This summer, I got on a phone call with the CEO of startup Springboards. The company develops large language models that focus on generating more diverse responses, aiming to break out of the groupthink trap that mainstream large models have fallen into. Right at the start of our call, he said something I still remember to this day: "We always say we're a self-loathing AI company. None of us are even sure we actually believe in what we're doing."

I joked back: "Then I guess I'm a self-loathing AI reporter."

I do love my job, but I really can't stand many forms of the technology I'm covering right now: it's twisted by hype, swept up by fanatics, and it's everywhere—you can't even escape it.

We're far from the only ones with this contradictory mindset. Around the world, having mixed feelings about AI has become the norm.

A Pew Research Center survey conducted in June 2026 found that more American adults believe AI will bring more harm than good to individuals and society as a whole, while the share of people holding optimistic expectations is much lower, with young people showing the strongest pessimism.

Stanford University's *AI Index Report 2026* notes that more than half of respondents globally say AI products and services make them feel uneasy.

Gallup poll data from May this year shows that 71% of American adults oppose building new AI data centers in their local residential areas. By comparison, the share of respondents opposing new nuclear power plants is 53%.

In a March NBC poll this year, AI's public approval rating was even lower than that of ICE—U.S. Immigration and Customs Enforcement, which has long been mired in controversy over its tough law enforcement style and has consistently had poor public reputation.

Contrary to the steadily declining public approval, the number of AI users is skyrocketing.

Data from market analysis firm Sensor Tower shows that in May this year, ChatGPT's monthly active users exceeded 1 billion. Google DeepMind's Gemini is close on its heels, hitting 950 million monthly active users in July.

The same Pew survey mentions that half of American adults report having used chatbots—more than double the figure in 2023, and a quarter of those users use chatbots every day.

This usage habit has spread across the globe. The Organisation for Economic Co-operation and Development counts 38 member states, covering most of the world's high-income economies, and its statistics show that more than one-third of adults have used a generative AI tool in the past three months.

All the surveys point to two facts that hold true at the same time: a lot of people genuinely don't like AI, and at the same time, a lot of people can't live without it.

Some people guess these are two completely separate groups of people, but I don't buy that. Unless the Venn diagram of AI skeptics and frequent AI users almost completely overlaps, these numbers don't add up logically.

There's another explanation: the more people interact with AI, the worse their opinion of it becomes. At least existing data supports this correlation: developed countries in the Northern Hemisphere with the highest AI penetration generally have more pessimistic public attitudes toward the technology; developing countries in the Southern Hemisphere with lower penetration have more optimistic public attitudes.

In my view, when people criticize AI, they rarely point the finger at the technology itself. Most of the anger targets the practices of the tech companies behind it: they push AI into every accessible scenario they can, and constantly hype it up to the public, claiming this technology will bring the biggest social and economic upheaval in generations. When they go that far, what good mood can they expect from ordinary people?

This isn't a new story. We saw almost exactly the same plot play out during the social media boom of the past two decades. Despite the growing tide of criticism against big tech, billions of people still flocked to Facebook and Twitter, and Google Search went through the exact same phase.

Back when social media was taking off, ordinary people had very little say. If you wanted to leave a platform, you had to give up all the content and social connections you stored there, so you either gritted your teeth and stayed, or abandoned everything you'd built and started over from scratch.

When it comes to AI, ordinary people still have room to make an impact. At least for now, policymakers have far stronger willingness to push regulation than they did back then. All 50 U.S. states have either passed or proposed bills regulating AI development and deployment, with the total number of relevant bills across the country exceeding 2,100—a tenfold increase in three years.

There is also still room for choice on the supply side. There are already many top-tier open-source large models on the market today, with capabilities matching the closed-source products from Google, OpenAI, and Anthropic. When consumers have options, the pressure of market competition can be passed on to the companies.

I don't want to be naïve about this. Trillion-dollar market cap tech companies are massive, and they aren't easily moved by external forces. But the future trajectory of AI has never been a foregone conclusion, unlike what these companies have implied to the public.

I asked the CEO of Springboards back then: if you don't really believe in AI, why are you building a new model? He answered that there's no going back on the development of large language models, but we can still steer it in a different direction.

I hope to see more of this "different direction" going forward: AI can clearly articulate its own limits, and more importantly, it can stop constantly acting like it's going to take over the world any day now.

This article was compiled and translated from *MIT Technology Review*, written by Will Douglas Heaven.