Cognitive Behavioural Therapy: the Next 10 Years?

Over the next decade, large language models (LLMs), neural networks, and big data are poised to reshape both the theoretical foundations of Cognitive Behavioural Therapy (CBT) and its practical application within services such as Improving Access to Psychological Therapies (IAPT) in the UK. And here's some (AI-assisted) thoughts about how:....

At the theoretical level, CBT has traditionally been grounded in information-processing models that emphasise the interrelated causality of cognition, emotion, and behaviour. To do so, they use fairly simplistic, binary logic to explain the structures involved. LLMs—systems that simulate linguistic reasoning and generate human-like dialogue—are likely to push CBT theory toward a richer, more computationally sophisticated 'fuzziness'. These models demonstrate how thought patterns, emotional responses, and linguistic framing can be modelled probabilistically. This could lead to new conceptions of “automatic thoughts” and “interpersonal schemas” not as static cognitive structures but as dynamic, probabilistic patterns of language and meaning that can be mapped, visualised, and modified computationally.

In terms of neural networks, their ability to model complex, non-linear relationships between variables can deepen psychological understanding of how cognitive, emotional, and physiological processes interact. For example, by using remotely recorded EEGs in the workplace to sequence functionally connected brain structures interacting as (in vivo) thought processes, it will become possible to measure intellectual and emotional labour in a way that has not been possible previously. So future CBT theory may become more biologically and computationally integrated—linking neural representations of belief updating, attention bias, and emotional regulation with behavioural outcomes. This could bridge the current gap between cognitive models and neuroscience, helping to refine therapeutic targets and tailor interventions to individual neurocognitive profiles.

From a practice perspective, particularly in IAPT, the convergence of LLMs and big data could transform access, assessment, and delivery. Natural language processing systems trained on vast datasets of therapy transcripts, patient feedback, and outcome measures could help clinicians identify patterns predictive of recovery or relapse. AI-assisted triage tools might personalise tele-care and tele-medicine pathways, while conversational agents—if used ethically and under supervision—could offer guided self-help or between-session support that mirrors CBT principles. Big data analytics would allow IAPT services to move beyond population-level metrics to truly personalised mental healthcare, dynamically adapting treatment protocols in real time based on individual progress and contextual data (e.g., mood tracking, wearable data, or linguistic markers).

However, these transformations will raise deep ethical and epistemological questions: Can algorithmic empathy ever substitute for human understanding? How might data-driven interpretations of cognition reshape our notion of mental distress? And what happens when the model’s predictions challenge a therapist’s judgment? The next ten years will likely see not only a technological revolution in CBT practice but also a philosophical reckoning about the nature of therapeutic knowledge itself—how it is generated, validated, and shared between humans and machines.