On Thu, Apr 11, 2013 at 3:21 AM, Ben Goertzel <[email protected]> wrote:
> Running experiments aimed at teaching you, the developer, useful > things about your system is one thing... > > Running experiments aimed at impressing others, is a different thing... > > Of course we have run OpenCog component code on many example problems > to learn what it does.... A couple simple examples of Fishgram and > Link2Atom were presented in papers at the AGI-12 conference, for > example... Are these papers online? The closest I could find was http://wiki.opencog.org/w/Fishgram and http://wiki.opencog.org/wikihome/images/d/d9/Link2Atom.pdf Regarding Link2Atom, it looks like it was still a proposal about a month before the conference paper deadline. It was yet to be determined if it would be implemented with Fishgram, MOSES, or something else. Are there any experimental results? > I understand that we do not yet have impressive examples of the whole > OpenCog system using its integrated intelligence to carry out > intelligent behaviors. Having folks like you repeatedly point this > out, doesn't actually make progress faster... Actually, you do have the virtual puppy video, I recall. The problem is that it is hard to measure the result. I think it would be more interesting if there was some specific task that could be compared with other approaches, for example, text prediction accuracy, OCR, speech recognition, or image recognition. I realize each one is "narrow AI". But you should be able to show that the same general approach works on all of them. > Building AGI is a complicated process; OpenCog is a big design being > implemented and specified-in-detail by a small team with various > distractions... > > The attitude expressed by you, Tintner and others that our progress is > held back because we are ignorant of basic science, engineering or > testing methodology -- is pretty absurd.... But I'm not going to > waste much time arguing with you people about it ;p ... too much work > to do... No, I think your progress is being held back by lack of funding, insufficient computing power, lack of access to training data, and an overly ambitious goal. I would be interested in working on OpenCog if I thought the approach could succeed. Unfortunately I don't. It's not that I have a better approach. It's that I have a lot of reasons to believe that the cost is much higher than you think it is. IMHO, AGI is going to be solved by lots of people independently working on little bits of the problem, i.e. narrow AI. It is too big for any person or small group to do it all. -- -- 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
