Four Important AI Questions

Freddie deBoer

Oct 19, 2026

263

101

43

Share

In anticipation of my book’s release NEXT WEEK (!), I recently appeared on Christianity on the Spectrum and Autism Confidential, two podcasts representing points of view which have been systematically marginalized by “neurodiversity” activists and which deserve your attention.

Two recent useful pieces on AI were published recently. Not useful in the sense that the people who wrote or featured in them are right, but useful in showing the reality distortion field that the media has erected around this topic.

First, of course, is Ezra Klein. His recent column is valuable because it’s such a stark statement of a decision that he made (subconsciously, probably) long ago: to insist that the only possible outcomes are AI doom or AI utopia, that the only possibilities are maximalist in one direction or another. He refuses to really entertain the possibility that, in fact, life will not be radically ruptured, that LLMs could prove to be significant and consequential software but ultimately just software, normal technology. He has crafted an exquisite mental safehouse for himself, such that he never has to entertain the possibility that tomorrow will be much like today - which is always a safe bet. He frequently says that “everyone he talks to” says AGI is coming soon, but “everyone he talks to” is someone with direct financial incentive in that being true. It’s a similar story with his show; Klein, now primarily a podcast host, uses the booking process in much the same way that Joe Rogan does - as a screening mechanism to make sure that he doesn’t have to encounter this particular opinion he doesn’t like, without having to actually forbid any opinion in one of his conversations. It’s an effective tactic, if you want to appear evenhanded. But it doesn’t help us to understand anything.

preorder All In Your Head now

And this Kevin Roose interview with New York really epitomizes a certain AI Guy approach to defending the outlandish claims made about the technology, a lot of patronizing answers based on speculation and unfulfilled promises that Roose insists are simply going to be fulfilled. He and interviewer Benjamin Hart both exhibit a lot of condescension, suggesting that every skeptic is just a sad denialist who can’t accept the changes coming. That affect has become inescapable among the AI maximalists: put-upon, impatient, a countenance more in sorrow than in anger, paternal, forever theatrically sighing. It amounts to an assertion that skeptics have a fundamentally emotional relationship to AI… which is rich coming from guys who refuse to confront their own emotional relationship to the technology. If they get to suggests that skeptics are just afraid to face the future, I get to point out that they’re nerds who grew up dreaming of living in Star Wars and who now credulously rush to believe that we’re about to live in a much cooler sci-fi future. It bothers people when I point that out, how a refusal to let go of fantasy colors the AI debate. They don’t like considering people’s unspoken motivations. But if podcasters get to pull the world-weary Cassandra routine, then I get to point out that they seem unable to accept the likelihood that the world will just go on going on in the same old mundane boring way it always has.

Roose would like you to know that he finds skeptics very unserious, that we don’t engage sufficiently with the facts. Well, two can play at that game: I’m tired of AI maximalists dodging very basic questions about not just AI but about epistemology, the burden of proof, their own emotional motivations, and when and how they could be proven wrong. So here are four questions, just four little questions from little old me. If you’re an AI hype-type and would like to actually argue, here are questions that you should be able to answer.

If everything is changing, why has nothing changed?

This is the big one, and it’s the one that most often prompts this phenomenon of people pretending that rolling their eyes is an argument: if we’re living in the AI revolution, why does nothing seem revolutionary? If this is the most consequential technology of our lifetimes, and maybe in human history, why aren’t average people experiencing more consequences? Why has day-to-day life barely changed for most people? Why is so much of human life in 2026 indistinguishable from 2016, or indeed from 2006 or 1996? Another way to frame this question is to ask, What have LLMs enabled human beings to do that they simply couldn’t do before? I periodically ask this question on Notes, and the first wave of responses is always filled with “Are you joking???” and “Uh, have you been under a rock for four years???” and similar - arguments to incredulity. Then I challenge people to actually name changes that are required to meet any reasonable definition of revolutionary change: they have to be significant, material (as in, not happening purely in software), and relevant to ordinary people. And they struggle. Some examples I’ve gotten have included automating note-taking, avoiding phone calls with the cable company or other corporate entities, streamlining laborious spreadsheet work…. This is all quite convenient, I’m sure. But that’s not exactly species-altering, history-shaking epochal change - and it’s not going to pay for the trillions of dollars that LLMs have to generate in order to earn back their truly incredible costs.

Technologies that make things we could already do cheaper, or easier, or more efficient, or more automated, or safer, etc, can be important technologies, definitely. That kind of technology is in fact underrated. But to reach the heights that AI maximalists have imagined proposed predicted promised, we have to go much, much further than that. Self-driving cars are likely to be among the most consequential of current AI technologies, and I think the rise of self-driving cars as a mass option is more or less inevitable. (Mandating self-driving is no such thing; if you think it’s hard to get Americans to give up their guns….) But a) less than 1% of the population has been in a self-driving car to date, and b) cars already exist. Self-driving will be a real boon for safety, for convenience, potentially for efficiency. But the transition from human-driven cars to self-driving cars is nothing like as consequential as the transition from horse-based travel to the train or the automobile. Most of what LLMs are currently used for involves automating tasks that humans already do - economically, politically, and culturally meaningful, potentially, but not revolutionary. Meanwhile the more outlandish promises of unlimited clean energy, a cure for all disease, radical life extension, colonizing the stars, etc., are all firmly in the realm of science fiction even by the most generous and optimistic assumptions.

Now, I often get people suggesting that my standards for what constitutes significant change are too high. But I must again point out that the standards of significance here aren’t mine at all; they’re the standards created by the AI hype industrial complex. I didn’t say that LLMs constitute a radical break in human history, such that we’ll forever identify pre-LLMs as humanity’s “before” and post-LLMs as our “after.” The people hyping LLMs did that. I didn’t say that what’s happening is a revolution, a new epoch, a fundamental paradigm shift in world history. The people hyping LLMs did that. I didn’t say that we’re living through a new Industrial Revolution. The people hyping LLMs did that. I didn’t say that LLMs are a bigger deal than electricity or fire. The people hyping LLMs did that. I am always and only holding them to the exact standards they insist on; I’m taking their own rhetoric more seriously than they themselves do.

The ongoing failure of consequences to emerge that are a) significant, b) material, and c) of meaning to average people appears to bug the AI people. I think it bugs them a lot; you can see them picking at the idea over time, getting more and more fretful about it. They’re clearly bothered that no smoking gun has emerged, especially as it’s become clear that LLM protein folding is not leading to immediate massive breakthroughs. (And, for the record, protein folding is a pretty good example of LLMs assisting in what’s still fundamentally human research, not some new AI-directed paradigm.) The failure to find that one killer app for proving the efficacy of LLMs in the material world is making AI enthusiasts antsy - and making them sloppy.

Many were taken with the story of Claude running a lab and directing important biology experiments that discovered new enzymes. Surely the future has arrived! There are several problems here. The first is that Claude did not run the lab and did not direct the experiments. The reality seems to be pretty quotidian: ordinary scientists in an ordinary (albeit impressive) lab did some experiments and used Claude to analyze the results. And it doesn’t even appear that there was anything novel about that analysis; they just used Anthropic’s LLM to automate conventional analysis. That’s useful! I use LLMs to do regressions and make scatter plots etc. myself, where I once would have used R or JASP or similar. LLMs represent a consequential evolution in software, after all. But that’s not the same as an AI doing novel science, which is still in the realm of “plans to,” which is to say, speculation and extrapolation, where far too much AI hype lives. Let’s suppose, though, that the story was true as told, that Claude itself had decided on an experiment and ran it itself (whatever that means) and the outcome had been what it was - the discovery of a new but not-particularly-interesting enzyme. How meaningful is that, then? I would argue that it’s no more meaningful than if human scientists had discovered a novel enzyme that was of no particular interest other than being new. That is to say, you would never hear about it because it wouldn’t be worthy of reporting.

(I beg you to bear in mind the fact that everyone who depends on investors or grant funding has direct and major incentive to overstate the degree to which AI is aiding them in their work.)

When people search for a potential consequence of AI that would really move the needle, they often suggest that AI will cure cancer. This is misguided on all manner of reasons. Cancer, famously, isn’t one disease but rather hundreds of different diseases that share certain elements, each of which is driven by different mutations, in different tissues, and each individual cancer evolves inside each individual patient…. There won’t be one moment where cancer is cured any more than there’s ever going to be one cure for infection, writ large. Researchers in oncology tend to scoff at the notion that LLMs will cure cancer, even as they tend to acknowledge how the technology could be useful for analyzing results. After all, cancer research isn’t held back by a shortage of analysis or a failure to read all the literature, which is what LLMs are good at. It’s held back by what current AI models aren’t good at. LLMs can’t shorten the time it takes to see whether a tumor comes back, can’t make a drug less toxic by crunching more numbers, can’t get test subjects to adhere to a study’s protocols or stop them from dropping out. LLMs can help at the margins by sorting data, by suggesting potential molecules to try, by speeding up paperwork. But “AI will cure cancer” mistakes a problem that’s fundamentally a matter of stubborn tissues for a problem of intelligence, which misunderstands both the disease and the technology.

Here’s the bigger perspective: as miraculous as such a thing would be, as much of a human victory it would be, eliminating every cancer on Earth would only increase human life expectancy by about three years. I certainly need no lectures about how incredible it would be if humanity no longer had to face cancer. (Trust me.)