tag:blogger.com,1999:blog-1928622184607631910.post8213063906979156256..comments2023-10-22T01:56:21.243-07:00Comments on Where Do We Go from Here?: Who Pays the Bill for Cities? Dave Aldenhttp://www.blogger.com/profile/04365271229524041881noreply@blogger.comBlogger2125tag:blogger.com,1999:blog-1928622184607631910.post-48849668770142794302013-01-07T11:33:48.268-08:002013-01-07T11:33:48.268-08:00Dan, thanks for the comment. As you say, you'...Dan, thanks for the comment. As you say, you're decades ahead of me in grasping the AI issues, so I'll defer to you.<br /><br />However, I do have one thought. On the air bag deployment situation, I reluctantly agree with the attorneys. Not because the heuristic system would be better but because it would be more easy to defend in a courtroom. It's astonishing how anti-scientific courtrooms can be. I sat on a jury about five years ago. The evidence involved statistics and physics. Both attorneys presented arguments that a reasonably alert high school science student could have shot full of home. But I was the only one of the jurors to note the errors. And that greatly affected the decision.Dave Aldenhttps://www.blogger.com/profile/04365271229524041881noreply@blogger.comtag:blogger.com,1999:blog-1928622184607631910.post-90659047071646390072012-11-28T13:04:30.282-08:002012-11-28T13:04:30.282-08:00On the Gary Marcus article: a friend of mine who h...On the Gary Marcus article: a friend of mine who has worked in the car industry described building a neural network system to trigger, I think, air bag deployments. Give it a whole bunch of samples of accelerometer data from situations where the air bags should and shouldn't deploy, and the neural network determined how to make that decision in every sample they could send at it.<br /><br />And then legal made them change it to a system based on heuristics that probably (in his opinion) has more false positive triggers, because there was no way to document for sure what the neural network deployment decider would do in absolutely all circumstances.<br /><br />What mattered was the determinism of the systems.<br /><br />I suspect that Prof. Marcus's concerns are, first, decades in the future. But free will and determinism essentially come down to "at some point we don't understand the system" vs "we understand the system", and as we learn more and more, that former becomes "we can't distinguish the behavior of the system from random".<br /><br />So, as someone who's been looking at AI issues for decades, I think he's got his philosophical blinders on: The machines will do what we tell them to. At some point we may choose to use statistical models, or we may use heuristics. We may build systems that are so complex that we don't understand them, but we will still be the creators.<br /><br />We don't question the ethics of the land mine itself, though we question the ethics of deploying the land mine. If anything, I think we'll evolve towards understanding further how humans are deterministic and lack free will, rather than thinking that cars have intent.Anonymousnoreply@blogger.com