Another AI failure…
Wednesday, August 7th, 2024Social media is looking awfully over-the-top now as the 2024 presidential election nears. For example, Elon’s X (formerly known as Twitter) put out a Kamala Harris ad, still with her voice but with all the words changed. I don’t know if this was a scurrilous use of AI (more a scurrilous abuse by Elon?), but I immediately logged on to my account and deactivated it. (Apologies to my loyal 1000+ faithful followers there who might have been surprised by my actions, but they know as well as I do that X has been going downhill ever since Elon walked in with his kitchen sink—not the gold one, of course—and I’d diminished my activity on X correspondingly ever since he took over. I closed my Facebook account even earlier!)
But AI (even the current primitive forms—see my previous post) changes things, doesn’t it? AI takes bad social media to worse levels. As if that “blue screen of death” seen around the world wasn’t enough—that was on CrowdStrike and Microsoft, by the way, not AI. I wasn’t on the internet at the time, and I didn’t return to it until I backed up my laptop and the “all clear” was given. (What occurred with Delta Airlines seems to be an additional problem…caused by their ancient and sickly software?)
I’m not talking about either X or CrowdStrike/Microsoft here. The first is out of my life (so is Elon Musk); the second can be avoided by minimizing my time on the internet (I already do that!). No, I’d like to discuss a report in The NY Times recently that current AIs are bad mathematicians, as if they were students good in the humanities but terrible at math. No surprise, really, although none of this matters much for authors and readers. (Or, better said, it makes it worse for both equally?)
Unlike most fiction writers, I’ve probably forgotten more math than they ever knew or will know. That’s not bragging. It’s just offered as evidence that I can well believe that The Times is partially correct. The type of recent AI that’s been in development has severe limitations; they mostly stem from the fact that current AI versions are just glorified search engines. Sure, the better ones can create stuff by putting what they find in a search (or what’s fed them) together in a more coherent whole. That’s the artificial part. But there’s little actual intelligence, as in human intelligence.
That’s where the math problems originate: How it puts things together is based on probability, more specifically, a calculation of likelihood that its creation is a correct portrait of reality. You’ve seen that with facial recognition software: Some unidentified face is 69% more likely to be some John Doe than someone else. The rest of math, most of it in fact, lies outside the sphere of probability and statistics, and that’s an exact science, not about likelihoods: Normally, an answer is either mathematically correct, or it’s not. Consequently, current AI can’t do math because it can’t handle exactness; everything’s a maybe, not a solid yes or no. Any AI guru who claims otherwise is a con man. (Or woman.)
But maybe superior AIs in the future can do better? More data too might mean better predictions. Sci-fi writers should have an AI give an answer like the following: It’s probable that X event occurs with a probability of 0.99999 with an error of plus or minus 0.00000342. That might not be exact enough to control a starship slipping through multiple quantum states of the Universe, but for general purposes, it’s better than a paltry 0.69. (Actually, likelihoods differ from probabilities, but you get the idea, right? And the statistics being used to make the predictions can give different results too—Bayesian stats are the most common.)
“What about quantum computing?” you might say. Yep, that could change everything. I’ll let you know after I’m able to purchase my first quantum laptop!
***
Comments are always welcome. (Please follow the rules listed on the “Join the Conversation” web page.)
“The Chaos Chronicles Trilogy Collection.” Futuristic AI plays an important supporting role all through the three novels found in this bargain sci-fi/thriller bundle. The reader can follow human beings flight from a dystopian Earth to first contact on a planet in the 82 Eridani system; an intergalactic war with different and xenophobic ETs; and battles with a deranged human, a psychotic sociopath out to control the galaxy. All the AIs here are artificial (even if partly constructed from living circuitry) and extremely intelligent, unlike current AI software. And man, they can they do math! This three-novel ebook bundle is available wherever quality sci-fi is sold.
Around the world and to the stars! In libris libertas!