« America is the Education of the World | Main | Traveling »

March 10, 2007



Your songs and voice are very sweet, beautiful and lovely. Made me realize smart people can sing too! Kudos.


Back in the day, he used to be pretty good at playing piano too.


Here's an interesting YouTube clip about predicting technological growth. This sort of evidence is what underlies materialist philosophical paradigms such as mine and Nato's.



It's a fascinating talk but there are two areas in which I think he's too optimistic: biological reverse-engineering and its special constituent, cognitive reverse engineering.

In the former case, my unease with his sunny predictions springs primarily from his citation of advances in sequencing DNA: a task that, for all its technical sophistication, is fairly straightforward. Parsing the interplay of homeobox genes in a multicellular organism, on the other hand, seems like a far less immediately tractable problem.

In the latter, I feel that he's taking the boxological cognitive models as being more complete and more detailed than they are relative to the brain's actual machinery. "Complete" and "detailed" need some unpacking to explain my reservations.

By "complete," I mean, "addressing all that needs to be addressed." It seems unlikely that cognitive science will boil down to enumeration and description of cognitive subsystems any more than enumeration and description of the proteins for which genes code exhausts developmental genetics. There's a huge systems integration task that seems likely to remain fairly opaque for longer than it will take to make human-speed computers.

"Detail" needs less explanation on its face - the drafters of boxological charts now they are outlining toy models rather than getting every last bit in there. On the less obvious side, I think it's fairly likely that there are many fine-but-critical points that are invisible to us without a level of visibility that's frankly implausible to me in anything like the near future. I don't mean quantum effects or anything as hopeless as that; I just mean that it seems almost certain that some important neural behaviors will only be observable in situ and at the several-neuron resolution.

That's not to say I think AI won't be achievable until we figure it all out. Really what I mean is that I think that we'll probably achieve it without exactly understanding what's going on. The boxological models will get us close enough to fail interestingly, and then combinations of trail-and-error, monkey-see-monkey-do, and perhaps some harnessing of evolutionary design will get us the rest of the way there. It's not unlike the history of aircraft design, really.


I *will* say I find his optimism quite cheering.


I think your reservations are beyond the scope of his lecture. His main point is that progress has been exponential on every observable level, in all fields. He said that it's basically impossible to make predictions in the particular, but that predictions in the general are very easy to make and study. All he has to do is look at the trends in his graphs to see how soon it will be until we have X capabilities to do Y achievements with Z resources. He's not really trying to be optimistic, he's trying to be realistic, and I think his arguments are very sound from a basic statistics perspective.


His overall project certainly seems defensible; I just cite a couple of instances to which his project should either not be generalized or to which he has perhaps applied the wrong curve. These are fundamentally niggling, but since they form a part of the concrete predictions he makes from his project, they will certainly impact peoples' judgement of the larger claim. "If everything's advancing at this exponential rates, why don't we have aircars?" a skeptic might ask - the answer being that the sort of technology in question isn't the same as the theory would predict to be on the exponential curve. A technical issue, but one with teeth.

The comments to this entry are closed.

My Photo

Only use a payday cash advance as a last resort.


Blog powered by Typepad