David Deutsch on Artificial Intelligence
Quantum computation physicist David Deutsch recently published an essay on why we haven't been able to develop a general-purpose artifical intelligence. I think it's well worth reading and in particular, making an effort to understand him, for which it will be helpful to read his book The Beginning of Infinity, which explains some of the key assertions he makes in the essay.
Some comments:
1. I totally agree with him that the core functionality missing so far in attempts to produce artificial general intelligence ('AGI') is "creativity, the ability to produce new explanations." (Note that Deutsch's term 'creativity' is a bit different from common usage.)
2. Deutsch concludes that "it is plausible that just a single idea stands between us and the breakthrough. But it will have to be one of the best ideas ever."
Based on my experience with roadblocks I've encountered in a wide range of disciplines, my guess is Deutsch is wrong: rather, all AGI attempts thus far lack not one, but at least two missing ingredients. I think it's possible that some of what we lack may not even be novel. But the reason for the roadblock is that adding any proper subset of what's missing makes the implementation weaker or intractable.
In this regard, I definitely sympathize with the camp that believes that all we need is more powerful hardware. It seems quite possible that this is one of the missing ingredients. But I agree with Deutsch that we're missing something else, too. Faster computation alone is not going to produce creativity.
3. Despite my agreement with Deutsch, I see a big hole in the logic we share: evolution was able to produce human intelligence -- including in particular the ability to create effective explanations -- without having any ideas at all. Just computation (very broadly defined) and time (billions of years).
How fast can we simulate this process? Do we know how to speed up the simulation, for example, by guiding evolution?
Some comments:
1. I totally agree with him that the core functionality missing so far in attempts to produce artificial general intelligence ('AGI') is "creativity, the ability to produce new explanations." (Note that Deutsch's term 'creativity' is a bit different from common usage.)
2. Deutsch concludes that "it is plausible that just a single idea stands between us and the breakthrough. But it will have to be one of the best ideas ever."
Based on my experience with roadblocks I've encountered in a wide range of disciplines, my guess is Deutsch is wrong: rather, all AGI attempts thus far lack not one, but at least two missing ingredients. I think it's possible that some of what we lack may not even be novel. But the reason for the roadblock is that adding any proper subset of what's missing makes the implementation weaker or intractable.
In this regard, I definitely sympathize with the camp that believes that all we need is more powerful hardware. It seems quite possible that this is one of the missing ingredients. But I agree with Deutsch that we're missing something else, too. Faster computation alone is not going to produce creativity.
3. Despite my agreement with Deutsch, I see a big hole in the logic we share: evolution was able to produce human intelligence -- including in particular the ability to create effective explanations -- without having any ideas at all. Just computation (very broadly defined) and time (billions of years).
How fast can we simulate this process? Do we know how to speed up the simulation, for example, by guiding evolution?