Pages

Saturday 6 April 2024

Alternative view of AI

 

3 comments:

Sam Vega said...

The Reddit scam looks to be particularly dodgy, and he's right about the declining quality in Google searches - I'd noticed that. There is a general tendency to think that "technology" improves in the abstract, rather than being fuelled by money that players will hoover up by fair means or foul.

Mind you, that Sasha robot is getting very realistic these days.

Sobers said...

My first thought when I read up on the whole AI thing as it exploded last year was 'Garbage in, Garbage out'. There's absolutely no way these things can be 'taught' manually, with every fact double checked before being presented to the AI model. They will just be data scraping the net and assuming that everything they find is true. Which we all know is nonsense. These things are actively making 'facts' up, and everyone is just writing this behaviour off as 'teething troubles', when in fact its proving the whole AI model as currently set up (which to throw stupendous amounts of questionable data at big enough computers, and assume when they regurgitate some of it that is 'intelligence') is fatally flawed.

I see the current AI based stock market tech boom as being a replay of the last tech boom, when all you had to do was have a website and suddenly your dog food shop was an internet giant, and worth billions. It wasn't, and reality reasserted itself eventually, as it will with this AI boom too. It may take longer this time, as the lure of the technology is strong, and it does work a lot of the time. The problem is going to be that it doesn't work ALL of the time, and thus anyone who implements it in their business is going to discover the hard way when it screws up. Imagine an AI doctor that gets 99.9% of drug prescriptions right, but occasionally throws out a fatally wrong one, or an AI accountant who gets your taxes right nearly all the time, but occasionally tells you about a completely made up tax law, and you commit tax fraud. And so on and so forth.

A K Haart said...

Sam - I tend to use Bing for searching these days. Not much better but the AI part makes it easier to be explicit. Simple searches are often crowded out by popular or commercial guff. I haven't seen it, but is that Sasha the film?

Sobers - that seems to be the issue, it works a lot of the time but doesn't work all of the time. Even if it is demonstrably better than any human, our expectations will rise towards perfect performance and acceptance could be hard going.

A major problem could be political. Anyone with access to genuinely powerful AI could in principle validate or invalidate any political narrative, decision, expert advice or statistical claim. It could be too democratic to be tolerated so maybe in one way or another it won't be tolerated.