Talk:Backward chaining

Untitled
I changed The example to reflect changes I made to Forward chaining. --CH

1. If Fritz croaks and eats flies - Then Fritz is a frog 2. If Fritz is a frog - Then Fritz is green

The conclusion in the 5th paragraph: Fritz croaks and eats flies, so must be green; Fritz is green, so must be a frog.

This is inconsistent with the rules of inference. Did you mean "Fritz croaks and eats flies, so must be a frog; Fritz is frog, so must be a green"? --DL

This example is just wrong - it is forward, not backward chaining.

Why is the backward chaining example the same as the forward chaining example if they are different?

This article isn't very clear.

134.225.254.250 08:32, 23 April 2007 (UTC)


 * Some of the Prolog material online contains academic discussion on this topic. Unless someone beats me to it, I'll try to look it up again for stub improvement. Hotfeba 19:00, 28 July 2007 (UTC)


 * I think it's pretty clear. And the example should be the same because the difference is not in the rules but in the inference algorithm using it. Goal-driven is backward (working from the conclusion back to the antecedent) and data-driven is forward (working from antecedent to conclusion). I only have an issue with the analogy used with "top-down" and "bottom-up" -- IMO data driven is bottom up (because you start with the nitty-gritty detail and end at an abstraction) and goal driven is top-down (begin with abstraction and work your way to the detail.) Think of language parsers as special examples of it: if you work the grammar rules from sentence down to parts and try to match words, you do top-down parsing, goal-driven. If you start with the words finding rules which match the sequence you see, then you do bottom-up parsing, data driven. My other concern is that the rules should not mention Fritz but a variable. I will change that. Gschadow 15:41, 29 September 2007 (UTC)

Merge
The article is about computer science terms. If anything, they should be merged to Expert system. WLU 22:07, 30 April 2007 (UTC)

Oppose the Merge
It's a close one. I'm usually for merging, it seems like everyone wants to create a Wikipedia article but far fewer want to make them good. Sorry, I'm editorializing. Anyway, I actually came to this article thinking perhaps it should be merged. I just finished what was essentially a complete rewrite of the article on Expert systems. But as I look at this article it looks pretty good (surprisingly so actually a lot of the AI articles were not so great) and goes into more detail than I did or want to in the articles on Expert systems, knowledge-based systems, etc. So I recommend we keep it. RedDog (talk) 23:48, 4 December 2013 (UTC)

fritz may be a canary
The part of the explanation of the example where it is concluded that Fritz is a frog mentions "and not a canary". I'm pretty sure that we can't actually prove Fritz isn't a canary. We, as humans, intuitively know that something cannot be both a canary and a frog, but there is no rule to that effect in the knowledge base, thus the algorithm cannot conclude it. I think that phrase should be deleted. I'm not going to do it myself because I'm not 100% positive I'm right... 71.88.110.253 (talk) 20:44, 18 May 2008 (UTC)
 * I'm not 100% certain either but I think you are wrong. I think I know what you are referring to, it's a variant of the Frame problem and I think if we were talking about First Order Logic assertions or a theorem prover you would be correct. However, inference engines are not theorem provers, they are less powerful (there are deductions a theorem prover can make that an inference engine can't) but as a result they are a lot faster and more certain to complete. I think in this instance we would use the close world assumption that if we don't know X is true we assume it's false. Although come to think about it if I'm right about that it really is irrelevant that Joe is a frog. Still think it makes sense to keep but willing to be convinced otherwise. RedDog (talk) 23:56, 4 December 2013 (UTC)

Game Theory and Backward Chaining?
I've taken an overview class on Game Theory and I've used many expert system shells but I don't ever recall hearing anyone say that what you do by working backward in game theory is "backward chaining". I do see the point though, I actually remember when I was taking the Game Theory class and the professor was talking about it I thought "oh its analogous to backward chaining in an inference engine" but I don't recall ever seeing that mentioned in a book or article on either topic and if not I think we should remove it. I notice there are some refs here I need to check out perhaps they support it. RedDog (talk) 23:39, 4 December 2013 (UTC)
 * So I don't have the Russel and Norvig book but I did find the Computational Graphs book on Google books. There seemed like there might be a section of that book that talked about game theory but I couldn't tell because those pages were ones that Google didn't include. So I'm going to leave the statement as is for now. RedDog (talk) 00:01, 5 December 2013 (UTC)

Primates?
In the intro it says "but it [backward chaining] has also been observed in primates". There are no references for that and I doubt that the two existing references talk about primates, they are both computer science books. I think it's just takes us too far afield for no reason to leave that statement in and will delete it. RedDog (talk) 00:10, 5 December 2013 (UTC)

Assessment comment
Substituted at 19:47, 1 May 2016 (UTC)