Re:What China Can Teach the U.S. About Artificial Intelligence
by AndrewKemendo on Sunday September 23, 2018 @08:48PM
(#18054251)
Lee never describes what the practical result of "A.I. supremacy" actually means. The only other place that "supremacy" in terms of between nation states is used is with military matters, e.g. Air Supremacy [1].
Is he talking about General Intelligence? Is he arguing that there are radical efficiencies that ML gives you that you can't get elsewhere?
I haven't seen a clear vision of what he means by "supremacy."
Aside from that, and I am well aware of Mr. Lee's CV, his book, and his resulting wave of articles reads like it's from someone who has never actually built a product that functionally based around ML. That's likely because he seems to be writing from a general audience, but some of the points just don't seem to hold.
For example, that SV doesn't like to copy. If there is anything true it's that SV LOVES to copy. See: Apple and MSFT copying everything important from PARC, IG copying Snap etc...
On the points about "AI," the example of being "first" to some arbitrary goal like fully autonomous cars is a weird near-strawman. Let's say that Didi does get to Level 5 autonomy first. Does that mean the rest of the world's on-demand driving services will completely capitulate to them? Of course not. Standard economics apply. Deep Learning hasn't upended the rules of economics just yet. Maybe once we're at the point where labor inputs are near zero because of automation we can talk.
Now if Lee were to start making the argument that this stuff matters in national defense and intelligence matters, or about the race to general AI, then we would finally have a good conversation. However he's not making that argument because it's a little too early to have it, and nobody seems to want to have it anyway.
Re:What China Can Teach the U.S. About Artificial Intelligence
by lngnmn1 on Sunday September 23, 2018 @02:44AM
(#18049742)
That the notion of intelligence requires the notion of instincts and hardwired self-preservation, so that training/learning phase would be more efficient and meaningful.
Basically, we should try to build an animal, which has an astonishing amount of intelligence (unattainable by any ML yet) in the evolved to match the shared environment brain's structure) first, before trying attacking a human-like intelligence.
Animal mind first. All the pattern-matching on hardwired pre-build structure, augmented (transformed) by learning is here.
Is he talking about General Intelligence? Is he arguing that there are radical efficiencies that ML gives you that you can't get elsewhere?
I haven't seen a clear vision of what he means by "supremacy."
Aside from that, and I am well aware of Mr. Lee's CV, his book, and his resulting wave of articles reads like it's from someone who has never actually built a product that functionally based around ML. That's likely because he seems to be writing from a general audience, but some of the points just don't seem to hold.
For example, that SV doesn't like to copy. If there is anything true it's that SV LOVES to copy. See: Apple and MSFT copying everything important from PARC, IG copying Snap etc...
On the points about "AI," the example of being "first" to some arbitrary goal like fully autonomous cars is a weird near-strawman. Let's say that Didi does get to Level 5 autonomy first. Does that mean the rest of the world's on-demand driving services will completely capitulate to them? Of course not. Standard economics apply. Deep Learning hasn't upended the rules of economics just yet. Maybe once we're at the point where labor inputs are near zero because of automation we can talk.
Now if Lee were to start making the argument that this stuff matters in national defense and intelligence matters, or about the race to general AI, then we would finally have a good conversation. However he's not making that argument because it's a little too early to have it, and nobody seems to want to have it anyway.
[1] https://en.wikipedia.org/wiki/Air_supremacy
>... enormous data flows [give] a multidimensional picture of each Chinese user
>copying: imitating a successful business model and then tweaking it.
>encouraging local officials to make the changes needed in their area so that private A.I. companies can develop products
Dunno - guess it helps a bit
Basically, we should try to build an animal, which has an astonishing amount of intelligence (unattainable by any ML yet) in the evolved to match the shared environment brain's structure) first, before trying attacking a human-like intelligence.
Animal mind first. All the pattern-matching on hardwired pre-build structure, augmented (transformed) by learning is here.