No kidding. I've been thinking about this sort of thing ever since I first learned that there was such a thing as computation theory, but I've never heard of any proper treatment of the various subjects.
> Regarding recording, the main concern was that I didn’t want to inhibit open-ended discussion by recording everyone’s umms, uhhs, and half-baked thoughts for posterity. In addition, there didn’t seem like a pressing need for recording, since I’ve already essentially written a “course textbook” in the form of my essay (http://www.scottaaronson.com/papers/philos.pdf). But as I said, we will have student reaction essays for each class session and they will go up on the website.
Looks really interesting, but seems limited (maybe wisely) to analytic philosophy.
There is a whole strain of work from the 80s that was partly inspired by Heidegger and phenomenology AND the realization that some classical AI problems were computationally intractable. This work was controversial, to say the least, but influential in its way. I'd suggest that any course on this topic ought to at least touch on this work.
I can see why the Dreyfus article would be controversial, but so far (first ten pages or so) I found it very enlightening.
This really struck a chord with me:
> What the learner acquires through experience is
not represented at all but is presented to the learner as more and more finely discriminated situations, and, if the situation does not clearly solicit a single response or
if the response does not produce a satisfactory result, the learner is led to further refine his discriminations, which, in turn, solicit more refined responses.
Comparing that to the meaning of "learn" commonly used in ML to mean "finding optimal weights for some features" and you can see the gulf pretty starkly.
6 comments
[ 3.0 ms ] story [ 21.7 ms ] threadhttp://www.scottaaronson.com/blog/?p=755#comment-27693.
> Regarding recording, the main concern was that I didn’t want to inhibit open-ended discussion by recording everyone’s umms, uhhs, and half-baked thoughts for posterity. In addition, there didn’t seem like a pressing need for recording, since I’ve already essentially written a “course textbook” in the form of my essay (http://www.scottaaronson.com/papers/philos.pdf). But as I said, we will have student reaction essays for each class session and they will go up on the website.
There is a whole strain of work from the 80s that was partly inspired by Heidegger and phenomenology AND the realization that some classical AI problems were computationally intractable. This work was controversial, to say the least, but influential in its way. I'd suggest that any course on this topic ought to at least touch on this work.
http://mit.dspace.org/bitstream/handle/1721.1/6947/AITR-802....
http://leidlmair.at/doc/WhyHeideggerianAIFailed.pdf
This really struck a chord with me:
> What the learner acquires through experience is not represented at all but is presented to the learner as more and more finely discriminated situations, and, if the situation does not clearly solicit a single response or if the response does not produce a satisfactory result, the learner is led to further refine his discriminations, which, in turn, solicit more refined responses.
Comparing that to the meaning of "learn" commonly used in ML to mean "finding optimal weights for some features" and you can see the gulf pretty starkly.