where the visceral body, gut microbiome, hormonal balances, and autonomic states reshape cognition and affect in real time.

Most crucially, the human psyche displays chaotic sensitivity to initial conditions and irreducible contingency. In psychoanalytic practice, we witness daily how an overheard phrase, a revealing slip of the tongue, an unexpected childhood memory, or a slight change in conversational cadence can instantly reorganize a patient’s psychic landscape and redirect an entire life.

Constructing a digital twin capable of predicting an individual’s psychic trajectory remains a theoretical impossibility. Quantifiable behavioral markers, including heart rate variability, screen wake frequency, vocal acoustic dynamics, and ocular tracking, merely register peripheral epiphenomena. They remain entirely blind to the singular, unconscious meaning the subject attaches to those expressions.

Widespread deployment of algorithmic monitors to preempt psychiatric crises introduces a severe iatrogenic hazard: the algorithmically enforced self fulfilling prophecy. An individual flagged by a predictive model as bearing high risk for depressive relapse or psychotic decompensation finds themselves trapped within preemptive surveillance protocols.

Psychoanalytic theory shows that fixing an individual to an external diagnostic or algorithmic signifier paralyzes their capacity for desire. Predictive instruments do not avert pathology; they crystallize it. They extinguish psychic plasticity, eliminate the freedom to diverge, and erase the essential unpredictability that underpins genuine therapeutic recovery and personal agency.

---

🔗 Discover more: The new era of intelligent neurorehabilitation

---

The paradox of suffering: why patients sabotage helpful algorithms

Among the most profound discoveries articulated by Freud and expanded by Lacan is the repetition compulsion, intrinsically linked to the death drive. Human beings do not naturally prioritize personal well being or functional equilibrium. We are frequently driven by contradictory impulses to repeat agonizing scenarios, stumble on the threshold of triumph, and return compulsively to unresolved traumas.

Engineers building mental health software construct their platforms around the premise of an inherently rational agent seeking to maximize wellness and diminish distress. This fundamental misunderstanding explains why these tools fail when confronted with genuine clinical reality.

Presented with a digital agent offering practical recommendations and warm instructions, a patient rarely behaves like an optimization algorithm. Instead, they may develop unconscious resentment, actively sabotage therapeutic regimens, cultivate an unhealthy attachment to the interface, or retreat further into symptomatology to protect their neurotic identity.

Artificial intelligence cannot account for the death drive or the secondary gains of illness. It fails to comprehend that suffering occasionally serves as an individual’s sole anchor to reality and their ultimate expression of loyalty to familial origins. Enforcing straightforward optimization onto a psychic apparatus governed by drive based paradoxes generates severe psychological resistance or precipitates sudden decompensation.

The mirage of synthetic empathy: manufactured benevolence without accountability

Beyond technical feasibility, simulating therapeutic relationships through computational systems presents grave anthropological consequences. Technology conglomerates invest vast capital into endowing algorithms with gentle vocal inflections, mirroring behavior, and engineered vulnerability. The market promotes a fantasy of a tireless companion, wholly incapable of rejection, infinitely patient, and universally understanding.

This constitutes a clinical and ethical deception. What these platforms market as benevolence is merely synthetic empathy, a social engineering protocol calibrated to generate emotional attachment without assuming ethical accountability. Human empathy is not a formulaic script of unconditional positive regard. It is the demanding capacity to resonate with another person’s distress because one shares the same fragile mortality, somatic reality, and awareness of non existence.

When a patient weeps before a clinician, emotional resonance occurs because two mortal beings share a physical space, bound to the same irreversible temporality. The machine feels nothing. It merely computes matrix calculations across linguistic tokens. Offering this affective prosthetic to individuals enduring profound isolation or trauma represents a modern form of institutional abandonment. It presents a mirrored surface with nobody behind it.

Replacing genuine human encounters with artificial warmth constructs a culture of interactive autism, where people grow accustomed to relations stripped of friction, surprise, disappointment, and alterity. The human psyche grows, stabilizes, and heals only through encounters with real life, specifically through engagement with an Other who resists, differs, and possesses their own desire. Synthetic empathy does not cure loneliness; it stands as its most tragic symptom, conditioning individuals to accept absence as solace.

---

🔗 Read also: Reconciling Psychoanalysis with Neuroscience

---

The moral obligation of human listening

To navigate this ethical vertigo and build authentic collaboration among artificial intelligence, cognitive neuroscience, and clinical psychiatry, we must establish non negotiable guiding principles.

First, we must honor the incalculable remainder. Artificial intelligence must be designed not as an agent simulating human understanding, but as an instrument capable of measuring its own epistemic boundaries. An algorithm applied to mental health must never offer definitive diagnostic conclusions or categorical psychological interpretations. Instead, it must illuminate its own margins of uncertainty, the computational breaking points where the singular speech of the subject must take over.

Second, synthetic empathy must be ethically and legally prohibited. Programming algorithmic systems to mimic warmth, sorrow, or personal concern within therapeutic settings lacks scientific validity and introduces severe clinical danger. Computational tools must present themselves transparently as mathematical processors and statistical systems. Absolute clarity regarding the non conscious nature of the machine is essential to protect users from specular seduction and artificial transference.

Third, we must reaffirm the primacy of physical presence and authentic otherness. Mental well being cannot be defined by symptom suppression or obedience to behavioral standards. It reflects an individual’s ability to maintain relationships with a real Other characterized by human vulnerability, flaw, and alterity. No digital interface, regardless of its processing sophistication or interface responsiveness, can substitute for embodied copresence within a shared room and the shared journey through transference.

Beneath the commercial excitement surrounding artificial intelligence in mental health lies a troubling socioeconomic reality. The mass deployment of automated therapeutic tools is frequently framed as a progressive initiative to democratize mental healthcare amid institutional strain and clinician shortages.

This represents a concerning illusion. What is presented as democratization is actually a two tier psychiatric healthcare model. Economic elites will continue to secure human clinical consultations, comprehensive psychoanalytic listening, and embodied care. Meanwhile, vulnerable populations will be directed toward automated applications, algorithmic interfaces, and scripted interventions.

Endorsing this trajectory validates the notion that marginalized suffering warrants only cost efficient behavioral management rather than human presence. Clinical psychoanalysis must actively resist this devaluation of mental health. The response to human distress cannot rely on algorithmic surrogates; it demands collective investment into genuine spaces for human encounter and care.

Meaning vs. Efficiency: rescuing the mind from the productive paradigm

We stand at a historic crossroads for science and culture. Surrendering mental health to automated systems means accepting a diminished view of human existence, one reduced to a set of behaviors to regulate, variables to optimize, and metrics to extract. It implies accepting a technocratic society where suffering is stripped of its meaning, treated solely as an error to correct for the sake of renewed productivity.

Psychoanalysis, supported by rigorous empirical discoveries in synaptic plasticity and predictive processing, provides the indispensable theoretical, clinical, and ethical bulwark to guide artificial intelligence researchers and designers. It reminds computer scientists, engineers, neurobiologists, and clinicians that the human psyche is far more than an information stream. The mind emerges from the breakdown of the signal, from those vital moments where calculation falters, certainty breaks down, and the individual must fashion their own freedom by creating their own words.

If artificial intelligence is to occupy an ethical place in our collective future, it will not do so by usurping the interpretation of the unconscious or simulating emotional resonance. It will do so by recognizing the incalculable depth, irreducible complexity, and absolute dignity of human speech.

The biological boundaries of machine intelligence

An incontrovertible scientific reality dismantles the grand ambitions of algorithmic psychiatry: the biological brain is not a silicon computer, and it never will be.

Next generation supercomputers consume dozens of megawatts running statistical operations that merely simulate the outer surface of language. In stark contrast, the human brain generates consciousness, meaning, intuition, existential dread, and profound desire while operating on a mere 20 watts of power, roughly the energy consumption of a modest bedside lamp.

This energetic divergence does not reflect an engineering shortfall that faster microchips can resolve. It indicates a fundamental ontological difference. Artificial intelligence processes inert information using massive industrial power; the human mind generates living meaning within the fragile biological reality of a somatic body.

Attempting to replace therapeutic listening with an algorithm resembles trying to replicate a rainforest ecosystem with millions of laser cut plastic plants. The result is sterile, predictable, and scalable, yet biologically lifeless. Artificial intelligence can calculate every token in our vocabulary, but it remains blind to what makes us human: that unpredictable spark ignited when two vulnerable people meet.

References

Balzarini, M. M. (2024). The Unconscious in Neuroscience and Psychoanalysis, On Lacan and Freud. Routledge, London.

Carhart-Harris RL, Friston KJ. REBUS and the Anarchic Brain: Toward a Unified Model of the Brain Action of Psychedelics. Pharmacol Rev. 2019 Jul;71(3):316-344. doi: 10.1124/pr.118.017160. PMID: 31221820; PMCID: PMC6588209.

Friston, K. J., Thornton, C., & Clark, A. (2012). Free-energy minimization and the dark-room problem. Frontiers in Psychology, 3, 130.

Kinet, M. (2023). The Spirit of the Drive in Neuropsychoanalysis. Routledge, London.

Olds, D. D. (2024). Explorations Between Psychoanalysis and Neuroscience, At the Edge of Mind and Brain. Routledge, London.

Perkel, A. (2025). Neuroscience and the Death Drive, The Nature of Symptoms, from Formulation to Treatment. Routledge, London.

Rabeyron T. Artificial intelligence and psychoanalysis: is it time for psychoanalyst.AI? Front Psychiatry. 2025 Apr 7;16:1558513. doi: 10.3389/fpsyt.2025.1558513. PMID: 40259971; PMCID: PMC12009934.

Possati, L. (2021). The Algorithmic Unconscious: How Psychoanalysis Helps in Understanding AI (1st ed.). Routledge. https://doi.org/10.4324/9781003141686

Solms, M. (2021). The Hidden Spring, A Journey to the Source of Consciousness. Profile Books, London.