Without AI everybody reading this will face oblivion in less than a century, in most cases much less, with AI maybe not. And do you really believe the measures you propose can stop the advance of AI over the long run? When it comes to data processing, and therefore of intelligence, electronics has a HUGE advantage over biology and nothing is going to change that, your proposals are just rearranging the deck chairs on the Titanic. Having said that I thought your new book was thoughtful and I enjoyed it, even though I disagreed with much of it.
As a professional AI alignment researcher, we really have no idea wtf we’re doing in general, but we have some ideas. Some more time to work on the problem before the most powerful AIs get yet more powerful would help. A lot.
Also, if we do the “international ASI ban” idea (which is humanity’s best hope), that doesn’t mean the practical AI capabilities available to the public will stagnate. The AI-powered tools based on the existing models will get better and better! And this will indeed speed up alignment research… but humans won’t win the race against RSI unless we hobble the AI capabilities by restricting brute-force compute.
Without AI everybody reading this will face oblivion in less than a century, in most cases much less, with AI maybe not. And do you really believe the measures you propose can stop the advance of AI over the long run? When it comes to data processing, and therefore of intelligence, electronics has a HUGE advantage over biology and nothing is going to change that, your proposals are just rearranging the deck chairs on the Titanic. Having said that I thought your new book was thoughtful and I enjoyed it, even though I disagreed with much of it.
John K Clark
As a professional AI alignment researcher, we really have no idea wtf we’re doing in general, but we have some ideas. Some more time to work on the problem before the most powerful AIs get yet more powerful would help. A lot.
Also, if we do the “international ASI ban” idea (which is humanity’s best hope), that doesn’t mean the practical AI capabilities available to the public will stagnate. The AI-powered tools based on the existing models will get better and better! And this will indeed speed up alignment research… but humans won’t win the race against RSI unless we hobble the AI capabilities by restricting brute-force compute.