If anyone ever gets serious about advancing artificial intelligence, they'll fund The Hutter Prize for Lossless Compression of Human Knowledge<http://prize.hutter1.net/>at something like the risk adjusted net present value of artificial intelligence.
If that happens, truly interesting things will start to emerge. On Thu, Jan 17, 2013 at 6:50 PM, Jed Rothwell <jedrothw...@gmail.com> wrote: > Mark Gibbs <mgi...@gibbs.com> wrote: > > >> What if Watson concludes LENR is a load of baloney? >> > > Seriously? That would not surprise me. I have been reading about Watson > and the methodology, because I am interested in natural language > processing, translation and so on. > > Watson does not do much synthesizing. It parses the natural language > sources, determines what they "mean" (in a sense) and then matches them to > the inquiry. That's fabulous, and useful, but not creative. > > Suppose you input into Watson only mass media sources, the Scientific > American, Nature and a few other leading journals. You do not add anything > from J. Electroanal. Chem. or other specialized journals that have actual > papers about cold fusion. Then you ask Watson "what is cold fusion." It > would say something similar to what Ouellette wrote; i.e., it is > pathological science, never replicated, blah, blah, blah: > > > http://blogs.scientificamerican.com/cocktail-party-physics/2012/10/29/genie-in-a-bottle-the-case-against-cold-fusion/ > > This article, along with everything else in the Sci. Am. and Nature about > cold fusion, is a long string of fact-free nonsense, but it is perfectly > grammatical and it could be parsed and regurgitated by Watson. Ouellette > herself is essentially acting as a computer. She does no fact checking. She > does not try to eliminate internal inconsistencies. She simply parrots > whatever she read in Wikipedia, or some other Internet sewer. > > Watson echoes what it reads. > > If you were to input all of the papers at LENR-CANR.org into Watson than > its response would echo actual, peer-reviewed scientific literature, and > the conclusions therein, which are overwhelmingly positive. > > > Note that I was only kidding in the first message in this thread. There is > no chance Watson would actually synthesize a general view of cancer > treatment, similar to the way a human epidemiologist would. I think it will > only search out specific papers and studies to address specific questions. > It does actually "understand" the conclusions in these paper at some level, > since it can tell that a statement is positive or negative; i.e., treatment > X is likely to be effective for disease Y. (Or that it has not been found > effective.) What it cannot do is conclude that "thousands of papers have > statements describing effective treatments, but a study made outside the > field by epidemiologists proves that all of these papers are exaggerated or > wrong." That is a very high level of generalization -- or abstraction if > you will. That would be finding an answer far afield of the set of usual > answers, with no contextual clues or linked keywords. I do not think Watson > is capable of it. > > I suppose that someday computers will be capable of this. > > - Jed > >
