Beyond the Prompt: Why Fear Kills Curiosity and What That Means for AI Adoption

Opening Summary

Most organizations treat AI adoption as a skills problem. Wrong. Cognitive neuroscientist Chantel Prat argues it is a safety problem. Her research reveals that when people feel threatened by AI, worried about their job, their competence, or being judged, their brains shut down the very learning mechanisms needed to adapt. The hidden consequence is that companies framing AI as an existential threat systematically discourage curiosity. Leaders, AI educators, and anyone trying to build AI-native teams need to understand that the biggest barrier is not ignorance. It is fear. And fear does not just block learning. It makes people avoid the new technology altogether.

Key Insights & Analysis

The Curiosity Equation: Safety Before Learning

Prat introduced a model called PACE (Prediction, Appraisal, Curiosity, Exploration) that maps the neural pathway to learning. The first precondition is that you need to know you do not know. But that is not enough. The second precondition, appraisal of safety, is where most AI initiatives fail.

"Before it makes you feel curious, it says am I safe here? And I think that in the space of AI, we're talking about psychological safety, job safety, evaluative safety, like as someone else monitoring my performance."

Consider what this means. When a leader says "AI will transform our industry, adapt or die," they are triggering the appraisal system. The brain says "not safe." Curiosity shuts down. The team does not become more eager to explore AI; they become more defensive. The very message meant to spur action actually prevents the learning required to act.

The downstream effect compounds over time. People who feel threatened avoid using AI, which means they gain no experience, which makes them more anxious, which deepens the avoidance. The system reinforces itself. Breaking that loop requires addressing the safety appraisal before trying to teach skills.

Theory of Mind: The Unlocked Competitive Advantage

People who possess a strong theory of mind for AI, the ability to model why the AI behaves the way it does, consistently outperform those who do not. And here is the notable part: theory of mind is entirely learned.

"theory of mind seems to be entirely learned there was like zero genetic factor"

Prat distinguishes theory of mind from empathy. Empathy is fast, automatic, and works when someone's brain is like yours. Theory of mind is slow, expensive, and requires building a model of something fundamentally different from you. With AI, the default assumption is that because it talks like a human, it thinks like one. That is System 1 thinking. And it leads to wrong predictions. The people who succeed are those who do the hard work of understanding how a given AI was trained, what it is optimized for, and where its blind spots are.

The implication is that theory of mind is a trainable skill that creates compounding advantage. Each interaction with AI becomes more effective because you are building a better internal model of the system. And Prat suggests AI itself can be the training ground: a non-judgmental environment where you can practice modeling another intelligence and get feedback.

"The AI might be a good way to practice this theory of mind building, because with people... usually you just sound like a jerk and you get it wrong... But your AI, it's like you can start with it."

The Expertise Trap: Why Success Stifles Exploration

Prat's eggs Benedict analogy is deceptively simple. The more data you accumulate about what works, the higher the opportunity cost of trying something new. At 51, you have eaten a lot of eggs Benedict. You know exactly how good the known option is. A new restaurant is not just a gamble; it is a bet against all that experience.

This is the explore-exploit dilemma at the individual level, and it explains why AI adoption is hardest among the most experienced people in an organization. They are not being stubborn. They are being rational. Their existing strategies have been validated over decades. The new technology offers an undefined reward with a high probability of being worse than what they already do.

But systems thinking shows that the short-term rationality of exploiting existing expertise creates long-term vulnerability. The person who refuses to learn AI today might be perfectly productive for the next 18 months. But when the system shifts, when competitors start operating at different efficiency curves, when the next generation of AI tools becomes the default, that expertise becomes a liability. The decision that makes sense now creates a trap later.

The Curiosity Catalyst: What Organizations Get Wrong

Prat's PACE model makes a prediction about organizational behavior: if you want people to explore AI, you need to create conditions where they feel safe to not know. That means separating performance evaluation from learning. It means celebrating failed experiments alongside successful ones. It means leaders admitting their own ignorance.

The hosts noted a powerful dynamic: people deep in Phase 3 or Phase 4 of AI proficiency (building autonomous agents, connecting databases) can inadvertently scare people who are still in Phase 1 (using AI as Google). The enthusiasm of the advanced user triggers threat in the novice. The solution is not to dumb down the message; it is to explicitly label the fear and normalize the learning curve.

"How do we spark people's curiosity rather than trigger their fear? I think in fact maybe just like sitting with that question and asking our audience even to sit with it... is my enthusiasm scaring people or getting them curious?"

Key Action Items

• Start every AI training by addressing safety, not skills. Over the next quarter, introduce AI initiatives with explicit acknowledgment of job security concerns. Frame AI as augmentation, not replacement. This pre-condition is the bottleneck. Skip it and the rest will not stick.

• Build theory of mind as a team practice. Within the next month, have team members ask their AI tool questions about itself: "How were you trained? What are you optimized for? Where did your training data come from?" Debrief as a group to build shared mental models. This pays off in 6 to 12 months as everyone gets better at predicting AI behavior.

• Create a psychological safety metric for learning environments. Track whether team members feel safe admitting they do not know something about AI. If scores are low, address the appraisal problem before adding more tools. This is an immediate action that unlocks long-term learning.

• Design exploration time with undefined rewards. Model what Prat calls "input for input's sake." Block 90 minutes weekly for unstructured experimentation with AI, no deliverables, no KPIs. The discomfort of undefined ROI is exactly why it works: most teams will not tolerate it, creating an advantage for those who do. This becomes durable in 12 to 18 months.

• Reverse-mentor across the expertise gap. Pair AI-advanced employees with senior experts who are stuck in the exploration trap. The senior's job is not to learn the tool; it is to articulate why they are reluctant. The junior's job is to demonstrate without threatening. This builds theory of mind on both sides. Start this in the next month.

• Teach the PACE model explicitly. Run a 30-minute session explaining prediction, appraisal, curiosity, and exploration. Ask each person to identify where they get stuck. Most will say appraisal: "I don't feel safe to experiment." Address that directly. This is a one-time investment that creates a shared vocabulary for learning.

• Practice the English to English translation. When communicating about AI across skill levels, consciously adapt your message to the listener's context. Ask yourself: "If I were in Phase 1, would this feel exciting or threatening?" This pays off immediately in improved collaboration and reduces the hidden cost of fear-based resistance.