Another Angle On "Artificial Intelligence"

     Investing is essentially a more-or-less honest confidence game: you convince investors your enterprise will succeed while not guaranteeing it and show 'em (a subset of) the books.  It either succeeds or fails and the investors either get their money back, ideally with some kind of profit or they don't.  But the investors have to convince themselves it's worth the risk and the firm they invest in has to do the work, not take the money and run.

     Tech firms are usually selling smoke, mirrors and a cunning plan.  That's what all forms of artificial intelligence boil to: a cunning plan and some impressive-looking hardware.  Good AI, bad AI -- either way, it's a massive server farm, building algorithms on algorithms to crunch through massive amounts of data in the hope of emergent patterns.  Chatbots are very, very sophisticated, context-dependent word-prediction machines; imagebots do the same thing with pictures.  The process between input data plus user prompts and output is opaque, and it was slow going for a long, long time.  (How long?  Marvin Minsky never did get a robot to build an analog Heathkit TV set, despite the very clear instructions that come with the kit.  Heathkit as a major kit company has been gone since 1992, analog TV since 2009, Dr. Minsky since 2016.  He bought that TV kit before I graduated high school, back when Gerald Ford was in the White House.)

     Then in 2022 (as recently as that!), a Google AI engineer convinced himself that a chatbot called LaMDA was self-aware.  He tried to hire a lawyer for it and went to the press.  Google fired him: that work was a trade secret.

     But mark what happened carefully: a person who interacted with AI in depth convinced himself it was alive, and that was the dawn of our current AI boom.  It didn't build a Heathkit,* or take over the world,† or help mastermind a political revolution.‡  It managed to seem real enough for someone to believe there was a ghost in the machine.  It got his confidence.

     "AI" is a con game, and it gets better at the con every day.  Does it get better at "intelligence?"  Probably not, but it certainly gets better at convincing people, especially those with money and a will to believe, that it is either intelligent, sentient, or on the verge of one or the other or both.  And it may not be an "honest con" in the way investing in general is: there may be zero chance of the actual payoff, or as close to zero as makes no difference: there will probably never be a ghost in the machine.  And that matters.

     The bottom is liable to fall out once it hits its peak, in something akin to the dotcom boom and crash, leaving fortunes for the promoters and ashes for the investors.
____________________
* Minsky, op. cit.
 
† Colossus: The Forbin Project, both the book and the film.
 
‡ The Moon Is A Harsh Mistress, Heinlein, Robert A.