The AI Backlash Has a Point; I Think It Is Aiming at the Wrong Target.
Thumbs Down
I am pro AI.
I should probably get that out of the way before someone arrives with a digital pitchfork.
I use AI regularly. I think it is astonishing technology, and I suspect we are still fiddling with the knobs on something that will ultimately be every bit as consequential as the personal computer or the Internet.
I also think some of the people complaining about AI have a point. Actually, several points.
The mistake, in my view, is assuming that because AI creates problems, the logical answer is less AI. Many of the things making people angry are not really problems with artificial intelligence itself. They are problems with how companies deploy it, how governments regulate it, how humans use it, and our peculiar habit of inventing things first and deciding what the rules ought to be sometime around Thursday. That distinction matters.
Start with jobs.
A March 2026 Quinnipiac University Poll found that 70 percent of Americans believe AI is likely to reduce the number of job opportunities. At the same time, people are increasingly using AI for research, writing, work projects, and data analysis (Quinnipiac University Poll, 2026). There is something fascinating about that contradiction. We’re worried about the machine while simultaneously asking it to help with Tuesday's spreadsheet.
The fear is not ridiculous. Employers have an economic incentive to accomplish more work with fewer people. Pretending otherwise insults everyone involved. If five employees can suddenly accomplish the work of seven, somebody in accounting is eventually going to notice. My objection is to the assumption that displacement is the only possible outcome.
AI can also increase what an individual worker is capable of doing. The goal should be augmentation before elimination. Use AI to chew through repetitive research, summarize routine material, find patterns in data, draft the tedious first pass, or handle administrative sludge. Let humans keep the parts requiring judgment, accountability, experience, empathy, and the occasional realization that the computer has completely lost the plot.
Creatives also have a legitimate concern, and this is where AI enthusiasts sometimes become unnecessarily smug. Artists and writers are not simply afraid of a fancy new paintbrush.
Recent peer-reviewed research examining generative AI and creative work identified legitimate risks involving loss of autonomy, deskilling, professional precarity, isolation, and a gradual shift from creating something toward merely curating what an AI has produced. The same research, however, also found potential benefits, including broader access to creative tools and the consolidation of tedious tasks (Montefiore et al., 2026).
That seems like a remarkably sensible place to begin. Use AI to remove drudgery before using it to remove the artist. If I have spent decades learning to write, I do not particularly want my contribution reduced to clicking “regenerate” until the machine stops producing prose that sounds like a motivational poster in a dentist's waiting room. But if the same system helps me research, organize material, challenge an argument, or catch something I missed, I have not surrendered authorship. I have acquired another tool.
The question should be who is directing whom.
Privacy is a harder problem.
Pew Research Center found this year that 71 percent of Americans believe increased AI use will make their personal information less secure. Sixty-three percent also think AI is advancing too quickly. Perhaps more tellingly, 67 percent have little or no confidence that the federal government can regulate AI effectively, while 59 percent lack confidence that American companies will develop and use AI responsibly (Pew Research Center, 2026). I cannot wave that away.
AI systems consume data voraciously, and “trust us” is not a privacy policy. The answer has to include clearer limits on collection and retention, meaningful consent, stronger security, transparency about how information is used, and consequences when companies play fast and loose with people's data. We already know ways to build systems that expose less personal information and restrict access to sensitive data. What is missing is not imagination. It is consistent implementation and accountability.
Then there is the problem everyone who uses generative AI eventually encounters. Sometimes it makes things up. “Hallucination” is an almost endearing word for a machine confidently handing you nonsense.
The criticism is absolutely deserved. In a July 2026 peer-reviewed study of language models used for time series forecasting, researchers found that conventional measures of accuracy could hide significant reliability problems. More encouragingly, they also found that structured tokenization consistently reduced hallucinations across the datasets they examined (Abdullahi et al., 2026). The study concerns a specific type of AI forecasting, so it would be improper to claim that this particular fix solves hallucinations everywhere. It does demonstrate that these failures can be measured and that at least some of them can be mitigated through better engineering.
AI should also become better at saying three very important words: “I don't know.”
Then we have deepfakes, misinformation, and the rapidly deteriorating usefulness of “I saw the video.”
Quinnipiac found that 28 percent of Americans had shared a video they later discovered was AI-generated. When asked about AI-generated material in political advertising, 38 percent wanted it banned, and another 45 percent wanted disclosure requirements (Quinnipiac University Poll, 2026).
This problem worries me considerably more than an AI-generated picture of the Pope wearing an improbable coat. We are entering a world where convincing evidence can be fabricated cheaply, but there is an equally nasty problem lurking behind it. Once everyone knows fake evidence is easy to manufacture, genuine evidence becomes easier to deny.
The technical answer includes better provenance, authentication, detection, and labeling. The social answer is harder. We are going to have to become considerably more skeptical about where digital material came from without becoming so cynical that we believe nothing at all.
Finally, there is the physical cost of all this supposedly ethereal intelligence. AI runs somewhere.
It uses electricity. Data centers use water and occupy actual communities populated by actual humans who may be considerably less excited about the AI revolution when their electric bill arrives. In the Quinnipiac survey, 65 percent opposed an AI data center being built in their community. Among opponents, electricity costs and water consumption were leading concerns (Quinnipiac University Poll, 2026).
Again, the critics have a point.
The answer is not to pretend computation has no environmental cost. It is more efficient hardware and models, better energy planning, sensible siting, transparent reporting, water conservation, and making sure communities receiving the burden also receive meaningful benefits.
There is a pattern here. The backlash against AI is telling us something useful.
People do not want to lose their livelihoods. Artists do not want their craft hollowed out. Nobody wants their personal information treated like free buffet shrimp. We do not want machines inventing facts, synthetic video poisoning our ability to know what happened, or data centers quietly handing their costs to everyone living nearby.
Fine, but those are design requirements.
I remain pro AI precisely because I do not think these problems are reasons to abandon it. They are reasons to build it better, regulate it intelligently, use it deliberately, and occasionally tell the people racing to deploy it that perhaps moving fast and breaking things becomes considerably less charming when the “things” are people's jobs, privacy, communities, and trust.
The backlash should not be dismissed. The backlash becomes useful when we treat complaints as specifications rather than obituaries for the technology. It should be listened to. But perhaps the question we ought to be asking is not whether AI is becoming too powerful. What if the real problem is that we have become very good at asking what AI can do, while remaining remarkably reluctant to decide what we should let it do?
References
Abdullahi, S., Danyaro, K. U., Chiroma, H., Koh, T. W., Zakari, A., & Aliyu, Y. (2026). Hallucination in time series large language models: An empirical investigation and analysis of mitigation strategies. Scientific Reports. https://doi.org/10.1038/s41598-026-62952-y
Montefiore, T., Formosa, P., Bankins, S., & Sahebi, S. (2026). The impacts of generative AI on the meaningfulness of creative work. Journal of Business Ethics. https://doi.org/10.1007/s10551-026-06342-4
Pew Research Center. (2026, June 17). Americans and AI 2026: Chatbots, smart devices and views on impact. Pew Research Center.
Quinnipiac University Poll. (2026, March 30). The age of artificial intelligence: Americans' AI use increases while views on it sour, Quinnipiac University Poll on AI finds; 7 in 10 think AI will cut jobs with Gen Z the most pessimistic. Quinnipiac University.
