AI is no longer a lab curiosity. It now sits inside therapy apps, search results, classrooms and workplaces, quietly shaping how we think and how we cope. In Episode 20 of the ReeThink Podcast, VR and clinical psychology pioneer Dr. Albert “Skip” Rizzo joins host Rudi Adigbli to explore how AI in mental health functions as a barometer of humanity’s health, revealing more about our systems and incentives than about the models themselves.

nstead of asking whether AI will “replace therapists,” this conversation asks a sharper question: What does our use of AI say about our ethics, our executive function and our capacity to care for each other?

Dr. Skip Rizzo is one of the world’s leading experts in virtual reality therapy and clinical applications of emerging tech. He directs medical virtual reality work at USC’s Institute for Creative Technologies and has spent decades building VR systems for PTSD, traumatic brain injury and rehabilitation.

In recent years, Skip has extended this work into AI-driven conversational agents for mental health, focusing on patient-facing tools that support, rather than replace, clinicians. This dual lens, deep clinical experience and hands-on AI development, makes him an ideal guide through the current hype cycle around AI in mental health.​

A key theme in Episode 20 is the difference between AI as decision support and AI as decision maker. That distinction sounds subtle, yet it defines whether AI in mental health becomes a force multiplier for clinicians, or a dangerous shortcut.

Used for decision support, cognitive AI tools can:

• Help clinicians structure assessments and surface overlooked risk factors.

• Summarize patterns across sessions that human memory might miss.

• Act as a non‑fatiguing assistant, freeing executive function for complex judgment.

Used as a decision maker, similar systems can:

• Encourage institutions to cut human staff on crisis hotlines, assuming AI can “handle it”.

• Give plausible but unsafe advice to people in acute distress, with no accountability.​

Skip shares concrete concerns about recent cuts to human-operated suicide helplines at the VA, and the risk that AI in mental health becomes an excuse to reduce human contact rather than expand access. He draws a hard ethical line: patient‑facing AI should extend human care and guide people towards real humans, not quietly replace them.​​

This is where AI and executive function intersect. When institutions outsource judgment to models, they are also outsourcing responsibility. The “barometer reading” drops as soon as cost optimization outweighs duty of care.​​

Outside formal healthcare, a quieter shift is unfolding. Knowledge workers, founders and employees are increasingly turning to general-purpose AI tools for psychological support. They ask LLMs about anxiety, burnout, relationship decisions and suicidal thoughts, often late at night and entirely outside any clinical context.

Episode 20 highlights three problems with this off‑label use of AI in mental health:

• Privacy and data risk – Sensitive emotional disclosures are stored and used to improve models, with unclear boundaries.

• False sense of safety – Outputs feel empathetic and coherent, but are not grounded in personal history or clinical training.

• No escalation path – Unlike a therapist, an LLM has no legal duty, supervision structure or clear protocol when risk appears.

At the same time, the episode acknowledges a brutal reality: the World Health Organization and recent workplace reports agree that a large share of people with mental health conditions will never see a therapist. This access gap is exactly where AI in mental health is being pulled in, sometimes thoughtfully, sometimes recklessly.

Here, AI again acts as a barometer of humanity’s health. It shows whether we treat mental health as a core public good that deserves human investment, or as a line item to be automated away.​​

Throughout the discussion, Skip emphasizes that good intentions are not enough. AI in mental health requires concrete guardrails and ongoing monitoring.

Emerging guidelines and best practices include:

• Safety before scale – Proving that a system does not cause harm in vulnerable states before deploying widely.

• Transparency and auditability – Making it clear when users are interacting with AI and logging decisions for later review.

• Continuous monitoring – Tracking model behavior over time as data and usage patterns change.

• Human escalation pathways – Designing AI workflows that actively push high‑risk users towards live clinicians, not away from them.​​

Skip points to recent research where one AI mental health chatbot showed promising results in a controlled trial, while another startup chose to shut down their product after concluding that for people in deep crisis, the tool became “dangerous, not just inadequate”. The difference lies in governance, not just in model quality.​​

For founders and investors, the takeaway is clear: AI in mental health is no longer a purely technical or ethical debate. It is an operational governance challenge. Systems need clear ownership, escalation authority and defined lines of accountability

Not all use‑cases in Episode 20 are about crisis or therapy. Some of the most exciting possibilities for AI in mental health lie in education and prevention.

Rudi introduces Manu Kapur’s “productive failure” framework, a learning approach where students tackle challenging problems before being shown the formal solution. This method is highly effective yet hard to scale, because it demands extensive preparation.

Here, cognitive AI tools can:

• Generate tailored problem sets and variations in real time.

• Provide stepwise hints rather than complete answers.

• Adapt difficulty to keep learners in a flow state rather than boredom or panic.

When combined with nervous system regulation performance, teaching students how to notice stress, reset their physiology and return to focus, this becomes a powerful mental gym for the next generation. The goal is not just better test scores, but resilient, self‑aware brains that can partner with AI instead of being overwhelmed by it.

In this context, AI in mental health is less about diagnosing disorders and more about training attention, emotion regulation and reflective thinking at scale.

One of the clearest red lines in Episode 20 concerns children. Both Rudi and Skip agree that it is unreasonable to place the burden of digital self‑regulation on developing brains.

Children and teenagers now interact with:

• Recommendation systems tuned for engagement rather than attention sovereignty.

• AI companions and chatbots that feel personal but are not accountable.

• Feeds filled with AI‑generated content that blurs reality and simulation.

Because executive function and prefrontal control mature relatively late, young users are especially vulnerable to persuasive design and AI in mental health that has not been designed with them in mind.

This leads to a simple, practical position:

• Stronger regulation for minors is non‑negotiable.

• Default settings, data collection and content exposure must be radically more protective.

• “Human‑in‑the‑loop” cannot be a slogan; it must translate into actual safeguards for young nervous systems.

Again, the barometer metaphor applies. How we let AI interact with children tells us more about our collective priorities than about the models themselves.

By the end of Episode 20, a pattern emerges. AI in mental health does not magically improve or destroy humanity. It amplifies what is already there.

When incentives prioritize quarterly earnings and cost reduction, AI becomes a sophisticated way to cut corners. Crisis lines shrink, human contact becomes scarce, and vulnerable people are quietly redirected to systems that were never built to carry that weight.

When incentives prioritize long‑term brain optimization for business, human dignity and nervous system health, the same technology becomes:

• A diagnostic aid for clinicians.

• A scalable teaching assistant for productive failure and deep learning.

• A digital companion that knows its limits and hands off to humans when necessary.

In that sense, AI in mental health is truly a barometer of humanity’s health. It reflects our ethics, our operational discipline and our willingness to keep humans at the center.

The future of AI in mental health will not be decided by algorithms alone, but by whether we are ready to build a mental gym culture that trains both people and systems to handle this new cognitive pressure.

ReeThink Incetives. ReeWire AI.