Re: Google engineer goes crazy, thinks AI is realby Rafiki 06/13/2022, 8:49pm PDT
blackwater wrote:
I've been hearing people debate AI for decades and it's old news at this point. But, every few years, some new AI milestone gets crossed. For example, computers defeated the human chess champion. Computers defeated the human Go champion. Computers can now do a pretty good job of recognizing objects. Computers can now do a pretty good job of driving.
It sure does seem like after a few more tactical retreats, the anti-AI guys are going to be pushed into the ocean.
No one made any tactical retreats because none of those are milestones in artificial intelligence, that's just media hype. The first 2 are probability and statistics problems (see: most AI should be Applied Statistics) solved by the engineering milestones of creating hardware powerful enough to calculate the probability of winning by cycling through all possible outcomes a billion moves in advance in a timeframe short enough for a practical game. The second one is also a statistics problem using pattern matching against a giant image database and is still pretty limited. And the 3rd is the closest to "AI," but is fairly unremarkable from a programming standpoint because we've had self-driving cars since racing games first added "CPU" as an opponent. Grand Theft Auto has been simulating entire cities full of cars obeying traffic laws and responding to arbitrary obstacles and events for 20 years. Real world self-driving cars are, once again, an engineering milestone of having hardware fast, sensitive, and, most importantly, cheap enough to integrate into consumer cars to feed data into the programs people already knew how to write.
I already gave the precise moment when AI can be achieved: when someone defines what intelligence is. Programmers can't code a simulation of something nobody can describe, but for some reason people can't just admit this to themselves so we get magical thinking setting arbitrary goals. When a computer can win a game! When it can match shapes and colors! When it can generate fat chicks in party hats! Somehow, by accident, intelligence and self-awareness are going to arise out of this. Humanity will flub its way into AI. Bullshit.
There's a famous quote about AI I can't find that said something to the effect of, "If we can't tell the difference between talking to a computer and talking to a person, is there one?" It's got this kind of lofty, philosophical air to it, but it's a nonsense question with a concrete answer.
Here's a dog trying to play fetch with a statue because it doesn't understand that the statue isn't real. It doesn't matter if you examine the situation from the dog's perspective, or "paradigm," or however you want to move the goalposts. The statue isn't real. My point is, just because we may be able to trick ourselves into believing something stupid doesn't make the stupid thing real. The fat man from Google believing in LAMDA or idiots from the 60s believing in ELIZA doesn't make them real AI. It doesn't matter how realistic a conversation looks or how good your $40 million neural network is at hotdog / not hotdog, if no one on earth is able to define how to be intelligent, then no one can code it, so we haven't achieved AI yet.