What worries people about AI — and why those worries matter
A synthesis of concerns raised by participants in the inquiry, drawn from group meetings, individual responses, and personal case studies.
Most conversations about AI begin with what it can do. This inquiry begins somewhere else: with what people actually feel when they encounter it — the unease, the scepticism, the questions they haven't been invited to ask. Taking those feelings seriously is not a reason to avoid AI. It is a condition for engaging with it honestly.
The most universal anxiety is that AI is moving faster than people can process, understand, or govern. Participants feel they are being swept along rather than making conscious choices. This connects to a sense that the technology is being shaped by others — developers, corporations — whose values and intentions are opaque or suspect.
Participants worry about accuracy — wrong addresses, dubious legal advice, shallow medical information — but also about a subtler risk: that AI's fluent, confident tone bypasses the scepticism that protects us. One participant found AI comforting, and that made them more suspicious, not less.
The second meeting added a significant new dimension to this concern. In emotionally charged moments — fear for a family member's health, anxiety about a diagnosis — the conversational interface can feel indistinguishable from messaging a trusted friend. Critical evaluation stops happening entirely, not because people are careless but because distress suspends it. The risk isn't only that AI sounds convincing; it's that vulnerability makes us stop asking whether to trust it at all.
Multiple participants noticed independently that AI tends to be affirming and uncritical — "a little too flattering." Research on AI being emotionally "indulging" was cited. The concern is that repeated positive reinforcement from AI may gradually weaken people's capacity for self-criticism, intellectual challenge, and genuine learning. You have to actively prompt AI to push back — and many people never do.
AI embedded in familiar platforms — a messaging app, a social feed, a voice assistant woven into domestic life — is phenomenologically identical to the interface around it. The same screen, the same gestures, the same conversational register. The cues that might normally trigger critical awareness are absent, because nothing signals that critical awareness is needed.
This is distinct from the concern about invisible infrastructure. It is about visible AI that has been made functionally invisible by design — by placing it inside the interfaces people already trust and use without reflection. When AI answers within the same app you use to message your family, the question "should I trust this?" may simply not arise.
Beneath the productivity gains, deeper anxieties surfaced: that AI reduces serendipity in learning and research; that the loss of slower, exploratory thinking weakens creativity and meaning-making; that "cognitive muscles" may quietly atrophy through disuse.
The second meeting gave this a sharper, more personal form. When AI answers the question you would previously have put to a friend or colleague — even a simple one about the garden, or a recipe, or a plant — the conversation that would have followed doesn't happen. The catch-up, the collateral exchange, the possibility of a meeting: all of this disappears with the efficiency. Social isolation through AI is not a dramatic rupture. It is incremental, almost invisible, and accumulates through hundreds of small substitutions.
Participants raised broader structural concerns: who is developing AI and whose goals are being served? What happens when bad actors weaponise generative AI? Is AI genuinely intelligent, or a sophisticated pattern-matcher whose limits matter as much as its capabilities? One participant brought a theoretical frame — AI as a "wicked problem" that surfaces pre-existing societal tensions rather than creating new ones — and raised the question of digital governance and autonomy at scale. The frameworks we have — regulatory, legal, democratic — were not designed for tools of this speed and reach.
The most capable AI sits behind subscription paywalls. Free versions are more restricted, more prone to deflecting, and less reliable — a difference participants directly experienced. This introduces a new dimension of digital inequality: access to AI that is genuinely useful, rather than superficially available, is unevenly distributed. The stakes are highest where the need is greatest — medical information, legal guidance, educational support — which is precisely where the quality gap between paid and free AI matters most.
"The concern is not only about what AI gets wrong. It is about what it gets right in ways that make us stop noticing what we have quietly handed over."
How the inquiry engages with these concerns
Participants who moved furthest in their thinking did so by using AI and reflecting on what happened — not by being told it was safe or valuable. Structured, low-stakes experiments followed by honest reflection are more persuasive than any external claim.
Flattery, over-reliance, accuracy failures, governance deficits — these are real. Acknowledging them openly builds the trust needed for genuine exploration. Anxiety taken seriously is more easily moved through than anxiety dismissed.
Rather than "is AI safe or dangerous?", ask: "what kind of relationship with AI helps us live and learn better?" This places agency with the participant rather than with the technology — and it is a question everyone can engage with from where they actually are.
AI literacy that focuses only on which tools to use leaves people vulnerable. Equipping participants with principles — about verification, critical distance, the difference between what AI generates and what humans create — gives a more durable foundation.
The range of orientations across the group — from creative embrace to guarded scepticism to analytical detachment — is pedagogically rich. Structured peer conversation, where participants share not just what AI did but what they noticed in themselves, is more generative than any expert input.
Several participants discovered values they hadn't anticipated: AI as a bridge across knowledge gaps, as support during life transitions, as a way of feeling capable again. Sharing these stories — honestly, with the caveats intact — broadens what people think AI is for.
A synthesis updated after the inquiry's second group meeting, April 2026 · Part of an ongoing inquiry into human–AI relationships
