Originally published at: Get a glimpse into the future with this AI-run city | Boing Boing
…
Needs more bears.
2 Likes
3 Likes
7 Likes
Hopefully the video highlights to people that generative language models are not any kind of intelligence.
4 Likes
Ray Bradbury?
3 Likes
If ChatGPT can’t play chess, how’s it going to play SimCity?
1 Like
The problem is people keep moving the goalposts. People claimed that chess took real human intelligence and that it was ridiculous to imagine that a computer could beat a grandmaster. Then they could and people decided “well chess isn’t that hard; Go is a much harder game; computers could never master that” but of course they did that too. It’s true that even the largest artificial neural net is still far smaller than the number of neurons in a human brain, but this idea that there is something"magical" about human intelligence that a similarly large neural net couldn’t do is just special pleading akin to the “God of the gaps” idea that insists that there are supernatural things science can’t deal with.
1 Like
Yes and by doing this we learned a lot about intelligence and human intelligence. We learned that computational intelligence is vastly different from human intelligence. We thought that chess and go were the hard parts and that following instructions was easy, but we got it backwards.
The only way that we have AIs that are even this good is because they have been trained on literally (!) every single piece of knowledge that the trainers were able to get their hands on. This sort of approach is not scalable. I say this because tomorrow’s models will be trained on the output of today’s models since that’s most of the new data being produced these days and this is not a way to get better results.
I do think we are going to be able to squeeze a little bit more performance out of these models, at the cost of vast quantities of energy and hardware, but we’re going to hit a wall pretty soon.[1]
Yes…this is just a prediction by someone on the internet, so take it with appropriate grains of salt. ↩︎
3 Likes
I think the idea of running a city is conceptually unlike any game. All of the games we’ve been able to master are structurally friendlier to computers, even in cases where we can’t yet solve them. They have fixed, observable states, defined victory conditions, and a rigid rule structure. They are subject to iterative analysis and the ability to verify success. City planning is fundamentally within the realm of Wicked Problems and generally don’t meet the criteria. Wicked problem - Wikipedia The goalposts didn’t get moved, they been in motion forever and occasionally turn into willow trees
2 Likes
Philip K. Dick?
1 Like
Star Trek?
I haven’t read Morozow in years, I see he has a podcast on Allende’s tech vision which I think I will give a go to!
This topic was automatically closed after 5 days. New replies are no longer allowed.
