How Smart Are Those Machines, Really?

Currently there is fear and trembling about the alleged possibility that machines can attain humans’ alleged intelligence. Not much of a feat, you may think. My own idea is that they are not and will not be able to do so. They can only pile up words that have been injected into them; they cannot make value judgments. Yes, they can find that “democracy” appears 99 billion times in the holdings of the University of Michigan library, and they can discover that in 99% of those instances it’s regarded as a good, better, best thing for “our society” to have. They can write “essays” substantiating that view, with many references to books and quotations of arguments. But they cannot decide which of these arguments, if any, are actually true, in what sense they are true, and to what degree, if any, their trueness is important. (They may do better with deciding whether a drawing of a cat is truly a drawing of a cat.)

Let me put this in another way. Suppose I’m speaking before an audience, and I suddenly have a heart attack, and a doctor is summoned to figure out whether I’m dying. She can do that in a quantitative way. She can count my pulse, determine my temperature, and so on. She can pronounce me dying, still dying, and finally dead, but she cannot with quantitative methods answer the question of whether this is a good thing or not. To do that, she would have to use qualitative methods. She would have to determine for herself what is good, better, best. She would have to cope with the possibility that many people in the audience might think, for various and obvious reasons, that my death is a good thing, but that the one person who thinks otherwise might just be right. A machine would be able to replace some of her doctor functions, but it would not be able to replace her philosopher function. It might be able to fake the issue, by finding in its capacious memory such cliches as “any man’s death diminishes me,” or “a good professor is a dead professor”; it might fill 10 or 10,000 pages with summaries of arguments; but that wouldn’t be “thinking.”

If exposition of the law is just citation of cases and reproduction of formulaic arguments, and most readers can’t tell the difference, then why get outraged about a fake?

This is my position, and I’m sticking to it. Lately I’ve been both amused and disgusted by the reported ability of a computer to fake a legal brief, complete with references to phony cases, and then to act like a real lawyer and maintain its total innocence. These accomplishments, however, do not demonstrate the legal or literary intelligence of the computer; they demonstrate the lack of intelligence often displayed by people who participate in that genre of writing. If exposition of the law is just citation of cases and reproduction of formulaic arguments, and most readers can’t tell the difference, then why get outraged about a fake? I’m sure that “work” in my own profession, literary history and criticism, would pass for human just as easily. A look at the literature shelves in the library strongly indicates that mechanical reproduction has been going on for at least 50 years.

Back then, word salads were more likely to be served in the dingy cafeterias of the “learned” professions than under the circus tent of politics. It used to be that labor leaders howled their demands, and politicians “lashed out” in fearful bellowings. The sounds were human, all too human. The old school may now be perishing in its final eruption, the recent communications of Donald Trump. People wonder what he means when he excoriates Ron DeSantis for being worse than the Democrats about COVID, being a bully about Disney, being “sanctimonious,” etc. You can try all you want to figure out what he means. To me it’s like wondering what Godzilla meant when he ROARRERRD and smashed another building.

But this is now passé. The well-named salad is now the dish of choice. I don’t need to tell you what it is. You know it when you hear it. It is, for example, the universal language of pressure-group pronouncements, in which all the vegetables from “LGBTQ+ rights” to “healthcare rights of families” to “rights to self-determination of indigenous peoples” are mixed together, in the hope that every diner will find something to fork out and swallow. Or, to be fair, it’s the strange mixture you get at the bar and grill on the other side of the political divide, where they can toss you up almost any kind of mélange, with ingredients varying from “America first” to “we must project American power in the world.”

Biden has oodles more viruses, and they’re really nasty.

But the reigning salad chefs, right now, are in business on the left side of the street. One is Karine Jean-Pierre, America’s sweetheart, whose performances in the White House briefing room are tests of even liberal reporters’ powers of endurance. Her sayings could certainly be generated by a machine. I like to think of her as one. Not only does it fit the empirical evidence, but there’s already been a movie made about beings like her. It’s called Westworld.

Jean-Pierre’s boss, Joseph Robinette Biden, Jr., is virtual (pun intended) proof that the White House has become a Westworld set. You can tell by the type of errors he makes. Once a virus gets stuck in his program, it’s there for good. Jean-Pierre would be lost without “let me be clear,” “this president has always,” and “for that I refer you to [insert name of acronymic government agency, which will never reply].” But Biden has oodles more viruses, and they’re really nasty. They’re the ones that generate “come on, man,” “mega maga Republicans,” “the real deal,” “true story!”, “not a joke,” “I’m not kidding,” “no, I really mean it,” “guess what?”, “I shouldn’t say this, I’ll get in trouble,” and other absurdities that his software keeps running in an endless loop.

More interesting are certain sounds he emits that appear to come from a cheapo version of AI that’s programmed to match words from category A to words from category B, without any means of checking for matches with category C, the “Ridiculous-Untrue-Avoid” list. So, if category A is “bad things that happened to my son Beau,” it’s programmed to match with category B, “very bad places for things to happen,” and bingo! The selection is made, and we get, “My son Beau, who died in Iraq.” Or, category A is “my previous occupations,” and category B is “gnarly plebeian stuff,” and we get, “I used to drive an 18-wheeler.” But maybe category A is “good things about me,” which can be matched with something from category B, “accomplishments that have marked my presidency.” The protocol will not allow any choices of N-words: “nothing, null, nada, N/A, non-starter.” But it will allow reversals of bad words, such as “inflation,” “deficit,” and “illegal border crossings.” So we get, “I lowered inflation,” “I lowered illegal border crossings,” and “I lowered the deficit more than any other president in history.” (In every third iteration, the program adds “in history.”)

This is not a very good program, but it does create flights of imagination that some people regard as evidence of humanity. Consider what happened on June 8, when Thinking Machine JRB-JR held a press conference with the British prime minister. Cruising its data banks, the computer lit on “railroads” as a subject that’s safe to talk about, and “building” as a good, optimistic, “positive” match. But where should we build a railroad? Where is Joe Biden actually building a railroad? This is where the program got sophisticated. It checked a list of places where railroads already exist, found that they’re almost everywhere, and proceeded to option 2, which is a list of places where railroads have never existed. It came up with this: “We’re talking about building — and I had my team putting together with other countries as well — to build a railroad from the Pacific Ocean — from the Atlantic Ocean all the way to the Indian Ocean.”

Right. Except for that little temporary problem about “Pacific” and the flaw in the data that rendered all of southern Africa as railroad-needy . . . mission accomplished.

This is not a very good program, but it does create flights of imagination that some people regard as evidence of humanity.

But it wasn’t the last time when things went all geographic like that. You know how computers are — once they get something into their heads . . . if they try to fix something, they just make it worse. Or haven’t you ever pressed that button after “Do you want Windows to fix this problem?” So here’s JRB-JR one week later, on June 14, performing for a conclave of wealthy “environmental” organizations and announcing: “We have plans to build a railroad from the Pacific all the way across the Indian Ocean.”

Imaginative? Yes. Proof of intelligence? I have grave doubts.

But come on; we should all have fun with this stuff. Last week I found something that’s lotsa fun. It’s about one of my favorite subjects, the Titanic, and it has nothing to do with either Biden or the Titanic submersible disaster.

But before I tell you — Let’s have a word from the sponsor. I want you to know that about 20 years ago I wrote a book called The Titanic Story, in which I tried, among other things, to expose some of the persistent Titanic myths and fallacies. You can get the book from Amazon, cheap! Go and do it. A few years later, I wrote an article for Critical Review (vol. 15, 2003, pages 403–434), discussing the ways in which myths were created about the disaster. If you google that item, you can find a number of places where you can get it online. Please go and do that, too.

Perhaps we shouldn’t give machines too much of the blame. After all, they, like Joe Biden, are just trying to prove they’re smart (and with similar results).

Thank you. To continue — After writing those things, I got tired of trying to identify untruths about the Titanic; there were too many and they were too boring. But what I found last week was a marvelously amusing discussion of the kind of word salads that we’re now being served whenever we open a book or look at a screen. It’s by a guy named Mike Brady, who runs a YouTube channel called “Oceanliner Designs,” and it’s a review of a purported history of Titanic. See if you don’t think this history could have been written by “many men, many women, and many children,” as Dr. Johnson said about a poem he didn’t like — or, I might add, by many machines.

But perhaps we shouldn’t give machines too much of the blame. After all, they, like Joe Biden, are just trying to prove they’re smart (and with similar results). And another explanation for verbal garbage has emerged.

A “marketing agency” named Captiv8 (“captivate” — how clever) is reported to have been the PFC that was told by the lieutenant, a vice president of Anheuser-Busch, to fulfill an alleged “super clear mandate” from the top brass, which was to modernize Bud Light’s “brand” — which meant, in practice, introducing obnoxious “social media influencer” Dylan Mulvaney to promote the cheap, innocuous beer. That decision, as you know, resulted in an enormous backlash. Bud sales cratered, going down by one-quarter and staying there, with stocks sustaining billions in damages. Now, Captiv8, which had been lying low, is being discussed, and one of its claims has been noted — that it has more than 30 million internet “creators” in its “farm system.” At the same time, the company appears to have claimed that it has “more than 1 million influencers on platforms such as YouTube, TikTok, Instagram and Twitter.”

Well, never mind the difference in the numbers. Even a million “influencers” could sling a lot of trash onto the salad bar. Whether the influencers are actually AI or not.