Aaron, On Mon, Apr 1, 2013 at 1:13 PM, Aaron Hosford <[email protected]> wrote:
> The BIG question is: What good are a lot of 95% "facts". You can't rely on >> ANYTHING, and as soon as you start putting those "facts" together in >> combination, the accuracy falls WAY below 95%. The more "facts" that are >> strung together, the less accurate the results. For example, the likelihood >> of just 10 x 95% "facts" all being correct is only 60%., and the likelihood >> of getting 20 correct is only 36%. Hence, any "AGI" applying limitless >> computing capability to make sense of Wikipedia is only going to generate a >> lot of gibberish, possibly including some jewels, but without any capacity >> to separate the jewels from the broken glass without access to the real >> world. > > > This only makes sense if the probabilities of the list of "facts" cannot > be leveraged for consistency-based error correction. If I can take two 95% > likely facts and use them to identify a third which is inconsistent with > them, I can often correct that third one based on how it conflicts with the > first two. It doesn't make for guaranteed 100% accuracy, but it can > significantly improve the error rate. It's a similar principle to boosting > in machine learning. > http://en.wikipedia.org/wiki/Boosting_(machine_learning) This is why I > think it's important to integrate knowledge. > Yes, I describe this in my patent application. However, this is ONLY interesting for applications that are error-tolerant. Most proposed AGI applications involve stringing many facts together so the error rate will be increased, yet they are NOT particularly error-tolerant. Steve ------------------------------------------- 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
