Iris Care Group

Andrew Hider

Cognitive Spaces and the World States

Imagine walking into a room. Your brain instantly processes the room’s dimensions, lighting, temperature, the people present, and their behavior. You also have an emotional response—pleasant, unpleasant, or neutral. Automatically, your brain creates a representation of the room, which guides your actions. As you move through the space, this representation is continuously updated. Each person processes the room uniquely, resulting in different behaviors and reactions. These representations shape our understanding of what is true, who is safe, and how we relate to others. Consciousness illuminates the necessary aspects of our environment for decision-making and action. Our brains compute this information in ways that give us the experience of moving through time with a consistent identity, even as our minds evolve.

We create “world states” and navigate them using “cognitive spaces.” A “world state” can be thought of as a map of reality, while a “cognitive space” is the subjective experience of navigating that map.

The Challenge of Mental Health AI 

How does this concept relate to AI in mental healthcare? Current AI applications focus on automating assessment, diagnosis, and treatment—such as analyzing clinical records, suggesting treatment algorithms, and creating therapy chatbots. While these tools are valuable, they mainly rely on signal detection, extracting diagnostic indicators from health records. In mental healthcare, this approach risks oversimplifying the complex cognitive processes of experienced clinicians and can overlook the patient’s perspective. The challenge lies in developing AI systems that operate effectively in the nuanced landscape of mental health practice, while also providing meaningful support to clinical care.

AI can significantly reduce the administrative burden on clinicians, allowing them to spend more time with patients. Currently, mental health professionals dedicate much of their time to documentation, analysis, and treatment planning—tasks that AI could streamline without compromising clinical rigor. This shift could increase the time for face-to-face therapeutic sessions, where the most valuable clinical work occurs.

The Clinical Reality of Mental Healthcare  

Mental health practice, especially at secondary and tertiary levels, differs significantly from physical medicine, which often relies on evidence-based treatment heuristics and objective clinical investigations. Consider the diagnosis of a broken bone: a radiologist detects signals in imaging data, and once shown the x-ray, it’s clear that the bone is broken. However, conditions like depression, panic attacks, or intrusive thoughts are not visible on a scan. This clinical reality presents unique challenges for AI implementation in mental healthcare, demanding innovative approaches to development and validation.

In good mental health practice, clinicians typically develop “case formulations” to understand psychological states. They construct a “world state” based on physical observations (appearance, speech, behavior) and then access a cognitive space, using their training, clinical models, and evidence-based knowledge. This clinical understanding is shared and refined with the patient, who may accept, reject, or collaborate on improving the formulation based on their unique perspective. The formulation remains a hypothesis until treatment proves effective. For complex or hard-to-treat problems, this process can be lengthy and unpredictable.

“Mental health care is fundamentally about understanding and helping unobservable minds, and AI models may not always be able to mirror the complex clinical problem-solving required in healing”

Beyond Signal Detection: The Need for Shared Understanding  

Mental health conditions account for a significant portion of the global health burden, yet current treatments often fall short. Too many people suffer for too long. The future of AI in mental healthcare promises sophisticated agents capable of supporting expert clinicians with evidence-based reasoning, particularly for complex or rare cases where clinical judgment can mean the difference between effective and ineffective treatment. However, these models need to be developed through a mental health-specific evaluative lens, requiring new approaches to model training and evaluation:

● Developing "Veridicality" in Mental Health AI Models

Mental health is dominated by perceptions of subjective meaning and experience, and AI models in this space must be evaluated using novel metrics. Since shared understanding is critical—and “truth” is as much a function of the patient’s cognitive space as visible clinical measures—AI models must produce outputs that resonate with patient experiences while remaining clinically helpful. For example, when analyzing electronic health records or clinical session transcripts, AI models should not only align with the patient’s perspective but also provide actionable insights for clinicians. This necessitates the development of robust evaluation frameworks, possibly advancing the concept of “veridicality”—the idea that it’s more important for an AI system in mental health to align with the subjective state of the patient than to detect an externally verifiable clinical variable.

● Ensuring "Clinical Reach" and Patient Safety

Equally important is ensuring that AI systems have sufficient "clinical reach" in their reasoning processes. Mental healthcare, dominated by subjective experience, requires safety criteria prioritizing human input. Mental health care is fundamentally about understanding and helping unobservable minds, and AI models may not always be able to mirror the complex clinical problem-solving required in healing. The risk of AI models being mistaken for genuine human understanding, particularly in clinical decision-making, is significant. Failures in this domain could have severe consequences. This risk demands the development of consensus measures and thresholds for acceptable “clinical reach” in AI systems used in mental health settings.

The Future – Democratising Access to Expert Mental Health Care         

The potential for AI in mental healthcare extends beyond improving existing services to democratizing access to expert clinical care. By capturing and safely deploying the complex reasoning patterns of experienced clinicians, AI could help bridge the global mental health treatment gap. If this is done correctly, while maintaining safety standards through new consensus metrics, we could create solutions that not only advance mental healthcare delivery but also make expert clinical reasoning more accessible to practitioners and patients worldwide.

However, success in this field requires a sophisticated understanding of the unique cognitive tasks involved in mental healthcare. Development must prioritize collaboration with clinical experts to ensure that AI solutions enhance, rather than replace, human judgment. The key to progress will be balancing innovation with safety considerations. Solutions that show measurable improvements in treatment outcomes while preserving the essential human elements of care will naturally find their place in healthcare systems worldwide.