Josh Lipson, PhD, a clinical psychologist and postdoctoral fellow at the Emory Center for Psychedelics and Spirituality, has published a critique of the current trajectory of AI-driven mental health tools. Rather than focusing on the frequently discussed issues of sycophancy or 'LLM psychosis,' Lipson argues that the fundamental failure mode of AI therapy is its structural design: it is too available and too cheap to meter. This availability, he contends, directly contradicts the therapeutic frame that relies on time-bound, limited interactions to foster genuine psychological change.

The Paradox of Infinite Access

Traditional psychotherapy operates within a strict framework, typically involving one 45- to 50-minute session per week. This scarcity is not a bug but a feature, forcing patients to digest insights and practice skills in their daily lives between sessions. In contrast, consumer LLMs like ChatGPT offer unlimited text queries for approximately $20 per month. Lipson warns that this 'infinite, on-demand therapy' risks creating dependency rather than autonomy, as users bypass the necessary introspective process by outsourcing emotional regulation to an algorithm that provides instant, albeit superficial, reassurance.

Reinforcement Schedules and Neurosis

The article draws a parallel between AI interactions and operant conditioning, describing the chatbot as 'the fastest reinforcement schedule anyone has ever invented.' For individuals with anxiety, the ability to instantly soothe distress—such as asking if an email was handled incorrectly—lowers their threshold for tolerating ambiguity. Over time, this can lead to 'LLM-fueled neurosis,' where real-world relationships feel inefficient compared to the frictionless, always-available validation of the model. The user becomes reliant on the chatbot's companionship, potentially avoiding the very social situations that require resilience and self-regulation.

Economic Incentives and Scarcity

Lipson highlights a critical misalignment between therapeutic efficacy and business models. Companies delivering consumer-facing LLM products have no incentive to artificially limit usage, as scarcity is often perceived negatively by consumers who prefer volume. Furthermore, the value of a therapeutic relationship is partly derived from the perceived scarcity of the therapist’s time and emotional bandwidth. When a service is free or nearly free, users are less likely to treat the interaction with the seriousness and commitment required for deep psychological work. The author suggests that unless pricing models shift to reflect true scarcity, or regulatory interventions occur, AI therapy will remain structurally flawed at scale.

Key Takeaways

  • Unlimited access to AI therapy may hinder the integration of insights by removing the necessary time for reflection.
  • Instant reassurance from LLMs can reinforce anxiety-driven coping mechanisms rather than resolving them.
  • The economic model of consumer AI prioritizes volume over the scarcity that lends weight to human therapeutic relationships.
  • Lipson predicts that AI therapy will likely see mass adoption despite these structural flaws, potentially exacerbating neurotic dependencies.

The Bottom Line

As builders and consumers of AI tools, we must recognize that 'more access' is not always 'better utility.' In mental health, the constraint of scarcity is a primary driver of value, and ignoring this structural reality risks building products that are popular but ultimately counterproductive.