The Crisis
Adoption Problem

There is a pattern I keep seeing, in my research, in organizations I study, and in my own professional environment. It is so consistent that I have started calling it a problem — not in the colloquial sense, but in the precise sense: a structural feature of how humans and institutions relate to AI that reliably produces bad outcomes.

The pattern is this: organizations and individuals adopt AI tools most aggressively exactly when careful evaluation is most critical — and least likely.

I call this the Crisis Adoption Problem.

What My Research Found

My dissertation examined what happens when human operators work with AI-assisted decision systems under time pressure. The finding I keep returning to — the one that surprised even me — was not that people trusted AI too much. It was when they trusted it most.

Under low time pressure, participants engaged meaningfully with AI outputs. They questioned anomalies. They cross-referenced. Their trust tracked their understanding.

Under high time pressure, that relationship collapsed. Trust in the AI system actually increased — even as comprehension decreased. Participants leaned harder on the AI precisely when they had the least capacity to evaluate whether leaning on it was warranted.

I called this the understanding-trust gap. But the more I thought about it, the more I realized it wasn't just a finding about individual operators. It was a finding about systems. And about organizations. And about the moment we are living in right now.

Scaling Up: From Operators to Organizations

What happens to an individual operator under time pressure happens to institutions under institutional pressure. Budget crises, competitive threats, workforce shortfalls, security emergencies — these are the organizational equivalent of a countdown clock. And in those moments, the AI vendor who shows up with a confident demo and a clean dashboard looks like a lifeline.

The procurement process compresses. The pilot evaluation shortens or disappears. The critical questions — What does this system actually optimize for? How does it fail? Who is accountable when it does? — get deferred. Not maliciously. The people making these decisions are under real pressure. They are not being reckless. They are doing what humans under pressure always do: they reach for the tool that reduces uncertainty fastest, and they trust it more than the evidence warrants.

This is the Crisis Adoption Problem. Not a bug in the adoption process. A feature of how human cognition and institutional incentives interact under stress.

Why This Matters More Now

AI tools have never been more capable, more accessible, or more aggressively marketed. The pitch cycle has accelerated dramatically. And many of the domains where AI is being adopted fastest — healthcare, defense, education, public safety — are exactly the domains characterized by chronic resource pressure, high stakes, and time-sensitive decisions.

In other words: the domains where the Crisis Adoption Problem is most likely to manifest are the same domains where its consequences are most severe.

A miscalibrated trust relationship between an operator and an AI system in a low-stakes commercial context is recoverable. In a clinical setting, a security operation, or an educational intervention, the same miscalibration looks very different.

This Is Not an Argument Against AI

I want to be precise about what I am and am not claiming here. I am not arguing that AI tools are dangerous, that adoption should slow, or that institutions should be more skeptical in a general sense. Many AI systems are genuinely useful. Some are transformative.

What I am arguing is that the adoption context shapes how humans relate to these tools — and that crisis conditions systematically bias that relationship toward over-reliance. The tool may be fine. The relationship to the tool, formed under pressure and never properly calibrated, is the problem.

My research operationalizes this as the understanding-trust gap: a measurable divergence between how much a user understands about an AI system's outputs and how much they trust those outputs. That gap widens under pressure. And wide gaps, in high-stakes domains, cost people.

What Would Help

The honest answer is that there is no clean fix. The Crisis Adoption Problem is structural — it emerges from the interaction of human cognition, institutional incentives, and market dynamics in ways that a checklist or a policy cannot fully address.

But a few things can make it less catastrophic:

Pre-crisis calibration. Organizations that have established deliberate relationships with AI tools before crisis conditions arise are far better positioned than those who first encounter a tool when urgency is already high. Trust calibrated under low-pressure conditions is more robust and more accurate.

Building for the gap. AI system designers who understand the understanding-trust gap can build for it — through transparency features, confidence indicators, and interface designs that make it harder to trust outputs you don't understand. My MICA framework research examines exactly this: how interface design shapes the gap.

Naming the pattern. The most underrated intervention is simply recognizing this dynamic when it is happening. If an organization can notice — we are in crisis conditions, this is when our evaluation is least reliable, we should slow down — that recognition alone can interrupt the pattern.

A Note From My Research

The MICA framework I developed, and the N=150 study comparing MURDOC and FACE interfaces, was fundamentally about what happens in that gap between understanding and trust. The interfaces that performed best were not the ones that produced the most accurate outputs. They were the ones that gave operators the clearest signal about when to trust and when to question.

That is, I think, the right design target for AI systems in high-stakes domains: not maximum accuracy in isolation, but maximum accuracy in the context of a human who is fallible, pressured, and making real decisions in real time.

The Crisis Adoption Problem is not going away. The pressure to adopt is only increasing. But if we understand what we are doing when we reach for AI under crisis conditions — and what it costs us to calibrate trust under pressure — we can at least make more deliberate choices about how we do it.

That seems worth working on.