AI Use Climbs as Public Sentiment Sours, MIT Tech Review Says

Executive Summary

• Public sentiment toward AI is souring even as usage climbs, with Pew Research Center data showing more US adults expect a negative personal and societal impact than a positive one, and pessimism strongest among young people (MIT Tech Review AI).
• Adoption is nonetheless accelerating: ChatGPT reached a billion monthly users in May and Google DeepMind's Gemini logged 950 million users in July, according to the market analysis firm Sensor Tower (MIT Tech Review AI).
• Half of US adults now say they use a chatbot, more than twice the share in 2023, and one in four report daily use, according to Pew (MIT Tech Review AI).
• Local opposition is measurable: a May Gallup poll found 71% of US adults would oppose a new AI data center in their area, compared with 53% who would oppose a new nuclear power plant (MIT Tech Review AI).
• Regulatory activity is expanding, with all 50 US states having passed or proposed AI laws, producing a patchwork of more than 2,100 bills—a tenfold increase in three years (MIT Tech Review AI).

Key Takeaways

• The gap between what people say about AI and how much they use it is best explained, in the author's account, by dislike of corporate deployment pressure rather than of the technology itself.
• Pessimism and adoption split geographically: the Global North, where adoption is highest, skews negative, while the Global South, with lower adoption, is more optimistic.
• AI differs from social media's techlash because switching costs are lower and political appetite for regulation is higher, leaving more room for consumer and policy pressure.
• The piece is an opinion-driven analysis by Will Douglas Heaven, not a study, and its central claim is framed as the author's interpretation rather than a measured finding.

The AI Love-Hate Gap MIT Tech Review AI Describes

MIT Tech Review AI's analysis opens with an anecdote that frames the whole piece: a conversation with the CEO of Springboards, a startup building an LLM designed to produce a wider variety of responses than mainstream rivals. The CEO described his own company as "self-loathing," saying, "We don't know if we really like what we're doing." The author, Will Douglas Heaven, extends the joke to himself and then to the broader market.

The contradiction he sets out to explain is empirical before it is philosophical. Survey after survey registers negative feeling. Stanford University research cited in the piece found more than half of people worldwide say AI products and services make them nervous. An NBC poll in March found AI less popular than ICE. Yet the same public keeps signing up. The numbers, Heaven argues, don't add up unless the groups overlap—unless the Venn diagram is becoming a circle.

His answer is that the dislike is not aimed at the technology. It targets the companies pushing it into as many parts of life as possible and telling the public to brace for the biggest social and economic upheaval in generations. Heaven writes that he does not want to be naive about moving trillion-dollar companies, but he argues what comes next is less inevitable than those companies imply.

Adoption Data Behind the Sentiment Split

The usage figures are the harder half of the paradox and the piece leans on third-party measurement to establish them. ChatGPT crossed a billion monthly users in May, per Sensor Tower, with Google DeepMind's Gemini close behind at 950 million users in July. Pew data shows half of US adults now use a chatbot, more than double the 2023 level, and one in four do so daily. The habit is not confined to the US: more than a third of adults across all 38 OECD countries report using generative AI tools in the previous three months.

Against that, the opposition is concrete rather than abstract. Gallup's May poll put opposition to a new local AI data center at 71%, well above the 53% who would oppose a new nuclear power plant. That gap matters for anyone siting infrastructure. It suggests the friction point is not model capability but physical and visible deployment.

Heaven offers two possible readings and rejects the simplest. It could be two distinct populations, but he doubts it. It could be that using AI breeds negativity, and there is a correlation: the Global North, where adoption is highest, is more pessimistic, while the Global South, where adoption is lower, is more optimistic. He presents that as correlation only, then gives his own explanation.

Related: Oura 2026: Ring 5 Launches Blood Pressure Tracking Under FDA Carve-Out

Why the Comparison to Social Media May Not Hold

The historical parallel in the piece is social media's last two decades, when billions joined Facebook and Twitter despite a mounting techlash against the companies running them, and Google search followed a similar pattern. The key difference Heaven identifies is exit cost. Quitting a social platform meant losing content and connections, so users either accepted the terms or started over.

With AI, he argues, there is still room for user influence. Two mechanisms are cited. First, political appetite for regulation is stronger: all 50 US states have passed or proposed laws governing development and deployment, creating a patchwork of more than 2,100 bills and a tenfold increase in three years. Second, open-source alternatives to Google, OpenAI and Anthropic are already on the market, which Heaven says creates at least the potential for more consumer choice and the market pressure that comes with it.

The piece does not quantify how much leverage either mechanism actually delivers. It states the potential and leaves the outcome open. The closing appeal is for AI that is clear about what it can and cannot do and that does not posture as if it is about to take over the world—a standard Heaven attributes to the Springboards CEO's point that there is no walking back from LLMs, but there is room to make them behave differently.

For deeper context, see our AI analysis: "Rituals Confirms Customer Data Breach in Cosmetics Sector 2026".

Sentiment and Adoption Signals in the MIT Tech Review AI Report

The source does not specify geography for Springboards or Sensor Tower's measurement scope, so those cells state the limitation rather than supply an unsupported location. No named individual other than the springboards CEO and the author is identified, and the source does not name that CEO.

What This Means for Practitioners

For enterprise buyers, founders and procurement teams, the practical signal is that adoption data and sentiment data must be read together rather than substituted for each other. Usage is expanding across the US and the OECD, which supports continued investment in deployment. But the Gallup finding that 71% of US adults would oppose a local AI data center, against 53% for nuclear power, indicates that siting and visible infrastructure carry reputational and permitting risk that model benchmarks will not capture. The regulatory patchwork of more than 2,100 state bills means compliance planning cannot assume a single national standard. Treat open-source alternatives as a live procurement option, as the source frames them as a source of consumer and market pressure.

MIT Tech Review AI Implementation Risks

Disclosure: this section is independent editorial analysis and is not sponsored, reviewed or approved by the source publication.

Additional coverage: Top 10 Drones Investment Opportunities in 2026

Three risks follow directly from the supplied text. First, sentiment risk: if negative public feeling is concentrated among younger users and strongest in high-adoption regions, then customer-facing AI features may face resistance that does not show up in usage counts. Second, infrastructure risk: local opposition measured at 71% against data centers is a deployment constraint for any organization that depends on new compute capacity in the US. Third, regulatory fragmentation risk: with all 50 states active and more than 2,100 bills introduced or passed, a tenfold increase in three years, multi-state operations face shifting obligations rather than a stable baseline.

The source does not establish that sentiment has caused measurable adoption declines, nor does it quantify regulatory costs. The evidence to watch is whether the correlation the author notes between high adoption and pessimism holds as usage spreads, and whether the political appetite for regulation he cites translates into binding rules. Source note: MIT Tech Review AI.