Re: [agi] uncertain logic criteria
On Thu, Sep 18, 2008 at 3:06 AM, Ben Goertzel [EMAIL PROTECTED] wrote: Prolog is not fast, it is painfully slow for complex inferences due to using backtracking as a control mechanism The time-complexity issue that matters for inference engines is inference-control ... i.e. dampening the combinatorial explosion (which backtracking does not do) Time-complexity issues within a single inference step can always be handled via mathematical or code optimization, whereas optimizing inference control is a deep, deep AI problem... So, actually, the main criterion for the AGI-friendliness of an inference scheme is whether it lends itself to flexible, adaptive control via -- taking long-term, cross-problem inference history into account -- learning appropriately from noninferential cognitive mechanisms (e.g. attention allocation...) (I've been busy implementing my AGI in Lisp recently...) I think optimization of single inference steps and using global heuristics are both important. Prolog uses backtracking, but in my system I use all sorts of search strategies, not to mention abduction and induction. Also, currently I'm using general resolution instead of SLD resolution, which is for Horn clauses only. But one problem I face is that when I want to deal with equalities I have to use paramodulation (or some similar trick). This makes things more complex and as you know, I don't like it! I wonder if PLN has a binary-logic subset, or is every TV probabilistic by default? If you have a binary logic subset, then how does that subset differ from classical logic? People have said many times that resolution is inefficient, but I have never seen a theorem that says resolution is slower than other deduction methods such as natural deduction or tableaux. All such talk is based on anecdotal impressions. Also, I don't see why other deduction methods are that much different from resolution since their inference steps correspond to resolution steps very closely. Also, if you can apply heuristics in other deduction methods you can do the same with resolution. All in all, I see no reason why resolution is inferior. So I'm wondering if there are some novel way of doing binary that somehow makes inference faster than with classical logic. And exactly what is the price to be paid? What aspects of classical logic are lost? YKY --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=114414975-3c8e69 Powered by Listbox: http://www.listbox.com
Re: [agi] uncertain logic criteria
I'm in the process of reading this paper: http://www.jair.org/papers/paper1410.html It might answer a couple of your questions. And, it looks like it has an interesting proposal about generating heuristics from the problem description. The setting is boolean rather than firs-order. It discusses the point about resolution being slow in practice. --Abram Demski On Tue, Sep 23, 2008 at 3:31 AM, YKY (Yan King Yin) [EMAIL PROTECTED] wrote: On Thu, Sep 18, 2008 at 3:06 AM, Ben Goertzel [EMAIL PROTECTED] wrote: Prolog is not fast, it is painfully slow for complex inferences due to using backtracking as a control mechanism The time-complexity issue that matters for inference engines is inference-control ... i.e. dampening the combinatorial explosion (which backtracking does not do) Time-complexity issues within a single inference step can always be handled via mathematical or code optimization, whereas optimizing inference control is a deep, deep AI problem... So, actually, the main criterion for the AGI-friendliness of an inference scheme is whether it lends itself to flexible, adaptive control via -- taking long-term, cross-problem inference history into account -- learning appropriately from noninferential cognitive mechanisms (e.g. attention allocation...) (I've been busy implementing my AGI in Lisp recently...) I think optimization of single inference steps and using global heuristics are both important. Prolog uses backtracking, but in my system I use all sorts of search strategies, not to mention abduction and induction. Also, currently I'm using general resolution instead of SLD resolution, which is for Horn clauses only. But one problem I face is that when I want to deal with equalities I have to use paramodulation (or some similar trick). This makes things more complex and as you know, I don't like it! I wonder if PLN has a binary-logic subset, or is every TV probabilistic by default? If you have a binary logic subset, then how does that subset differ from classical logic? People have said many times that resolution is inefficient, but I have never seen a theorem that says resolution is slower than other deduction methods such as natural deduction or tableaux. All such talk is based on anecdotal impressions. Also, I don't see why other deduction methods are that much different from resolution since their inference steps correspond to resolution steps very closely. Also, if you can apply heuristics in other deduction methods you can do the same with resolution. All in all, I see no reason why resolution is inferior. So I'm wondering if there are some novel way of doing binary that somehow makes inference faster than with classical logic. And exactly what is the price to be paid? What aspects of classical logic are lost? YKY --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ 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/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=114414975-3c8e69 Powered by Listbox: http://www.listbox.com
Re: [agi] uncertain logic criteria
On Tue, Sep 23, 2008 at 6:59 PM, Abram Demski [EMAIL PROTECTED] wrote: I'm in the process of reading this paper: http://www.jair.org/papers/paper1410.html It might answer a couple of your questions. And, it looks like it has an interesting proposal about generating heuristics from the problem description. The setting is boolean rather than firs-order. It discusses the point about resolution being slow in practice. First-order theorem proving is very different from propositional, the techniques do not transfer there. I'd be very delighted if you can show a paper about a superior algorithm for first-order =) YKY --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=114414975-3c8e69 Powered by Listbox: http://www.listbox.com
Re: [agi] uncertain logic criteria
PLN can do inference on crisp-truth-valued statements ... and on this subset, it's equivalent to ordinary predicate logic ... About resolution and inference: resolution is a single inference step. To make a theorem-prover, you must couple resolution with some search strategy. For a search strategy, Prolog uses backtracking, which is extremely crude. My beef is not with resolution but with backtracking. Another comment: even if one's premises and conclusion are crisp-truth-valued, it may still be worthwhile to deal with uncertain-truth-valued statements in the course of doing inference. Guesses, systematically managed, may help on the way from definite premises to definite conclusions... ben g On Tue, Sep 23, 2008 at 3:31 AM, YKY (Yan King Yin) [EMAIL PROTECTED] wrote: On Thu, Sep 18, 2008 at 3:06 AM, Ben Goertzel [EMAIL PROTECTED] wrote: Prolog is not fast, it is painfully slow for complex inferences due to using backtracking as a control mechanism The time-complexity issue that matters for inference engines is inference-control ... i.e. dampening the combinatorial explosion (which backtracking does not do) Time-complexity issues within a single inference step can always be handled via mathematical or code optimization, whereas optimizing inference control is a deep, deep AI problem... So, actually, the main criterion for the AGI-friendliness of an inference scheme is whether it lends itself to flexible, adaptive control via -- taking long-term, cross-problem inference history into account -- learning appropriately from noninferential cognitive mechanisms (e.g. attention allocation...) (I've been busy implementing my AGI in Lisp recently...) I think optimization of single inference steps and using global heuristics are both important. Prolog uses backtracking, but in my system I use all sorts of search strategies, not to mention abduction and induction. Also, currently I'm using general resolution instead of SLD resolution, which is for Horn clauses only. But one problem I face is that when I want to deal with equalities I have to use paramodulation (or some similar trick). This makes things more complex and as you know, I don't like it! I wonder if PLN has a binary-logic subset, or is every TV probabilistic by default? If you have a binary logic subset, then how does that subset differ from classical logic? People have said many times that resolution is inefficient, but I have never seen a theorem that says resolution is slower than other deduction methods such as natural deduction or tableaux. All such talk is based on anecdotal impressions. Also, I don't see why other deduction methods are that much different from resolution since their inference steps correspond to resolution steps very closely. Also, if you can apply heuristics in other deduction methods you can do the same with resolution. All in all, I see no reason why resolution is inferior. So I'm wondering if there are some novel way of doing binary that somehow makes inference faster than with classical logic. And exactly what is the price to be paid? What aspects of classical logic are lost? YKY --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?; Powered by Listbox: http://www.listbox.com -- Ben Goertzel, PhD CEO, Novamente LLC and Biomind LLC Director of Research, SIAI [EMAIL PROTECTED] Nothing will ever be attempted if all possible objections must be first overcome - Dr Samuel Johnson --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=114414975-3c8e69 Powered by Listbox: http://www.listbox.com
Re: [agi] uncertain logic criteria
No transfer? This paper suggests otherwise: http://www.cs.washington.edu/homes/pedrod/papers/aaai06b.pdf -Abram Demski On Tue, Sep 23, 2008 at 7:31 AM, YKY (Yan King Yin) [EMAIL PROTECTED] wrote: On Tue, Sep 23, 2008 at 6:59 PM, Abram Demski [EMAIL PROTECTED] wrote: I'm in the process of reading this paper: http://www.jair.org/papers/paper1410.html It might answer a couple of your questions. And, it looks like it has an interesting proposal about generating heuristics from the problem description. The setting is boolean rather than firs-order. It discusses the point about resolution being slow in practice. First-order theorem proving is very different from propositional, the techniques do not transfer there. I'd be very delighted if you can show a paper about a superior algorithm for first-order =) YKY --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ 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/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=114414975-3c8e69 Powered by Listbox: http://www.listbox.com
Re: [agi] uncertain logic criteria
On Tue, Sep 23, 2008 at 9:00 PM, Abram Demski [EMAIL PROTECTED] wrote: No transfer? This paper suggests otherwise: http://www.cs.washington.edu/homes/pedrod/papers/aaai06b.pdf Well, people know that propositional SAT is fast, so propositionalization is a tempting heuristic, but as the paper's abstract has stated, it can only apply to small domains. AGI is precisely a large-domain problem! YKY --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=114414975-3c8e69 Powered by Listbox: http://www.listbox.com
Re: [agi] uncertain logic criteria
On Tue, Sep 23, 2008 at 9:00 PM, Abram Demski [EMAIL PROTECTED] No transfer? This paper suggests otherwise: http://www.cs.washington.edu/homes/pedrod/papers/aaai06b.pdf Sorry, I replied too quickly... This paper does contribute to solving FOL inference problems, but it is still inadequate for AGI because the FOL is required to be function-free. If you remember programming in Prolog, we often use functors within predicates. My guess is that commonsense reasoning would make use of such functors as well. YKY --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=114414975-3c8e69 Powered by Listbox: http://www.listbox.com
Re: [agi] uncertain logic criteria
On Tue, Sep 23, 2008 at 9:20 PM, YKY (Yan King Yin) Sorry, I replied too quickly... This paper does contribute to solving FOL inference problems, but it is still inadequate for AGI because the FOL is required to be function-free. If you remember programming in Prolog, we often use functors within predicates. My guess is that commonsense reasoning would make use of such functors as well. Well, even when FOL-with-functions can be converted to function-free FOL, the blow-up may be too much for a commonsense KB. YKY --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=114414975-3c8e69 Powered by Listbox: http://www.listbox.com
Re: [agi] uncertain logic criteria
I don't know prolog's functors. But, I agree that the approach is fundamentally limited, because it is restricted to finite domains. -Abram Demski On Tue, Sep 23, 2008 at 9:20 AM, YKY (Yan King Yin) [EMAIL PROTECTED] wrote: On Tue, Sep 23, 2008 at 9:00 PM, Abram Demski [EMAIL PROTECTED] No transfer? This paper suggests otherwise: http://www.cs.washington.edu/homes/pedrod/papers/aaai06b.pdf Sorry, I replied too quickly... This paper does contribute to solving FOL inference problems, but it is still inadequate for AGI because the FOL is required to be function-free. If you remember programming in Prolog, we often use functors within predicates. My guess is that commonsense reasoning would make use of such functors as well. YKY --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ 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/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=114414975-3c8e69 Powered by Listbox: http://www.listbox.com
Re: [agi] uncertain logic criteria
On Wed, Sep 17, 2008 at 10:54 PM, Abram Demski [EMAIL PROTECTED] wrote: Pei, You are right, that does sound better than quick-and-dirty. And more relevant, because my primary interest here is to get a handle on what normative epistemology should tell us to conclude if we do not have time to calculate the full set of consequences to (uncertain) facts. Fully understand. As far as uncertain reasoning is concerned, NARS aims at a normative model that is optimal under certain restriction, and in this sense it is not inferior to probability theory, but designed under different assumptions. Especially, NARS is not an approximation or a second-rate substitute for probability theory, just as probability theory is not a second-rate substitute of binary logic. It is unfortunate that I had to use biased language, but probability is of course what I am familiar with... I suppose, though, that most of the terms could be roughly translated into NARS? Especially independence, and I should hope conditional independence as well. Collapsing probabilities can be restated as generally collapsing uncertainty. From page 80 of my book: We call quantities mutually independent of each other, when given the values of any of them, the remaining ones cannot be determined, or even bounded approximately. Thanks for the links. The reason for singling out these three, of course, is that they have already been discussed on this list. If anybody wants to point out any others in particular, that would be great. Understand. The UAI community used to be an interesting one, though in recent years it has been too much dominated by the Bayesians, who assume they already get the big picture right, and all the remain issues are in the details. For discussions on the fundamental properties of uncertain reasoning, I recommend the works of Henry Kyburg and Susan Haack. Pei --Abram On Wed, Sep 17, 2008 at 3:54 PM, Pei Wang [EMAIL PROTECTED] wrote: On Wed, Sep 17, 2008 at 1:46 PM, Abram Demski [EMAIL PROTECTED] wrote: Hi everyone, Most people on this list should know about at least 3 uncertain logics claiming to be AGI-grade (or close): --Pie Wang's NARS Yes, I heard of this guy a few times, who happens to use the same name for his project as mine. ;-) Here is my list: 1. Well-defined uncertainty semantics (either probability theory or a well-argued alternative) Agree, and I'm glad that you mentioned this item first. 2. Good at quick-and-dirty reasoning when needed --a. Makes unwarranted independence assumptions --b. Collapses probability distributions down to the most probable item when necessary for fast reasoning --c. Uses the maximum entropy distribution when it doesn't have time to calculate the true distribution --d. Learns simple conditional models (like 1st-order markov models) for use later when full models are too complicated to quickly use As you admitted in the following, the language is biased. Using theory-neutral language, I'd say the requirement is to derive conclusions with available knowledge and resources only, which sounds much better than quick-and-dirty to me. 3. Capable of repairing initial conclusions based on the bad models through further reasoning --a. Should have a good way of representing the special sort of uncertainty that results from the methods above --b. Should have a repair algorithm based on that higher-order uncertainty As soon as you don't assume there is a model, this item and the above one become similar, which are what I called revision and inference, respectively, in http://www.cogsci.indiana.edu/pub/wang.uncertainties.ps The 3 logics mentioned above vary in how well they address these issues, of course, but they are all essentially descended from NARS. My impression is that as a result they are strong in (2a) and (3b) at least, but I am not sure about the rest. (Of course, it is hard to evaluate NARS on most of the points in #2 since I stated them in the language of probability theory. And, opinions will differ on (1).) Anyone else have lists? Or thoughts? If you consider approaches with various scope and maturity, there are much more than these three approaches, and I'm sure most of people working on them will claim that they are also general purpose. Interested people may want to browse http://www.auai.org/ and http://www.elsevier.com/wps/find/journaldescription.cws_home/505787/description#description Pei --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ 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/ Modify Your Subscription: https://www.listbox.com/member/?; Powered by Listbox:
Re: [agi] uncertain logic criteria
On Thu, Sep 18, 2008 at 4:21 AM, Kingma, D.P. [EMAIL PROTECTED] wrote: Small question... aren't Bbayesian network nodes just _conditionally_ independent: so that set A is only independent from set B when d-separated by some set Z? So please clarify, if possible, what kind of independence you assume in your model. Sorry, I made a mistake. You're right that X and Y can be dependent even if there is no direct link between them in a Bayesian network. I am currently trying to develop an approximate algorithm for Bayesian network inference. Exact BN inference takes care of dependencies as specified in the BN, but I suspect that an approximate algorithm may be faster. I have not worked out the details of this algorithm yet... and the talk about independence was misleading. YKY --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=114414975-3c8e69 Powered by Listbox: http://www.listbox.com
Re: [agi] uncertain logic criteria
On Thu, Sep 18, 2008 at 1:46 AM, Abram Demski [EMAIL PROTECTED] wrote: Speaking of my BPZ-logic... 2. Good at quick-and-dirty reasoning when needed Right now I'm focusing on quick-and-dirty *only*. I wish to make the logic's speed approach that of Prolog (which is a fast inference algorithm for binary logic). --a. Makes unwarranted independence assumptions Yes, I think independence should always be assumed unless otherwise stated -- which means there exists a Bayesian network link between X and Y. --b. Collapses probability distributions down to the most probable item when necessary for fast reasoning Do you mean collapsing to binary values? Yes, that is done in BPZ-logic. --c. Uses the maximum entropy distribution when it doesn't have time to calculate the true distribution Not done yet. I'm not familiar with max-ent. Will study that later. --d. Learns simple conditional models (like 1st-order markov models) for use later when full models are too complicated to quickly use I focus on learning 1st-order Bayesian networks. I think we should start with learning 1st-order Bayesian / Markov. I will explore mixing Markov and Bayesian when I have time... 3. Capable of repairing initial conclusions based on the bad models through further reasoning --a. Should have a good way of representing the special sort of uncertainty that results from the methods above Yes, this can be done via meta-reasoning, which I'm currently working on. --b. Should have a repair algorithm based on that higher-order uncertainty Once it is represented at the meta-level, you may do that. But higher-order uncertain reasoning is not high on my priority list... YKY --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=114414975-3c8e69 Powered by Listbox: http://www.listbox.com
Re: [agi] uncertain logic criteria
Prolog is not fast, it is painfully slow for complex inferences due to using backtracking as a control mechanism The time-complexity issue that matters for inference engines is inference-control ... i.e. dampening the combinatorial explosion (which backtracking does not do) Time-complexity issues within a single inference step can always be handled via mathematical or code optimization, whereas optimizing inference control is a deep, deep AI problem... So, actually, the main criterion for the AGI-friendliness of an inference scheme is whether it lends itself to flexible, adaptive control via -- taking long-term, cross-problem inference history into account -- learning appropriately from noninferential cognitive mechanisms (e.g. attention allocation...) -- Ben G On Wed, Sep 17, 2008 at 3:00 PM, YKY (Yan King Yin) [EMAIL PROTECTED] wrote: On Thu, Sep 18, 2008 at 1:46 AM, Abram Demski [EMAIL PROTECTED] wrote: Speaking of my BPZ-logic... 2. Good at quick-and-dirty reasoning when needed Right now I'm focusing on quick-and-dirty *only*. I wish to make the logic's speed approach that of Prolog (which is a fast inference algorithm for binary logic). --a. Makes unwarranted independence assumptions Yes, I think independence should always be assumed unless otherwise stated -- which means there exists a Bayesian network link between X and Y. --b. Collapses probability distributions down to the most probable item when necessary for fast reasoning Do you mean collapsing to binary values? Yes, that is done in BPZ-logic. --c. Uses the maximum entropy distribution when it doesn't have time to calculate the true distribution Not done yet. I'm not familiar with max-ent. Will study that later. --d. Learns simple conditional models (like 1st-order markov models) for use later when full models are too complicated to quickly use I focus on learning 1st-order Bayesian networks. I think we should start with learning 1st-order Bayesian / Markov. I will explore mixing Markov and Bayesian when I have time... 3. Capable of repairing initial conclusions based on the bad models through further reasoning --a. Should have a good way of representing the special sort of uncertainty that results from the methods above Yes, this can be done via meta-reasoning, which I'm currently working on. --b. Should have a repair algorithm based on that higher-order uncertainty Once it is represented at the meta-level, you may do that. But higher-order uncertain reasoning is not high on my priority list... YKY --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?; Powered by Listbox: http://www.listbox.com -- Ben Goertzel, PhD CEO, Novamente LLC and Biomind LLC Director of Research, SIAI [EMAIL PROTECTED] Nothing will ever be attempted if all possible objections must be first overcome - Dr Samuel Johnson --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=114414975-3c8e69 Powered by Listbox: http://www.listbox.com
Re: [agi] uncertain logic criteria
On Wed, Sep 17, 2008 at 1:46 PM, Abram Demski [EMAIL PROTECTED] wrote: Hi everyone, Most people on this list should know about at least 3 uncertain logics claiming to be AGI-grade (or close): --Pie Wang's NARS Yes, I heard of this guy a few times, who happens to use the same name for his project as mine. ;-) Here is my list: 1. Well-defined uncertainty semantics (either probability theory or a well-argued alternative) Agree, and I'm glad that you mentioned this item first. 2. Good at quick-and-dirty reasoning when needed --a. Makes unwarranted independence assumptions --b. Collapses probability distributions down to the most probable item when necessary for fast reasoning --c. Uses the maximum entropy distribution when it doesn't have time to calculate the true distribution --d. Learns simple conditional models (like 1st-order markov models) for use later when full models are too complicated to quickly use As you admitted in the following, the language is biased. Using theory-neutral language, I'd say the requirement is to derive conclusions with available knowledge and resources only, which sounds much better than quick-and-dirty to me. 3. Capable of repairing initial conclusions based on the bad models through further reasoning --a. Should have a good way of representing the special sort of uncertainty that results from the methods above --b. Should have a repair algorithm based on that higher-order uncertainty As soon as you don't assume there is a model, this item and the above one become similar, which are what I called revision and inference, respectively, in http://www.cogsci.indiana.edu/pub/wang.uncertainties.ps The 3 logics mentioned above vary in how well they address these issues, of course, but they are all essentially descended from NARS. My impression is that as a result they are strong in (2a) and (3b) at least, but I am not sure about the rest. (Of course, it is hard to evaluate NARS on most of the points in #2 since I stated them in the language of probability theory. And, opinions will differ on (1).) Anyone else have lists? Or thoughts? If you consider approaches with various scope and maturity, there are much more than these three approaches, and I'm sure most of people working on them will claim that they are also general purpose. Interested people may want to browse http://www.auai.org/ and http://www.elsevier.com/wps/find/journaldescription.cws_home/505787/description#description Pei --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=114414975-3c8e69 Powered by Listbox: http://www.listbox.com
Re: [agi] uncertain logic criteria
--- On Wed, 9/17/08, Abram Demski [EMAIL PROTECTED] wrote: Most people on this list should know about at least 3 uncertain logics claiming to be AGI-grade (or close): --Pie Wang's NARS --Ben Goertzel's PLN --YKY's recent hybrid logic proposal It seems worthwhile to stop and take a look at what criteria such logics should be judged by. So, I'm wondering: what features would people on this list like to see? How about testing in the applications where they would actually be used, perhaps on a small scale. For example, how would these logics be used in a language translation program, where the problem is to convert a natural language sentence into its structured representation and convert it back in another language. How easy is it to populate the database with the gigabyte or so of common sense knowledge needed to provide the context in which natural language statements are interpreted? (Cyc proved it is very hard). For a lot of the problems where we actually use structured data, a relational database works pretty well. However it is nice to see proposals that deal with inconsistencies in the database better than just reporting an error. -- Matt Mahoney, [EMAIL PROTECTED] --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=114414975-3c8e69 Powered by Listbox: http://www.listbox.com
Re: [agi] uncertain logic criteria
On Wed, Sep 17, 2008 at 9:00 PM, YKY (Yan King Yin) [EMAIL PROTECTED] wrote: --a. Makes unwarranted independence assumptions Yes, I think independence should always be assumed unless otherwise stated -- which means there exists a Bayesian network link between X and Y. Small question... aren't Bbayesian network nodes just _conditionally_ independent: so that set A is only independent from set B when d-separated by some set Z? So please clarify, if possible, what kind of independence you assume in your model. Kind regards, Durk Kingma The Netherlands --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=114414975-3c8e69 Powered by Listbox: http://www.listbox.com
Re: [agi] uncertain logic criteria
Good point, this applies to me as well (I'll let YKY answer as it applies to him). I should have said conditional independence rather than just independence. --Abram On Wed, Sep 17, 2008 at 4:21 PM, Kingma, D.P. [EMAIL PROTECTED] wrote: On Wed, Sep 17, 2008 at 9:00 PM, YKY (Yan King Yin) [EMAIL PROTECTED] wrote: --a. Makes unwarranted independence assumptions Yes, I think independence should always be assumed unless otherwise stated -- which means there exists a Bayesian network link between X and Y. Small question... aren't Bbayesian network nodes just _conditionally_ independent: so that set A is only independent from set B when d-separated by some set Z? So please clarify, if possible, what kind of independence you assume in your model. Kind regards, Durk Kingma The Netherlands --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ 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/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=114414975-3c8e69 Powered by Listbox: http://www.listbox.com
Re: [agi] uncertain logic criteria
YKY, Thanks for the reply. It seems important to me to be able to do more than just the fast reasoning. When given more time, a reasoning method should reconsider its independence assumptions, employ more sophisticated models, et cetera. By the way, when I say markov model I mean markov chain as opposed to markov network-- should have been more clear. In that context, 1st-order means conditioned on 1 past item. So when I say 1st-order model, I mean something like: a model that records conditional probabilities conditioned only on 1 thing. (So I might know the probability of winning the election given the fact of being male, and the probability given the fact of being over age 30, but to calculate the probability given *both*, I'd have to assume that the effects of each were independent rather than asking my model what the combined influence was.) These models allow facts to be combined fairly quickly, but are wrong in cases where there are combined effects (such as adding sugar makes it nice, adding salt makes it nice, but adding both makes it awful). 2nd-order means condition on only 2 items, and so on. Anyway, my vision is something like this: we first learn very simple (perhaps 1st or 2nd order) models, and then we learn corrections to those simple models. Corrections are models that concentrate only on the things that the simple models get wrong. The system could learn a series of better and better models, each consisting of corrections on the previous. Thus the system reasons progressively, first by the low-order conditional model, then by invoking progressive corrections that revise conclusions. So, what I really would like would be a formal account of how this should be done; exactly what kind of uncertainty results from using the simple models, how is it best represented, and how is it best corrected? Conditional independence assumptions seem like the most relevant type of inaccuracy; collapsing probabilities down to boolean truth values (or collapsing higher-order probabilities down to lower-order probabilities), and employing max-entropy assumptions, are runner-ups. --Abram On Wed, Sep 17, 2008 at 3:00 PM, YKY (Yan King Yin) [EMAIL PROTECTED] wrote: On Thu, Sep 18, 2008 at 1:46 AM, Abram Demski [EMAIL PROTECTED] wrote: Speaking of my BPZ-logic... 2. Good at quick-and-dirty reasoning when needed Right now I'm focusing on quick-and-dirty *only*. I wish to make the logic's speed approach that of Prolog (which is a fast inference algorithm for binary logic). --a. Makes unwarranted independence assumptions Yes, I think independence should always be assumed unless otherwise stated -- which means there exists a Bayesian network link between X and Y. --b. Collapses probability distributions down to the most probable item when necessary for fast reasoning Do you mean collapsing to binary values? Yes, that is done in BPZ-logic. --c. Uses the maximum entropy distribution when it doesn't have time to calculate the true distribution Not done yet. I'm not familiar with max-ent. Will study that later. --d. Learns simple conditional models (like 1st-order markov models) for use later when full models are too complicated to quickly use I focus on learning 1st-order Bayesian networks. I think we should start with learning 1st-order Bayesian / Markov. I will explore mixing Markov and Bayesian when I have time... 3. Capable of repairing initial conclusions based on the bad models through further reasoning --a. Should have a good way of representing the special sort of uncertainty that results from the methods above Yes, this can be done via meta-reasoning, which I'm currently working on. --b. Should have a repair algorithm based on that higher-order uncertainty Once it is represented at the meta-level, you may do that. But higher-order uncertain reasoning is not high on my priority list... YKY --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ 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/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=114414975-3c8e69 Powered by Listbox: http://www.listbox.com
Re: [agi] uncertain logic criteria
Pei, You are right, that does sound better than quick-and-dirty. And more relevant, because my primary interest here is to get a handle on what normative epistemology should tell us to conclude if we do not have time to calculate the full set of consequences to (uncertain) facts. It is unfortunate that I had to use biased language, but probability is of course what I am familiar with... I suppose, though, that most of the terms could be roughly translated into NARS? Especially independence, and I should hope conditional independence as well. Collapsing probabilities can be restated as generally collapsing uncertainty. Thanks for the links. The reason for singling out these three, of course, is that they have already been discussed on this list. If anybody wants to point out any others in particular, that would be great. --Abram On Wed, Sep 17, 2008 at 3:54 PM, Pei Wang [EMAIL PROTECTED] wrote: On Wed, Sep 17, 2008 at 1:46 PM, Abram Demski [EMAIL PROTECTED] wrote: Hi everyone, Most people on this list should know about at least 3 uncertain logics claiming to be AGI-grade (or close): --Pie Wang's NARS Yes, I heard of this guy a few times, who happens to use the same name for his project as mine. ;-) Here is my list: 1. Well-defined uncertainty semantics (either probability theory or a well-argued alternative) Agree, and I'm glad that you mentioned this item first. 2. Good at quick-and-dirty reasoning when needed --a. Makes unwarranted independence assumptions --b. Collapses probability distributions down to the most probable item when necessary for fast reasoning --c. Uses the maximum entropy distribution when it doesn't have time to calculate the true distribution --d. Learns simple conditional models (like 1st-order markov models) for use later when full models are too complicated to quickly use As you admitted in the following, the language is biased. Using theory-neutral language, I'd say the requirement is to derive conclusions with available knowledge and resources only, which sounds much better than quick-and-dirty to me. 3. Capable of repairing initial conclusions based on the bad models through further reasoning --a. Should have a good way of representing the special sort of uncertainty that results from the methods above --b. Should have a repair algorithm based on that higher-order uncertainty As soon as you don't assume there is a model, this item and the above one become similar, which are what I called revision and inference, respectively, in http://www.cogsci.indiana.edu/pub/wang.uncertainties.ps The 3 logics mentioned above vary in how well they address these issues, of course, but they are all essentially descended from NARS. My impression is that as a result they are strong in (2a) and (3b) at least, but I am not sure about the rest. (Of course, it is hard to evaluate NARS on most of the points in #2 since I stated them in the language of probability theory. And, opinions will differ on (1).) Anyone else have lists? Or thoughts? If you consider approaches with various scope and maturity, there are much more than these three approaches, and I'm sure most of people working on them will claim that they are also general purpose. Interested people may want to browse http://www.auai.org/ and http://www.elsevier.com/wps/find/journaldescription.cws_home/505787/description#description Pei --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ 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/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=114414975-3c8e69 Powered by Listbox: http://www.listbox.com