Artificial Intelligence—The End of the Life As We Know It? Part one

I will begin by answering the question: There is no way we can know. Artificial intelligence (AI) is neither good nor evil. It is an advanced computer method for analyzing streams of data and looking for patterns, then using those patterns to fill in gaps in our knowledge. Computers have been doing this for decades, and humans have been doing it for centuries with pen and paper and abacus. The difference is that we are surrendering our decision-making prerogatives to AI.

It would be very useful to have, if we do not already, an AI program to detect unusual patterns in our cells that may signal the onset of cancer, far earlier than a human observer could detect it. Then the program could alert a doctor, who would talk with the patient. This could save a life. But AI programs do more than this. They decide if the patient will, in fact, get cancer. They then alert insurance companies, which will then deny insurance coverage even if the cancer is a decade or more in the future. Insurance companies do not like to cover “pre-existing conditions” (in France, they have no choice), but what about future conditions? Too bad, buddy.

It is also useful to have AI programs that can detect possible clusters of terrorists or criminals. Law enforcement or military personnel could then take precautions. But what if a “terrorist cell” turns out to be a wedding in Gaza? Too bad, buddy. You get bombed. I wonder how many billions of dollars have been spent bombing what a human observer could easily have seen was the wrong place. We end up killing thousands of innocent people and creating enemies around the world.

Or an AI program identifies you as a criminal, and you get arrested. Eventually, you might be exonerated, but probably not for a long time, since, law enforcement assumes, computers are never wrong.

An AI program might identify the ideal investment for you to make with your huge stockpile of money. It could go ahead and invest it for you, since the opportunity might be lost otherwise.

In all of these cases, the human prerogative of making the decision was surrendered to the AI program itself.

An AI program can be racist, for example, if it is designed to be, or even allowed to be. Or it can use any metric of human appearance as a basis for discrimination. AI is designed to not let you know the basis of its decisions, because that basis will be a mathematical algorithm that would seem like gibberish to almost everyone. Facial recognition programs, for example, have a higher rate of misidentification among people of color than among whites. This is a problem that can be corrected, if someone wants to. You can bet that if Nazis had had AI, they would have built racism into it.

There is a fundamental delusion here, as explained by Yuval Noah Harari in his book Nexus: A Brief History of Information Networks from the Stone Age to AI. That delusion is: The more information you have, the closer you will come to understanding the truth. The problem is that much or most of the information that you have is going to be wrong, leading you in a random direction or even the wrong one. AI analyzes the information, and leads you, at best, in a random direction.

Harari’s books, this one and Sapiens and Homo Deus, are widely recognized as being the wisest and deepest books on human history. If you only read three books about human history, it should be his three books.

We all have our own AI horror stories. Mine is, fortunately, more funny than tragic. I publish books online through Kindle. One day, without warning, Kindle Publishing blocked one of my uploaded books. An AI program made the decision, a decision that was never explained to me. I tried solving the problem, and my account reappeared and disappeared over and over. Finally a human being (an underpaid staffer in India) got involved and reinstated my account and all my books. I never did find out what the problem was. Only the AI software knows. Maybe it scanned the cover image of my grandfather and decided that it was, in fact, a photo of Kim Kardashian. Is there a close resemblance? You be the judge.

Never heard of Yuval Noah Harari? His own book will explain why. The human brain is not designed by evolution to pay much attention to reams of data, no matter how important they may be. It is also not designed to pay much attention to big, overarching ideas, which is what he writes about. The human brain, evolving over millions of years in tribes, cares almost only about specific stories. We are good at remembering even complex story lines so long as they do not have too many complex ideas or too many characters. At most, the average person can remember just a part of the history of his or her own family or country. That is certainly the case with me; I know a lot about Cherokee history, especially my own family, but other aspects of history? Not so much.

If you have a racist AI program that picks out books for you to read, perhaps it would identify Yuval Noah Harari as African and that, therefore, he is biased or even stupid. Actually, he is Israeli. But by the time you have figured this out, you have already passed up reading his books.

The End of Creativity

Aldous Huxley wrote a short novel in which he saw a pickup truck speeding out of a Hollywood movie studio, overloaded with unsolicited screenplay manuscripts. The truck veered, and one of the manuscripts fell off. This didn’t really happen, of course. But, almost a hundred years ago, one of history’s great writers complained about how people in power (in this case, movie producers, but this would also include editors and literary agents) would barely if ever look at what they derisively called their “slush pile” of submissions.

Today, little has changed. The term “slush pile” is still standard. I have written a lot, and published very little, fiction. One of my stories ended up in a magazine that was printed and spiral bound at a copy shop. You cannot submit fiction directly to an editor; and very few literary agents, I suspect, even look at the submissions they receive. Out of hundreds of agents to whom I have submitted samples, only about three have responded personally. The others, I suspect, just have unpaid assistants write rejections or, if the submission is online, the software automatically sends out a rejection. I suspect that few of my submissions have even been looked at.

I could say all kinds of bad things about fiction agents. But the purpose of this essay is to defend just one aspect of their refusal to look at submissions. And that is, AI software can write fiction and submit it. How is an agent to know whether a real person wrote the novel? A robot can click on the “I am not a robot” button, I assume.

And they can write them really, really fast. Amazon self-publishing recently had to set a limit on the number of submissions a single source, such as an email, can make: three per hour. One can easily see that an AI program could write ten times that many novels, and keep doing it all night without coffee and without amphetamines. A single computer could write more novels than the whole world’s collection of novels before, say, 1990.

By definition, these novels are formulaic. They follow formulas. That’s what computers do. The resulting novels are very unlikely to have any deep thoughts, and the plot lines will almost certainly be lame. No agent or editor would call them good. By slightly altering the parameters, the AI programmer can produce a great number of very different novels, all bad.

But that doesn’t matter. Most readers cannot tell the difference between good fiction and bad. Maybe a hundred years ago a reader could distinguish them. And today, some readers still can. This probably includes you, intelligent reader. But people who can tell good novels from bad are not as big of a market as those who just want something to distract them. A publisher can pay almost nothing for an AI generated manuscript, and sell a lot of copies. Each such title would be immensely profitable. If good writers complained that a certain major publisher sold books that a computer may have written, a lot of readers would just tell us to stop griping and get with the modern world.

An important reason for this is that, during the twentieth century, the rules of good literature were jettisoned. Plots no longer had to make any sense, especially if they included dream sequences. Fantasy literature is particularly vulnerable to having plots that make no sense; the writer could just change the laws of nature when he or she wanted to. There needs to be no character development or beautiful description. Certainly no meaning-of-life stuff. Poetry is even worse, which is why most poets only read poems by other poets whom they know. Real human writers turned fiction into something that a computer would eventually be able to write. That time has come.

I have spent many hours sending things to fiction agents. Each one has a slightly different set of rules, and they will not read any submission that does not follow these rules. Should you include a summary, or not? A sample of the writing, or not, and if so how much? I’m not sure that it matters, because the submission will almost certainly not be read. Some agents include a list of rules, the last of which is always, I will look at the submission if it looks like something I would like. I doubt that, in the contentment of retirement, I will ever do this again. And the main reason is that neither agents nor publishers could possibly find a good, human novel in the mass of fake AI manuscripts.

I have been moderately successful at publishing nonfiction in those subjects about which I am an expert, even if not the top expert in the world: botany, evolution, ecology, scientific thinking. Check out my books at stanleyrice.com. A computer could write nonfiction, but it would quickly get recognized as fake, because of the unlimited number of errors it would contain. Publishers might even have legal liability if readers followed stupid advice from a random manuscript. For nonfiction, agents and editors want not just a good book but evidence of expertise. This doesn’t prevent fake nonfiction from being published. There are whole “scientific” journals that will publish anything even if it makes no sense at all and since the “journals” are online it costs almost nothing to publish them. Professors have lost their jobs from claiming fake papers as their publications. But it is harder to write fake nonfiction than fake fiction.

Writers in the movie industry have gone on strike over AI taking over their jobs. Most moviegoers would not know the difference between a human writer and a computer. But the stakes are higher, since in a movie you have immense production costs aside from the writing. If moviegoers recognized a movie as “a real stinker,” the producer would quickly be out of business. Maybe.

So my plan is to stop submitting any fiction or poetry. Each submission would go into a pile of mostly computer-generated submissions, potentially numbering in the billions of billions. A needle in a haystack, or a snowflake in hell, would stand a better chance. I will have them printed up for future generations of my family. Someday, in an underground vault, they might get found. And my poems, too. Sorry, Randy, I know you are a good poet but who else can know it? I notice that neither one of us lives off of our fiction or poetry income.

A similar thing is happening in music. A good composer can still outdo a computer. Even when a Huawei program finished Schubert’s unfinished symphony, it required a little human help from composer Lucas Cantor. I listened to it. It simply did not sound like Schubert. But it was competently written and scored. Computers can provide all kinds of sounds that no musical instrument can produce, but composers still need to tell the computers what to do.

But maybe not for long. During much of the twentieth century, composers on university faculties prided themselves on writing music that was distinctly unpleasant to listen to. Their philosophy seemed to be that you are not supposed to actually enjoy music; it is supposed to be a psychological experience, and if you do not connect with the composer’s music, there is something wrong with you. This was the dominant philosophy during the years I took music courses at the university. I heard from two music graduate students about this. I was considering sitting in on a composer’s forum. A graduate student asked me, Do you use notes? I said of course I used notes. Then don’t bother with the forum, she said. Another graduate student told me that the “new music” of the twentieth century created only one emotion in her: terror.

Within a couple of decades, this approach to music was starting to die away. Perhaps they did not know it, but these out-of-touch-with-musical-pleasure composers were sowing the seeds of their own destruction. If a computer could write music that is just as good as the self-proclaimed leaders of musical innovation, then someday they will.

A lot of careers are being replaced by AI. Even some physicians I have consulted stopped to look up my symptoms on WebMD. I got lots of different diagnoses. Some of them would have been funny had they not been intended as serious

Robots have been making cars for decades.

And now they are writing novels. These days, poor Aldous Huxley would have been left mouldering in the dust.