Imagine a person who confidently and condescendingly tells you that he knows the answer to every question, except that he constantly gets facts wrong and cannot reason his way out of a paper bag. Oh, and he is almost as unlikable as Ted Cruz. Is he J.D. Vance, or is he AI? (Admittedly, AI is not a "person," but are we truly sure about J.D.?)
Early last month, I had some fun writing a couple of columns about large-language models -- which have, quite inaccurately, come to be called Artificial Intelligence (AI) -- and how bad they are. On June 3 and June 9, with reference to my lone previous column mentioning AI from last Fall, I stipulated that my critique was not based on tech expertise but simply on the fact that AI has been shown again and again -- no matter how much it has supposedly improved over time -- to be all but useless.
I captured the fundamental problem in a quote from a podcaster that I used in both recent columns: "If you presented a calculator to me and said, 'This'll do any calculation you want, but it'll get a bunch wrong, and you don't know how often, and you don't know which ones will be wrong, so you have to manually check them all anyway,' isn't that useless?" And that certainly applies to J.D., who will confidently spit out words upon words that can in no way be taken at face value. Worse than AI, however, people would be wise simply to assume that what J.D. is saying is always wrong.
Indeed, he openly admitted -- or, more accurately, proudly announced -- that he will lie whenever he feels like it: "If I have to create stories so that the American media actually pays attention to the suffering of the American people, then that's what I’m going to do." This means that although AI is "dumb" in the sense that it naively accepts as fact anything that it scrapes from the internet, Vance deliberately steers away from the most reliable sources of facts and merely makes up things that he or the billionaires who created him would find useful.
Before going further with this comparison between the error-riddled automaton and AI, however, I should take a moment to offer a few relevant side notes.
First, I have complained in my previous columns that the "AI Overview" result at the top of Google searches cannot be turned off. Indeed, it is now apparently the plan to stop providing the results of old-fashioned searches entirely, although we will see about that. Even as it is, however, at best a person has to try to skip past the AI commentary at the top of the page without having one's attention caught by something possibly false, scrolling down to where the old-fashioned results begin. Happily, some tech types have come up with a workaround that only provides pre-AI search results. It is called "&udm=14 the disenshittification Konami code," and the link is here: https://udm14.com/ You're welcome.
Second, for those who would like to read a truly informative and insightful column explaining for laypeople (like me) why AI is not even supposed to be intelligent, the editors of The New York Times op-ed page pleasantly surprised me by having a very smart expert on its list of occasional columnists: Zeynep Tufekci, a Princeton sociologist whose "research revolves around politics, civics, movements, privacy and surveillance, as well as data and algorithms," according to her "about" page. In her most recent column (June 30, 2026), after offering a few amusing anecdotes involving super-silly AI errors, Tufekci lays it out clearly:
Large language models are not reasoning machines. They’re plausibility engines. It’s not just that they don’t test their outputs to make sure they’re correct or logical, or that they fail to do so in certain instances. They can’t, and they’ll never be able to on their own. They can only assess which answers are probable, based on the data on which the models have been trained. And that holds true whether they’re trained on the full breadth of human output or only on peer-reviewed scientific articles. It’s baked into the way they operate.
This helped me understand more precisely why the "improved" AI models never seem to improve. And Tufekci also makes it clear why the "data on which the models have been trained" is not the only inherent weakness of the technology.
On that data problem, I noted in my most recent column on this topic the now-infamous "MechaHitler" incident, which became the leading example of the fundamental garbage-in-garbage-out problem with AI, because that problem is fundamental to the internet itself.
I recently encountered this problem in a way that I had not seen before. I was visiting London, meaning that I was traveling on the Underground, which locals call the Tube. Because I am a hard rock-loving Baby Boomer, I cannot hear the words "the Tube" without thinking of this lyric from The Who's "Who Are You?" in which Roger Daltrey sings about a hungover morning after: "I took the tube back out of town, Back to the Rolling Pin." A few years ago, I wondered whether the Rolling Pin was a real bar and, if so, if it still exists somewhere in London. So I googled it.
On this most recent trip, I recounted to a loved one what I had learned when I researched that question online. It seems that although the guys in The Who came out of London, they (like most of the British Invasion bands of the 1960's) spent a lot of time in New York City. Although Daltrey (singing writer Pete Townshend's lyrics) used the words "the Tube," in fact the story happened in New York, with Daltrey returning to a local bar the morning after a bender. I thought that was interesting, and it was consistent with the way people use idioms in sometimes inaccurate ways.
The problem is that the story is apparently completely false, so when I wrote "in fact" in the paragraph above, I was wrongly repeating falsehoods that I had found on the internet. How do I now know that they are false? At this point, that is almost a metaphysical question, but I can at least note two things that I discovered after I had recounted my story to that admirably patient audience of one.
First, every possible search configuration that I used this time gave me the result that the story did happen in London. (That the opening lyric refers to "a Soho doorway" is genuinely ambiguous, however.) Second, not only is the current information inconsistent with what I found a few years ago, but the counter-intuitive story seems to have completely disappeared from the internet, as far as I can determine. To be clear, I am aware that one can have faulty memories or even be subject to the Mandela effect (or is it the Mengele effect?). My recollection of that story about the Rolling Pin, however, seems nearly impossible to explain with such phenomena. It is too specific and involved to be a hazy memory with a few facts wrong, and it is certainly not an example of a Mandela effect, because it is not shared.
In any case, I found this whole thing especially illuminating, because it is not merely about bad information being available online. It also shows that information can come and go. In fact, now that I am publishing this anecdote here, the internet again has at least one searchable source in which "Who are You?" supposedly takes place in New York.
Again, however, it is not simply that AI will always have to rely on outside-generated data that (per Tufecki) it cannot possibly assess for accuracy. AI is also very bad at using what it regards as facts correctly, and when it is correct, it is repetitive. In my June 3 column, for example, I reproduced some of the lyrics from an AI-generated song that was created in response to a friend's prompt involving some of the key locations in my recent nomadic existence. The AI-generated lyrics repeatedly used certain phrases to refer to various places: Washington DC was "a tailored suit" and "marble"; Florida was "heat" and "gators"; and the lyrics kept going back to Amsterdam's "canals." In my penultimate paragraph, I was in full sarcasm mode:
Maybe AI will get much better at all of this over time, but maybe not. [I]t could be that AI will continue to be simply bad. It is, at the very least, not obvious that additional programming could make it choose nontrivial -- to say nothing of accurate -- shorthands for various places. If you knew one of those annoying people who constantly referred to New York as "the city that never sleeps" and gave him "the Big Apple" and "the city so nice they named it twice," he could probably use those three options interchangeably. That would not, however, make him interesting.In some sense, however, my assessment was too kind, because the evidence to date suggests that the more likely path would be for an AI lyricist to refer to New York as "the Windy City" or "the Big Easy" as frequently as Gotham or "the city that never sleeps." AI is, as I suggested in today's headline, not only ridiculous but irredeemably ignorant and possibly dangerous.
All of which brings us back to J.D. vis-a-vis AI. Who likes AI? Some normal people do, I suppose, but at this point it is mostly down to a few creepy tech billionaires, because AI is useful to them. Who likes J.D.? Even fewer normal people appear to affirmatively like the VP, while everyone else seems to be completely turned off by him. But again, a few creepy tech billionaires think that he is useful to them.
This is because both are easily manipulable. AI sycophantically admits its errors and apologizes. J.D. changes his views without hesitation, and he says that his lies are for the greater good. He also, however, has an almost mystical capacity to make himself look ridiculous. Lecturing the Pope to "be careful" when speaking about theology is a self-own of biblical proportions (wink), and J.D. also says self-negating things out loud. As The Guardian recently reported, for example, Vance has admitted that he has no idea how diplomacy works:
On Friday Vance went on Fox where he expressed frustration with diplomatic protocols vis-a-vis Iran and said: "I’ve gotta be honest with you – I don’t really understand these things. [But] I’m trying to be respectful. Given my position in the last year-and-a-half, I have to care about diplomatic protocols all of a sudden."
Another important side note: Speaking of the Pope, his recent encyclical against AI was an example of moral leadership in its purest form. After what happened to Francis, we should all hope that Leo steers clear of Trump's Number Two.
But back to J.D. and Iran. This is the man who was supposed to be negotiating a diplomatic end to his boss's disastrous war of choice, and he said out loud on the TeeVee box that the whole diplomacy thing puzzles him. In addition, as Senator Chris Coons (D-Del.) reportedly put it at the time: "The easiest way to tell that even the White House knows this deal is a disaster? Instead of trumpeting it as his own deal, Trump is putting the eventual blame on" our hapless Vice President.
Back when Antonin Scalia was alive, I occasionally heard this from his defenders: "You might not agree with him, but ya gotta admit that he's a great thinker." That was never true (as I explained in a 2022 column), but at least it had a certain prima facie plausibility to it. In the end, he was mostly bombast and overweening, unearned self-regard. Vance has all of that, but without even the most basic case for presuming that there is any brilliance lurking under the surface. (And no, going to an elite law school does not prove anything. At all.)
In the end, J.D. is the worst version of what AI would be if it were (slightly) more human.
- Neil H. Buchanan