Quote:
Originally Posted by andyh2000
There's a tendency in AI that as soon as computers become able to do something that previously only humans could do, it no longer counts as intelligence. Playing chess, recognising images etc. Sure, we don't have artificial general intelligence yet but tell someone 20 years ago that you'd be able to speak in natural language to a computer, asking it general knowledge questions, and it would provide useful answers; well, I think they'd have called that intelligence.
Andrew
|
No, the Chess and other Computer successes don't do it like we do. It's brute force examining many options and giant databases.
And the speaking and getting clever answers is a lie, depending on 1000s of cheap human workers.
Watson Health system was nothing to do with how Watson won Jeopardy. It was a failure.
We used to think brain size was important. Makes no difference to humans and Dolphins may be no "smarter" than dogs or horses. Rooks & many other crows might be "smarter" than chimps, dogs or dolphins.
Also many animals can learn a large vocabulary (with some crows and maybe even British starling as able to say words as grey parrots) and correctly connect words or symbols or sign language, but not one verified case of actual language. Some people are quite angry with Noam Chomsky's ideas on human language.
People too easily anthropomorphize animal behaviour or deterministic computer programs like chat bots. See Eliza, ALICE and all the current online ones. The more convincing ones are not much better than Eliza, just more rules and bigger database.
People misunderstand the Turing Test. Turing's suggestion wasn't that it would prove AI if passed, but that it should be possible and it would not indicate real or hard AI.
Today AI is a marketing term that's replaced Expert Systems. Computer Translation goal was once true grammar. Google abandoned that. They started with EU texts, the same things in different languages. Google Translate is sort of OK to your own language, but it's not much better than programs 20 years ago. It's a database with phrase and word matching. A giant Rosetta stone. Machine Learning is nothing of the sort. It's human curated data fed to the multi-level pattern matching database. Computer Neural Networks are a kind of data-flow pattern matching (not recognition) and are nothing to do with biological neural systems. Marketing lingo.
Both "The Diamond Age" and "The Ensorcelled Maid" examine the fake nature of AI today.