Doctoral Research · Drexel University DBA · 2025
Doctoral Thesis
Why people champion AI they don't fully trust — three years and 715 survey responses on what actually makes someone become an advocate for generative AI. The data didn't say what I expected.
The conventional wisdom goes something like this: trust AI, and you'll advocate for it. Distrust it, and you won't. Simple, symmetric, intuitive. It's the same logic we apply to almost every technology adoption story.
My dissertation — three years, a quantitative study of 715 working professionals, and more SPSS output than I care to admit — set out to test that logic. It didn't hold.
What the data revealed is messier, more interesting, and arguably more useful for anyone thinking about how organizations actually deploy and scale AI. Trust and distrust are not two ends of the same dial. They are two separate systems. And only one of them reliably predicts whether someone will go out and tell others to adopt generative AI.
The setup
My research examined a specific behavior: advocacy intention — the likelihood that a current GenAI user would actively recommend and promote the technology within their organization. Not passive acceptance. Active championship.
I ran participants through two scenarios. In one, they were asked to imagine using GenAI to summarize a document — a relatively low-stakes task. In the other, they were asked to imagine using GenAI to generate financial advice — higher stakes, higher consequence if the system gets it wrong. Same technology. Different risk context.
The theoretical backbone came from two sources. The first is Suchman's (1995) legitimacy framework — the idea that people evaluate institutions (and technologies) across four dimensions: do I understand it (cognitive), does it align with ethical norms (normative), does it actually work for me (pragmatic), and does it comply with rules and regulations (regulative)? The second is the Lewicki and colleagues tradition of treating trust and distrust not as opposites but as qualitatively distinct psychological constructs, each with its own antecedents and consequences.
What I expected to find
My original model was built on what seemed like a defensible assumption: legitimacy builds trust, trust drives advocacy, distrust works in reverse. Higher risk moderates both relationships — making trust more fragile and distrust more potent.
Clean. Logical. Wrong.
Well — not entirely wrong. But the data forced a significant revision, and the revision turned out to be more instructive than the original hypothesis.
Finding one: legitimacy is not a single thing
The first discovery was methodological but had real consequences. When I ran the principal component analysis, legitimacy didn't cohere into a single construct. It split cleanly into four: cognitive, normative, pragmatic, and regulative — each exerting its own independent influence on advocacy.
That last point is worth pausing on. Organizations often lean heavily on regulatory compliance as a trust signal. This has been audited. This is certified. This meets the standard. The data suggests that framing alone doesn't move the needle for advocacy. What people care about more is whether they understand the system, whether it feels ethical, and whether it does what it promises.
Compliance is a floor, not a ceiling.
Finding two: distrust did not drive advocacy decisions
This is the finding that surprised me most — and that I think matters most for practitioners.
Trust was a strong, significant, consistent predictor of advocacy. When participants had confidence in GenAI's reliability, integrity, and ability to deliver — they advocated. That part of the hypothesis held.
Distrust was different. It didn't directly and significantly suppress advocacy. Instead, distrust was strongly tied to perceived risk. It activated in high-stakes situations — but even then, it didn't translate into refusal to advocate unless other factors, particularly legitimacy perceptions, had also deteriorated.
In simpler terms: distrust and advocacy can coexist. You can carry misgivings about a technology and still tell your colleagues they should use it. We see this in the real world constantly — people who use and recommend tools they have genuine concerns about, because the utility outweighs the anxiety, or because the trust dimension is strong enough to carry the weight.
Finding three: risk doesn't moderate — it precedes
My original model treated perceived risk as a moderator: risk would amplify or dampen the relationship between trust and advocacy. That's not what happened.
Instead, perceived risk operated as a direct antecedent. When participants were placed in the high-stakes scenario, their perceived risk increased — which then cascaded into reduced legitimacy perceptions and lower trust levels. Risk reshaped how people evaluated the system before they ever got to the question of advocacy.
This has practical implications for how organizations frame AI deployments in high-stakes domains — healthcare, finance, defense, legal. The goal isn't to convince someone that the risk is low. The goal is to ensure that even in acknowledged high-risk contexts, the system remains comprehensible, ethically grounded, and demonstrably useful.
What this means if you're deploying AI
The summary for practitioners is this: trust and distrust are separate problems requiring separate interventions. You cannot build enough trust to dissolve distrust. You have to address each on its own terms.
Building trust means transparency about how the system works, clarity about where it can fail, and demonstrated reliability over time. It means explainable AI — not as a compliance feature, but as a cognitive legitimacy investment.
Managing distrust means something different: it means engaging directly with the specific concerns that generate it. Fear of bias. Anxiety about job displacement. Apprehension about who controls the system and under what conditions. These don't dissolve through trust-building. They require direct acknowledgment, honest dialogue, and visible accountability mechanisms.
The dimensions of legitimacy give you a practical diagnostic. If adoption is stalling, ask which dimension is failing:
Is it cognitive — people don't understand how it works? Fix: explainability, user education, clear documentation of decision logic.
Is it normative — people feel it's ethically misaligned? Fix: visible ethical commitments, bias audits, stakeholder engagement.
Is it pragmatic — people don't see real-world value? Fix: concrete use-case demonstrations, ROI evidence, accessible success stories.
Is it regulative — compliance concerns? Fix: necessary but rarely sufficient. Layer the other three on top.
The honest limitations
This study is a snapshot. It relied on a professional sample — not a general population — and the findings may not hold in cultures or sectors with very different baseline attitudes toward technology risk. The measurement captured intentions, not actual advocacy behaviors; the gap between what people say they would do and what they actually do remains an open question.
The GenAI landscape also moves fast. The models available when I was designing this research are already several generations old. As explainability improves, as regulatory frameworks solidify, as hallucinations become less frequent — the legitimacy calculus will shift. Longitudinal research that tracks these perceptions over time is needed.
I say this not to undercut the findings, but because intellectual honesty about what a study can and cannot claim is part of what makes research worth reading in the first place.
Why I wrote this
I've spent more than two decades deploying technology at scale — in government, at AWS, at Google Cloud. The question of why some technologies get championed and others stall has practical stakes far beyond academic interest.
What I found is that the answer isn't primarily about the technology. It's about the sociotechnical scaffolding around it — the legitimacy people ascribe to it, the trust it earns over time, and the very specific, targeted work required to address the distrust that coexists with that trust.
Advocacy isn't the absence of doubt. It's a decision people make in the presence of doubt — when cognitive, normative, and pragmatic legitimacy are strong enough to carry the weight.
That's a more complicated truth than the simple trust-equals-adoption story. It's also, I think, a more useful one.
Committee chair: Dr. David Gefen, Professor of Decision Sciences, Drexel LeBow College of Business.