On Fri, May 3, 2013 at 9:35 AM, Ben Goertzel <[email protected]> wrote: > On Fri, May 3, 2013 at 9:20 PM, Mike Tintner <[email protected]> wrote:
>> There is not the slightest element of AGI progress – there is no program >> that can be generative – that can “learn and generalize beyond its original >> domain” per you. >> >> Any discussions or predictions of AGI have to *start* from an explanation of >> that fact. None do. > > My explanation for this is that cognitive synergy is both necessary for > achieving AGI with limited computing resources, and "tricky" (in a technical > sense), > > http://multiverseaccordingtoben.blogspot.hk/2011/06/why-is-evaluating-partial-progress.html > > You may not like this explanation, but that's your problem ;p Well, I don't like it either. There is no evidence for synergy as the key to intelligence. All of the evidence is in the other direction. The best predictors are ensemble models. Initial progress is rapid. Adding more models just gives you incremental improvement. Intelligent systems are necessarily complex. Watson is a good example. It has hundreds of modules running in parallel, each able to answer a small (overlapping) subset of questions. Watson does not fail if not all of the modules are working. It just gets a slightly lower score. Likewise with the PAQ data compressors, with hierarchical neural vision systems, and with humans who suffer brain damage. Progress is slow (not zero) because AGI is hard. Work is proceeding in 3 directions: 1. Narrow AI: attempts to get machines to do one thing that humans do well (recognize faces, navigate, transcribe speech, etc). 2. AGI: attempts to get machines to do everything that a human could do. 3. Industrial IT: attempts to get machines to do one thing that a group of millions of humans could do. Lots of money goes into (3), because solving (2) is not very useful if the solution costs more than hiring a human. Money only goes into (1) to the extent that it helps (3). But look where the money is going. We have language translators that translate 100 different languages at rates of hundreds of pages per second. No human could do that, but a large group of humans could. We have search engines that index petabytes of text, and image recognition software that scans billions of images. -- -- Matt Mahoney, [email protected] ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657 Powered by Listbox: http://www.listbox.com
