Old MacBezos Had a Server Farm…
Free-associating there, a little. Pardon me.
Seems AI is on a lot of people’s minds these days. I, along with many, have my doubts:
My opinion: there are a lot of physical processes well suited to the very fancy automation that today is called AI. Such AI could put most underwriters, investment analysts, and hardware designers out of a job, like telegraph agents and buggy whip makers before them. I also think there’s an awful lot of the ‘we’re almost there!’ noise surrounding AI that has surrounded commercial nuclear fusion for my entire life – it’s always just around the corner, it’s always just a few technical details that need working out.
But it’s still not here. Both commercial nuclear fusion and AI, in the manner I am talking about, may come, and may even come soon. But I’m not holding my breath.
And this is not the sort of strong AI – you know, the Commander Data kind of AI – that gets human rights for robots discussions going. For philosophical reasons, I have my doubts human beings can create intellect (other than in the old fashioned baby-making way), no matter how much emergent properties handwavium is applied. Onward:
Here is the esteemed William Briggs, Statistician to the Stars, taking a shot at the “burgeoning digital afterlife industry”. Some geniuses have decided to one-up the standard Las Vegas psychic lounge routine, where by a combination of research (“hot readings”) and clever dialogue (“cold readings”), a performer can give the gullible the impression he is a mind reader, by training computers to do it.
Hot readings are cheating. Cons peek in wallets, purses, and now on the Internet, and note relevant facts, such as addresses, birthdays, and various other bits of personal information. Cold readings are when the con probes the mark, trying many different lines of inquiry—“I see the letter ‘M’”—which rely on the mark providing relevant feedback. “I had a pet duck when I was four named Missy?” “That’s it! Missy misses you from Duck Heaven.” “You can see!”
You might not believe it, but cold reading is shockingly effective. I have used it many times in practicing mentalism (mental magic), all under the guise of “scientific psychological theory.” People want to believe in psychics, and they want to believe in science maybe even more.
Briggs notes that this is a form of the Turing Test, and points to a wonderful 1990 interview of Mortimer Adler by William F. Buckley, wherein they discuss the notions of intellect,. brain, and human thought. Well worth the 10 minutes to watch.
In Machine Learning Disability, esteemed writer and theologian Brian Niemeier recounts, first, a story much like I reference in my tweet pasted in above: how a algorithm trained to do one thing – identify hit songs across many media in near real time – generates an hilarious false positive when an old pirated and memed clip goes viral.
Then it gets all serious. All this Big Data science you’ve been hearing of, and upon which the Google, Facebook and Amazon fortunes are built, is very, very iffy, no better than the Billboard algorithms that generated the false positive. Less obvious are people now using Big Data science to prove all sorts of things. In my gimlet-eyed take, doing research on giant datasets is a great way to bury your assumptions and biases so that they’re very hard to find. This, on top of the errors built in to the sampling, the methodology and algorithms themselves – errors upon errors upon errors.
As Niemeier points out, just having huge amounts of data is no guarantee you are doing good science, in in fact multiplies to opportunity to get it wrong. Briggs points out in his essay how easily people are fooled, and how doggedly they’ll stick to their beliefs even in the face of contrary evidence. You put these things together, and it’s pretty scary out there.
I’m always amazed that people who have worked around computers fall for any of this. Every geek with a shred of self-awareness (not a given by any means) has multiple stories about programs and hardware doing stupid things, how no one could have possibly imagined a user doing X, and so (best case) X crashes the system or (worse case) X propagates and goes unnoticed for years until the error is subtle, ingrained and permanent. Depending on the error, this could be bad. Big Data is a perfect environment for this latter result.
John C. Wright also gets in on the AI kerfuffle, referencing the Briggs post and adding his own inimitable comments.
Finally, Dust, a Youtube channel featuring science fiction short films, recently had an “AI Week” where the shorts were all based on AI themes. One film took a machine learning tool, fed it a bunch of Sci Fi classics and not so classics, and had it write a script, following the procedure used by short film competitions. And then shot the film. The results are always painful, but occasionally painfully funny. The actors should get Oscar nominations in the new Lucas Memorial Best Straight Faces When Saying Really Stupid Dialogue category: