There is a particular kind of discomfort in building something you are not sure you should be building. Will Douglas Heaven, a senior AI journalist at MIT Technology Review, discovered a version of that feeling during a conversation with the CEO of Springboards, a startup developing a large language model designed to produce a wider range of responses. The executive's confession was candid, almost startling: "We often say that we're a self-loathing AI company. We don't know if we really like what we're doing."
Heaven's response — that it made him a "self-loathing AI journalist" — was a joke. But the underlying sentiment it points to is anything but funny. It captures something true and strange about where we are with artificial intelligence right now. People distrust it. Researchers question it. The public is turning against it. And yet usage is climbing faster than almost any technology in recent memory. This is the AI popularity paradox, and understanding it matters more than the industry's boosters would like to admit.
Why People Hate AI But Can't Stop Using It
Contempt and compulsion rarely coexist so openly. With social media, users took years to articulate their ambivalence. With AI, the discomfort arrived almost simultaneously with the products themselves.
Part of the explanation is structural. AI tools are embedded everywhere now — in search engines, in writing software, in customer service portals, in healthcare triage systems. Opting out is less a personal choice than a negotiation with infrastructure. A knowledge worker who finds AI-generated summaries philosophically objectionable still encounters them in their inbox, their productivity suite, their research tools. The technology has become, as Heaven put it, inescapable.
But that only explains passive exposure. It does not explain the millions of people who actively open ChatGPT, Microsoft Copilot, or Google Gemini every day knowing full well how they feel about them. For that, you need a different frame: utility friction. The tools work well enough, often better than the alternatives, for specific tasks. Grammar checking, code debugging, first-draft generation, summarizing long documents — in these narrow applications, the value proposition is real and immediate. The philosophical objections are real too, but they live in a different mental register than the deadline pressure at 4 p.m.
That gap — between what we think about a technology and what we do with it — is exactly where the AI popularity paradox lives.
Public Sentiment Is Souring on Artificial Intelligence
The shift in public opinion is measurable and accelerating. Across surveys from the Pew Research Center, tracking conducted by Gallup, and Edelman's annual Trust Barometer, a consistent picture is emerging: trust in AI companies is declining, and anxiety about AI's societal impact is rising. As Heaven reported, polling now shows more people expect AI to have a negative impact on society than a positive one — a significant reversal from the cautious optimism that characterized early ChatGPT coverage in 2023.
Read next Laika's Wildwood: Stop-Motion Fantasy at TIFF 2026Pew Research surveys from recent years have documented growing concern among Americans about AI's role in hiring, surveillance, healthcare decisions, and content generation. Younger adults — the demographic most likely to use AI tools daily — are not exempt from this anxiety. They are often simultaneously the heaviest users and the most worried about long-term consequences.
The reasons behind souring sentiment are not mysterious. High-profile failures have accumulated: AI-generated misinformation spreading before elections, deepfake scams targeting ordinary people, hiring algorithms quietly discriminating in ways that are difficult to detect and harder to challenge. The technology is warped by hype, as Heaven observed — announcements routinely outpace capabilities, and when reality catches up to marketing, trust erodes.
There is also a more diffuse unease that polling struggles to fully capture: the sense that AI is being deployed not because it is the best solution, but because it is cheap, scalable, and good for quarterly earnings. When a company replaces its customer support team with a chatbot that cannot understand context and apologizes for everything, users notice. They remember.
AI Usage Keeps Skyrocketing Despite Backlash
None of this negativity shows up in usage data. ChatGPT reportedly reached 100 million monthly active users faster than any consumer application in history — a record it set within roughly two months of its public launch. McKinsey's State of AI research has consistently documented year-over-year increases in enterprise AI adoption across nearly every industry sector. Mobile app downloads for AI assistants continue to set records on both major platforms.
This is the data point that should make everyone pause. When people say they distrust something and then use it anyway, we usually interpret that as cognitive dissonance. But it is worth asking whether the framing is slightly wrong. Distrust of AI and frequent use of AI may not be contradictory positions at all. They may be a perfectly rational response to a technology that has been thrust into the center of daily life before society had any meaningful say in how it should be governed, limited, or distributed.
People are not confused. They understand exactly what they are dealing with. They just do not have a better option in many cases, and the specific utility of the tool in front of them is real, even when the broader phenomenon terrifies them.
Inside the Industry: Even AI Builders Have Doubts
The Springboards CEO's admission is striking not because it is unusual, but because it is unusually honest. Inside the AI industry, versions of that ambivalence are common. Researchers who spent years working on safety problems have left major labs citing concerns about the pace of deployment versus the pace of alignment research. Engineers who built the systems that now generate persuasive political content have spoken publicly about regret.
Heaven noted that the technology has been "steered by zealots" — a pointed observation from someone who covers it full-time. The reference is to a strain of AI accelerationism that treats any concern about deployment risk as either ignorance or obstruction. In that worldview, the answer to every problem caused by AI is more AI, faster. Critics within the field find this posture genuinely alarming.
The self-loathing that the Springboards CEO described is not paralysis. It is a form of moral seriousness that the loudest voices in the industry often lack. Acknowledging that you are not sure your work is net-positive is not a reason to stop, but it is a reason to proceed carefully, to prioritize oversight, and to resist the narrative that speed is always virtue.
EmTech Future 2026: The Conversation the Industry Needs
MIT Technology Review's EmTech Future 2026 arrives at a moment when this tension is impossible to ignore. Events like EmTech matter precisely because they create space for the kind of honest reckoning that quarterly earnings calls and product launch events deliberately avoid. The AI popularity paradox is not a communications problem that better marketing can fix. It is a substantive problem about what the technology is being built to do, for whom, and at whose expense.
The conversations that need to happen in forums like this one are not primarily technical. The engineering challenges of AI are enormous but tractable. The harder questions are about governance: who gets to decide what AI systems are deployed in public services? Who bears liability when they fail? How should productivity gains be distributed rather than concentrated? What democratic mechanisms exist to slow or redirect development that a public majority opposes?
These are not abstract philosophical questions. They are the practical challenges that determine whether the current wave of AI development ends in something broadly beneficial or in a significant and hard-to-reverse transfer of power away from ordinary people.
What the AI Paradox Means for the Future of Technology
The AI popularity paradox is not a phase. It is a structural feature of how transformative technologies embed themselves into society before accountability structures exist to govern them. The internet went through something similar. Social media's consequences are still being worked out after two decades.
What makes AI different is the speed of deployment and the breadth of penetration. It is in hospitals and courtrooms and classrooms simultaneously, without the gradual rollout that allowed earlier technologies to be studied and regulated at a reasonable pace.
The self-loathing CEO, the ambivalent journalist, the user who distrusts AI but opens the app anyway — these are not outliers. They are the majority, navigating a technology that has outrun their ability to evaluate it, in a policy environment that has not caught up, embedded in products they did not fully choose. Acknowledging that tension honestly is not pessimism. It is the beginning of a more honest conversation than the industry has been willing to have.
Related coverage
- How 2008 Reshaped the US Energy Relationship
- Biotech's Future: MIT's Top Innovators Under 35 in 2026
- Future of edge computing — Complete Guide
Source: MIT Technology Review
