AI Skeptics Doubt a Slowdown Will Fix the Technology’s Real Harms

As talk of an artificial intelligence slowdown spreads through Silicon Valley, a growing chorus of critics is pushing back on a tempting assumption: that if the hype cools, the harms will cool with it.

For much of the past few years, the dominant story about AI has been one of acceleration — bigger models, larger data centers, richer valuations. More recently, the conversation has shifted. Investors question whether returns can justify the enormous sums being spent. Executives temper promises about what the next generation of models will deliver. Some researchers argue the field is approaching diminishing returns from simply scaling up.

For people who have spent years warning about AI’s downsides, that shift is not necessarily reassuring.

Why skeptics aren’t celebrating

The core of the skeptics’ argument is that the problems they have identified are not side effects of speed. They are features of how the technology is built, deployed and governed.

Automated decision systems already screen job applications, set insurance and lending terms, inform tenant screening and shape policing priorities. Those systems do not need to improve — or even work well — to affect people’s lives. A slower pace of model releases does nothing to remove a flawed tool that has already been installed in a hiring pipeline or a government agency.

The same logic applies to labor. Companies that have restructured teams around automation are unlikely to reverse course because the underlying technology plateaued. Once a workflow is rebuilt and headcount reduced, the disruption is largely locked in.

Critics make similar points about content. Synthetic text, images, audio and video are already cheap and widely available. Deepfakes, scam operations and low-quality machine-written material circulating online do not depend on further breakthroughs. The tools that exist today are sufficient.

Infrastructure that outlasts the hype

The physical footprint of the AI boom is another sticking point. Data centers under construction or recently completed represent long-term commitments of land, electricity and water, often negotiated with local governments under the promise of jobs and tax revenue. Those facilities and their energy demands remain whether or not the next model is a leap forward. Communities that raised concerns about utility costs and grid strain argue that a cooling market does not undo signed contracts or built infrastructure.

There is also the question of what the training data era left behind. Disputes over copyrighted work, scraped personal information and uncompensated creative labor are playing out in courts and legislatures, and a slower pace of development does not resolve them.

A risk of losing momentum

Perhaps the sharpest worry among critics is political. Much of the recent appetite for AI regulation has been driven by fear — of mass job loss, of election manipulation, of systems too powerful to control. If the perception takes hold that AI was overhyped all along, the urgency behind oversight efforts could evaporate before meaningful rules are written.

That would leave the current generation of systems in place, largely unexamined, while public attention moves on. For skeptics, the more useful framework is not how fast AI is advancing but who is accountable when it causes harm, what recourse affected people have, and whether companies must disclose how their systems make decisions.

Those questions do not have a speed setting. A quieter AI market may relieve pressure on investors and reduce the volume of grandiose claims. But as critics see it, transparency, liability and enforcement are the only tools that address the harms already in circulation — and none of them arrive automatically when the boom fades. Read More