undefined | Better HN

0 pointsmattdesl2y ago0 comments

> Similarly, nobody doubted that neural networks were capable of very interesting things - the holdup was the level of processing power needed to run them. As soon as that changes, useful applications abounded.

This is incorrect. AI has gone through multiple ‘winters’ where there were serious doubts and pessimistic attitudes toward its capabilities.

https://en.wikipedia.org/wiki/AI_winter

0 comments

acdha2y ago

Yes, I’m aware of AI winters but that doesn’t affect the point I was making: people knew neural networks were capable of the task since neuroscientists were studying how our own brains used them, but they were computationally daunting. “AI winter” didn’t mean that everyone gave up and assumed the theists were right that some supernatural power was the key part, it meant that developing commercially-viable technologies was harder than hoped.

That’s where the contrast with Bitcoin is so pronounced: in that case, the limitation isn’t technical but political - it’s been available to anyone who wants it for 15 years but most people don’t want it because you have to strongly share a certain ideology to prefer a slower, less secure, more expensive financial system. There is no technical improvement which will suddenly boost Bitcoin adoption the way GPUs and smart algorithms boosted neural networks because the inefficiency is the point.

mattdeslOP2y ago

the whole idea of an “AI winter” is that people did not feel that AI was capable and worth pursuing, the Wikipedia link makes that sentiment clear. The remaining computer scientists still working in AI space obviously continued developing the technology despite the lack of public interest and major funding, but the same happens in crypto: engineers are still developing new ideas and technologies within blockchain space (ZK proofs, verkle trees, etc). Bitcoin has stagnated but other technologies have not.

acdha2y ago

> the whole idea of an “AI winter” is that people did not feel that AI was capable and worth pursuing, the Wikipedia link makes that sentiment clear.

Try counting the number of times people mention phrases like “combinatorial explosion” or other limitations like the single/multi-layer perceptron argument, which are the kinds of problems we’re talking about where the issue was feasibility on the hardware available, or things like “expert systems” which were dead ends unrelated to neural networks.

Now, you can try to change the topic again to hypothesize that some non-blockchain technology will become popular but that’s no more relevant to this thread than the 80s expert systems people were to modern machine learning systems. Different technology and implementations having different results isn’t exactly disproving criticism.

1 more reply

j / k navigate · click thread line to collapse

0 comments

acdha2y ago

mattdeslOP2y ago

acdha2y ago

> the whole idea of an “AI winter” is that people did not feel that AI was capable and worth pursuing, the Wikipedia link makes that sentiment clear.

1 more reply

j / k navigate · click thread line to collapse