I spend a lot of time around people who build with AI. Developers, researchers, startup founders — the kind of folks who talk about context windows and agent architectures the way basketball fans talk about playoffs. In this bubble, AI is the most exciting thing happening. It’s the future, and it’s arriving faster than anyone expected.
Then I step outside that bubble and the conversation changes completely.
A new study from Pew Research dropped on June 17, and the numbers are brutal. Only 16 percent of Americans believe AI will have a positive impact on society over the next 20 years. Nearly 40 percent say it will be negative. Two out of three Americans think AI development is moving too fast. And here’s the kicker: the people who should be most excited about the future — young adults under 30 — are actually the most pessimistic. Only 14 percent of them see a positive future with AI.
That gap — between the people building AI and the people living with it — is the most important story in tech right now. And if you’re in the industry, dismissing these numbers as “people just don’t understand the technology” would be the worst thing you could do.
What the Numbers Actually Say
The Pew study is worth reading in full, but here are the numbers that should keep AI executives up at night:
Trust is almost nonexistent. Sixty-seven percent of Americans do not trust the U.S. government to meaningfully regulate AI. Fifty-nine percent do not trust AI companies to develop the technology safely. When both the regulators and the regulated are distrusted by roughly the same margin, you have a systemic credibility problem — not a messaging problem.
Usage and trust are completely decoupled. About 44 percent of U.S. adults now use ChatGPT. A quarter of Americans use AI chatbots daily. These are not people who reject technology — they’re actively using it. And yet only 16 percent think its long-term impact will be good. That’s not ignorance talking. That’s people who have tried the product and still have serious concerns about where it’s heading.
Half the country isn’t even on board. Fifty percent of Americans say they never use AI chatbots at all. Among people 65 and older, that number jumps to 75 percent. The primary reason isn’t lack of access or technical literacy — it’s a simple lack of interest, with no intention to adopt in the future. You can’t bridge a trust gap with people who’ve already decided the technology isn’t for them.
The Industry Has Earned This Skepticism
Let’s be honest: a lot of the skepticism is earned.
In just the past few weeks on this blog alone, I’ve covered stories that would make any reasonable person nervous about AI. Earlier this month, KPMG published an AI-generated report that was built almost entirely on fabricated data — hallucinations passed off as research by one of the world’s largest consulting firms. If KPMG can’t tell the difference between real analysis and AI confabulation, why should the average person trust anything generated by these systems?
Then there’s the security angle. OpenAI had to introduce a Lockdown Mode because AI agents — the very systems companies are racing to deploy everywhere — can be tricked into executing malicious commands, leaking sensitive data, or following instructions hidden in web pages. This isn’t theoretical. It’s happening right now, and the fixes are Band-Aids on a fundamentally unsolved problem.
Meanwhile, Google raised $85 billion for AI infrastructure — a number so large it’s hard to comprehend. Anthropic and OpenAI are racing toward IPOs with valuations in the hundreds of billions. The money pouring into AI makes the dot-com era look restrained. But here’s what nobody in those pitch decks is saying: the public — the actual users and customers — are increasingly uneasy about where all this is headed.
The Trust Paradox That Nobody’s Solving
Here’s the thing about trust: you can’t buy it, you can’t ship it as a feature, and you definitely can’t LLM your way out of a credibility crisis.
The AI industry is trapped in a paradox of its own making. On one hand, companies want us to believe AI is transformative — powerful enough to reshape entire industries, automate complex work, and maybe even achieve AGI. On the other hand, when something goes wrong — a hallucination, a jailbreak, a biased output — the response is almost always “well, it’s still early days, these are known limitations, the user should have known better.”
You can’t have it both ways. You can’t sell AI as revolutionary and then dodge accountability by calling it experimental. The public has figured this out.
And then there’s the speed problem. Two-thirds of Americans think AI is advancing too quickly. I understand why — when you see new models dropping every few weeks, each one more capable than the last, each one trained on more data than any human could consume in a thousand lifetimes, the pace feels reckless. Nobody voted for this speed. Nobody was asked whether they wanted AI integrated into every search result, every email client, every messaging app. It just happened.
What makes this particularly tricky is that some of the concern is legitimate, and some of it is fear of change — and those two things are very hard to untangle. But dismissing all skepticism as Luddism is lazy, and it’s exactly what the industry has been doing for the past two years. The Pew numbers suggest that strategy isn’t working. The recent government export ban on Anthropic’s most capable model — imposed with little explanation and no public consultation — is the kind of heavy-handed action that makes people wonder who’s really steering the AI ship.
What Should Actually Happen Next
I don’t think the answer is to slow down AI development. That ship has sailed, and frankly, I don’t want it to slow down — the technology has genuine potential to improve healthcare, accelerate scientific research, and make software development more accessible. I’ve seen what tools like GitHub Copilot and Claude can do for developers who used to struggle with boilerplate and documentation. That’s real value.
But the AI industry needs to do three things, and it needs to do them now:
First, stop treating public skepticism as a PR problem. The 16 percent number isn’t a messaging failure — it’s a trust deficit built on real incidents and reasonable concerns. When KPMG publishes AI hallucinations as analysis, when AI agents can be jailbroken with hidden text, when facial recognition systems show persistent racial bias — these aren’t “perception issues.” They’re product failures. Fix the products, and some of the trust will follow.
Second, give people genuine opt-out mechanisms. One of the reasons people feel uneasy about AI is that they didn’t choose to participate. AI summaries appeared in Google search results. AI features appeared in Gmail and Office. AI chatbots popped up in every app. There was no consent, no rollout, no choice. When people feel like technology is being done to them rather than for them, distrust is the natural response. Simple, honest opt-out buttons — not buried in settings menus — would go a long way.
Third, the AI safety conversation needs fewer promises and more proof. Every major AI company has a safety team and a responsible AI charter and a blog post about their commitment to ethical development. But when 59 percent of Americans don’t trust these companies to develop AI safely, those documents aren’t working. What would work: independent audits with published results. Clear liability when AI systems cause harm. Actual consequences for safety failures — not just PR statements.
Here in the Philippines, we feel these dynamics differently. AI adoption here is growing fast — Filipino developers are among the most active on GitHub, and tools like ChatGPT have become essential for students and professionals who use English as a second language. But the regulatory conversation is almost nonexistent. The US administration’s AI executive order set a direction, but it left enormous gaps that other countries are watching closely — and not necessarily filling.. We’re adopting the technology without having any of the safety conversations that countries like the U.S. and EU are at least attempting. That’s a gap we need to close.
Bottom Line
The Pew numbers should worry everyone in the AI industry — not because they’re unfair, but because they’re accurate. The trust deficit is real, it’s growing, and it’s being fed by genuine product failures, regulatory vacuum, and an industry culture that treats skepticism as ignorance.
Sixteen percent is not a number you fix with better marketing. You fix it by building AI that deserves trust — and right now, we’re not there yet.