Re: Re: [agi] Wbat would it take to move rapidly toward advanced AGI?
On Sat, Oct 16, 2010 at 5:15 PM, Matt Mahoney matmaho...@yahoo.com wrote: However, programming languages are fundamentally different from natural language in that (1) they have a precise grammar and semantics, That's an advantage. Since you can precisely define what you wish to express. Only when talking to computers. Humans don't learn language that way. Humans learn incrementally yet are still able to use partially learned languages. Ya, well I don't see any difference. Computers also learn incrementally, they start with bios, then kernel, then system services... same thing compilers, they have a base vocabulary (assembly), that's extended by the standard-library, which can be further extended by third-party libraries. humans also learn the meaning of words, by a complex set of sensual cues, associated with a word. Just as a computer can learn the meaning of a word, by reading in a library or dictionary what it means. partially learned languages can also be used, since functional correlations between one word and another can be identified. and (3) the complexity is on the order of 10^5 to 10^6 bits vs. 10^9 bits for natural language. Again it's an advantage, as it would require less resources to do so. You can't express much with such a small language. ? I'd say it's difficult to express stuff with the gooey inconsistency found in NLP. simply due to people not having common dictionary, or many people not using dictionaries. It's far easier to express yourself precisely, when there are standard words, and a consistent grammar. Another advantage of HSPL, is that it's easy to learn, due to it's simplicity. So could be used as an intermediary language. Especially for international negotiations, where clarity and precision, may be valued. --- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/8660244-d750797a Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=8660244-6e7fb59c Powered by Listbox: http://www.listbox.com
[agi] Re: [GI] Digest for general-intellige...@googlegroups.com - 10 Messages in 2 Topics
On 17 October 2010 18:20, Ben Goertzel b...@goertzel.org wrote: In other words, using formal grammar actually makes it harder to establish the connection at the NL-logic interface. IE, it is harder to translate NL sentences to formal grammar than to formal logic. KY Quite the opposite, actually. Translating an NL sentence to a *set* of grammatical trees, representing syntactically possible parses, is almost a solved problem. E.g. the Stanford parser or the link parser do that. Then, translating each of these grammatical trees into a *set* of formal logic expressions, each representing a possible semantic interpretation of the tree, is a partially-solved problem. E.g. OpenCog's RelEx and RelEx2Frame components and Cyc's NL subsystem both do that (in different ways), though not perfectly. So based on the current state of the art, it seems that turning NL into a formal grammar (e.g. a dependency grammar) is significantly less problematic than turning NL into logic, because forming the logic representation requires resolving additional ambiguity, beyond that which must be resolved to form the formal-grammar representation Agree; but would like to add several remarks: --part of the difficulty of applying logic of NL is the need to handle spatial reasoning (A is next to B and B is next to C therefore ...? C is not far from A) -- part of the difficulty of applying logic of NL is the need to handle more abstract reasoning (A is the major of B and majors are people therefore B is a person) (opencyc does this ... not badly) -- Some philosophers of mathematics e.g. Carlo cellucci (see 18 unconventional essays on the nature of mathematics) will stridently point out that, while classical logic is the format in which proofs are stated, it is not at all the method by which mathematicians generate new ideas -- they use reasoning by analogy, by allegory, by induction, and many others, to generate hypothesis which might be possible solutions to problems. I think that we should realize that the same techniques should be applied in AGI: we use reasoning by analogy not because it gives formally correct answers, but because it generates reasonable hypothesis which may or may not be true, but which can be examined in greater detail to see if they are true. These other, non-rigorous reasoning methods are all parts of what we might call intuition -- a set of hard-to-explain reasons why we think something might be true -- which must then be subjected to more rigorous analysis to see if yet more evidence can be found. In short, real-life, just like mathematics, is all about problem-solving and not theorem-proving (which is the last step of creating math, not the first). --linas --- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/8660244-d750797a Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=8660244-6e7fb59c Powered by Listbox: http://www.listbox.com
[agi] Re: [GI] Digest for general-intellige...@googlegroups.com - 10 Messages in 2 Topics
Linas, It seems to me that analogy falls rather simply out of relational probabilistic reasoning. Say we want to make an analogy between two entities A and B. We essentially look for predicates that hold for both A and B; ie, we look for a way to fill in the blank in A is like B, because _. Then, if we want to predict something about B, we know A belongs to the same reference class and can provide 1 piece if evidence concerning the behavior of entities in that reference class. --A On Mon, Oct 18, 2010 at 5:10 PM, Linas Vepstas linasveps...@gmail.comwrote: On 17 October 2010 18:20, Ben Goertzel b...@goertzel.org wrote: In other words, using formal grammar actually makes it harder to establish the connection at the NL-logic interface. IE, it is harder to translate NL sentences to formal grammar than to formal logic. KY Quite the opposite, actually. Translating an NL sentence to a *set* of grammatical trees, representing syntactically possible parses, is almost a solved problem. E.g. the Stanford parser or the link parser do that. Then, translating each of these grammatical trees into a *set* of formal logic expressions, each representing a possible semantic interpretation of the tree, is a partially-solved problem. E.g. OpenCog's RelEx and RelEx2Frame components and Cyc's NL subsystem both do that (in different ways), though not perfectly. So based on the current state of the art, it seems that turning NL into a formal grammar (e.g. a dependency grammar) is significantly less problematic than turning NL into logic, because forming the logic representation requires resolving additional ambiguity, beyond that which must be resolved to form the formal-grammar representation Agree; but would like to add several remarks: --part of the difficulty of applying logic of NL is the need to handle spatial reasoning (A is next to B and B is next to C therefore ...? C is not far from A) -- part of the difficulty of applying logic of NL is the need to handle more abstract reasoning (A is the major of B and majors are people therefore B is a person) (opencyc does this ... not badly) -- Some philosophers of mathematics e.g. Carlo cellucci (see 18 unconventional essays on the nature of mathematics) will stridently point out that, while classical logic is the format in which proofs are stated, it is not at all the method by which mathematicians generate new ideas -- they use reasoning by analogy, by allegory, by induction, and many others, to generate hypothesis which might be possible solutions to problems. I think that we should realize that the same techniques should be applied in AGI: we use reasoning by analogy not because it gives formally correct answers, but because it generates reasonable hypothesis which may or may not be true, but which can be examined in greater detail to see if they are true. These other, non-rigorous reasoning methods are all parts of what we might call intuition -- a set of hard-to-explain reasons why we think something might be true -- which must then be subjected to more rigorous analysis to see if yet more evidence can be found. In short, real-life, just like mathematics, is all about problem-solving and not theorem-proving (which is the last step of creating math, not the first). --linas -- Abram Demski http://lo-tho.blogspot.com/ http://groups.google.com/group/one-logic --- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/8660244-d750797a Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=8660244-6e7fb59c Powered by Listbox: http://www.listbox.com
