There is clear empirical reason for concern. In narrow ai we saw in 2016 a dramatic step function when a simple new deep learning algorithm describable with literally 300 lines of workable python code made Go game AI jump from mouse level play to g-d level play. If that happens again in the more general AI of LLM then we could be in la-la land, and we know for sure scientist are working to bring the ideas from the alphazaro go algorithm into the field of multi modal language models (aka chatgpt)
@_yak
4 ай бұрын
The inability of so many of our thought leaders to seriously engage with AI risk is what makes me the most pessimistic. Kevin points out that Einstein would lose against a tiger, without realizing that that could have been just as easily pointed out by a hardcore AI doomer. A human would lose a fight with a tiger, and yet humans dominate this planet, not tigers. Why? Intelligence. A feeble naked ape, designed by evolution to be good at surviving long enough to reproduce, has walked on the moon. Intelligence is a powerful force, and we don't actually know the limits of its power.
@mattd8725
4 ай бұрын
There is some problem with the assumptions in your reasoning. The first is that spreading everywhere is because of intelligence, or that intelligence is required for it. It is obviously true that many things that don't appear intelligent are everywhere, such as bacteria, dust or advertisements. Then is "spreading everywhere" a goal which is basically intelligent? Not really, expanding populations until the point of collapse does not seem to be proven to correlate to intelligence. We don't know the limits of intelligence? Perhaps not all limits, but we know that intelligence alone can't allow us to live on the surface of Mars with no significant life support. It doesn't really improve our odds of being ambushed by a tiger. Intelligence tells us to avoid tigers and not try to drive a poorly prepared VW camper van to Mars.
@_yak
4 ай бұрын
@@mattd8725 Pretty much every time you will see a tiger, you'll be safely outside of its cage at the zoo. Intelligence did that. It's not the only powerful strategy that evolution stumbled upon, but it is powerful. And no, we don't know its limits. It won't allow us to live on Mars without life support, but it might indeed allow us to live on Mars, or even spread out farther. With significantly greater intelligence, what more can be done? But here's my basic point: this thing about humans being weaker than tigers but still being the dominant species is an idea that you can hear AI safety people make. It seems like Kevin isn't aware of that. Even Bob makes points that betray a disengagement with that conversation. I simply don't know if the AI safety people are correct or not, but many of the ideas they have don't sound crazy to me and I wish more of my favorite public intellectuals (like Bob) would more seriously engage with them.
@dsuleyma
4 ай бұрын
Please get on Dr John Vervaeke and talk about Intelligence and AI. Would be very interesting.
@Anders01
4 ай бұрын
Interesting observation Kevin made about how the progress of the AI capacity itself may not be exponential (accelerating). I still believe that AI, robotics and automation will change the economy within years! Not in decades. For example I wouldn't be surprised if Chinese tech companies release armies of humanoid workforce robots next year connected to AI clouds.
@bn16fan
4 ай бұрын
The Master!
@gingerhipster
4 ай бұрын
"I think this is going to be decades"
@gingerhipster
4 ай бұрын
There isn't a regulatory force that can stop this.
@charlesalexanderable
4 ай бұрын
20:12 1-2% a year is still exponential
@human_shaped
4 ай бұрын
Hehe. Yes, good catch.
@ryandury
4 ай бұрын
How so? growth can be linear. It's not inherently exponential.
@ryandury
4 ай бұрын
2% year over year as linear growth turns $100 into $300 over 100 years. Exponential growth turns it into much more
@human_shaped
4 ай бұрын
@@ryandury Someone forgot to math. A repeated percentage increase is literally the definition of exponential.
@rey82rey82
4 ай бұрын
@@ryandurycompound interest
@aroemaliuged4776
4 ай бұрын
When you are child and grow into an adult with no ideological or arcane beliefs Everything is natural Doesn’t mean it is true or false But it does make you skeptical
@rey82rey82
4 ай бұрын
Embrace maximum risk
@rossw1365
4 ай бұрын
"ai will take control to better do a job" is just the paperclip argument stated more generally and like paperclip ai, it suffers from the same poor logic if ai is powerful enough to "take control", then it is likey also smart enough to know taking control is not a good idea and if ai doesn't know taking control is not a good idea, then it is prob not powerful enough to do it ie, ability and intelligence scale together and ais, like paperclip, that are able but not smart enough to know are a contradiction
@jakeinfactsaid8637
4 ай бұрын
This is a pretty bad conversation so far. i’m about a third of the way through and Kevin seems utterly out of his depth and his understanding of intelligence, risk, and what you can determine from logical analysis is very poor. His best argument seems to be, let it happen and that’s the only way to say anything instructive about the current moment. . Disappointing, please bring on someone else
@DLH.23
4 ай бұрын
Its ironic that his name was Steve Jobs since his tech is gonna take away all our jobs.....
@WattisWatts
4 ай бұрын
These techno optimists really depress me.
@human_shaped
4 ай бұрын
What a load of deist mumbo jumbo pointless crap at the end. The AI point of view is interesting to hear though. I don't agree but at least it was on a more rational basis.
Пікірлер: 27