I am not interested in writing a translator. I don't think I could write a program to understand the email I just wrote because I used ideas that are quite advanced. (Certainly too advanced for a contemporary agi program and probably too advanced for the majority of human beings to understand.)
I want to be able to study the problem of using text to enable an AGI program to learn through the use of conversation without resorting to a context-free language. I am certain that I could write an AI program that could learn if the conversation used a context-free language. But an AI program that was capable of learning using a context-sensitive language as a base for conversation is, in my opinion, capable of genuine general learning (as far as might be accomplished within the domain of text). So I am interested in the question of what do we need to do to get a program to learn through conversation. I am not interested in creating a context-free conversational language because my opinion is that natural language is based on the fact that the components of a sentence or a text can add to the understanding of the text. So simply resolving the text into some particular context-free interpretation is not enough for my plan. Parsing and resolving context-sensitive or ambiguous phrases is a part of understanding but there is more to it than that. Jim Bromer On Sat, Oct 26, 2013 at 4:43 PM, Matt Mahoney <[email protected]> wrote: > Do you think you could write a program to parse the email you just > wrote? Do you think you could write a program to translate it into > another language? > > Language evolved to be efficiently learnable on a very complex, > embodied, 1 petaflop, 100 TB parallel computer, whose software took 3 > billion years to write, after being fed training data over a 1 Gb/s > sensory input channel for several years. Are you surprised it's a hard > problem? > > On Fri, Oct 25, 2013 at 10:17 PM, Jim Bromer <[email protected]> wrote: >> I am convinced that it would be easy to get a text-based Learning-AI >> program learn to respond in fairly simple ways to simple texts. (And >> I will be in a position to try it out in the near future.) The >> question is whether this kind of ability has to be at the expense of >> an ability to integrate more sophisticated kinds of learning into it. >> >> I just do not see why people have not produced solid examples of >> simple learning using text-based AI unless the problem was either that >> they felt they needed to impress the skeptics or they became >> confounded by their own, more complicated use of language. >> >> Simple language does not have to be at the level of a programming >> language. I think that programming languages are "context free" >> because even though the apparent context may seem to violate the >> context of the substrings taken separately, any particular string >> (that is any grammatical string) will still only generate one >> particular output. >> >> So a computer could (genuinely) learn about simple strings that might >> not be context free and use them to generate different points. As >> long as this was kept relatively simple it should be completely >> feasible and it might be a good starting point to examine what was >> going on. (Even though a text only AI program would not be capable of >> applying its knowledge in a sophisticated way, it could still >> constitute genuine learning in my opinion because it would be able to >> learn new things within the domain of the text-based interactions.) >> >> So even though my data management system is neither simple nor >> sophisticated, I believe that I will be able to use it for simple but >> somewhat sophisticated kind of learning which would be general within >> the limits of the domain of text. >> >> Jim Bromer >> >> >> ------------------------------------------- >> AGI >> Archives: https://www.listbox.com/member/archive/303/=now >> RSS Feed: https://www.listbox.com/member/archive/rss/303/3701026-786a0853 >> Modify Your Subscription: https://www.listbox.com/member/?&; >> Powered by Listbox: http://www.listbox.com > > > > -- > -- Matt Mahoney, [email protected] > > > ------------------------------------------- > AGI > Archives: https://www.listbox.com/member/archive/303/=now > RSS Feed: https://www.listbox.com/member/archive/rss/303/24379807-f5817f28 > Modify Your Subscription: https://www.listbox.com/member/?&; > Powered by Listbox: http://www.listbox.com ------------------------------------------- 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
