Talk:Computational intelligence

From Wikipedia, the free encyclopedia

Seems like this article had some disagreements coming in over the weekend. This involves the extent and breakdown of the topic. A quick link check produced the following:

  • Bioinformatics or computational biology is about the sovling of biological problems
  • Bioengineering deals with bio-molecular and molecular processes, and includes biomedical- food- and agricultural engineering
  • Autonomous mental development is the investigating of evolution through computer programs
  • Computational finance is a form of finance which relies on mathematical methods
  • Computational economics is the branch of applied mathematics concerned with the financial markets
  • Intelligent Systems is a video game developer team of Nintendo Co., Ltd.
  • Emergence is the process of complex pattern formation from simpler rules (occurring over time or over disparate size scales)
  • Data mining is the extraction of implicit, previously unknown, and potentially useful information from data, used with a varied meaning

These topics are not subsets of CI, although they might be related. Machine learning and Expert systems should be removed from 'related topics' as is explained at the end of the first article paragraph. Lets keep the disagreements to the discussion. --moxon 17:43, 31 October 2005 (UTC)

Contents

[edit] Contradiction, confusion

I don't get it: First ci "either rejects fuzzy systems or ignores neural networks", then both are listed as part of ci. ??? And more generally, saying what the two sides (ci and ml) reject is not a clear way to explain what they are. This ci beginner needs help. Thx, "alyosha" 03:01, 20 December 2005 (UTC)

I edited it to try and solve the disambiguation: fuzzy rejects and neurals ignores stats. The deffinition of CI and ML is quite fuzzy, at the moment I think examples and differences are the simplest way to define then. They are 2 different classical approached to AI. ML builds heavily on statistics, CI does not. This is the most obvious difference. A rough definition might be "learning from empirical data". --moxon 15:27, 20 December 2005 (UTC)

[edit] Merge?

Do we really need a seperate article for a specific arbitrary, mechanical distinction on ways to achieve weak AI? How about this article gets redirected to a new article on the computational aproaches to AI? - Jake11...

Strong/Weak AI is a philosophical issue or matter of opinion and has no impact on the techniques followed. CI denotes distinct methods opposing conventional/neat AI. See the IEEE CI Soc. as referenced. This is an existing and growing branch of computer science & engineering and needs no redirection. --moxon 16:27, 16 March 2006 (UTC)

[edit] Not alternative to AI, etc.

Computational intelligence is a subset of AI. It is an alternative to GOFAI and the 'neats'. In addition Evolutionary computation is not equal to Genetic algorithms and Fuzzy logic is not an algorithm. --moxon 10:19, 16 January 2007 (UTC)

[edit] Is CI a subfield of AI?

The answer is: the question is controversial.
It comes down to the sources. Here is a survey of the sources given for this article:
This is not how it is used in the article's first reference: (Poole et al. 1998). They say: "Artificial Intelligence is the established name for the field we have defined has Computation Intelligence" on page 1. So they think of it as new name for the old field.
The standard AI textbook Artificial Intelligence: A Modern Approach doesn't seem to mention computational intelligence at all. We could assume that they don't think that the distinction made in the article is interesting enough to merit discussion.
Reading the IEEE CIS web site seems to support CI as a subfield, but only if one reads between the lines: knowledge based systems, for example, is missing from the list, as is logic programming.
The Computational Intelligence Group from Amsterdam sees CI a new subfied. So there's that.
The journal Computation Intelligence's description reads "This leading international journal promotes and stimulates research in the field of artificial intelligence (AI)." So this journal description explicitly does not recognize CI as a subfield.
The International Journal of Computational Intelligence lists the same areas as this article, and as before, logic programming, expert systems and knowledge based systems are all missing. So this would seem to support the idea of CI as subfield, but again, this is a supposition after the fact.
The Journal of Intelligence Technology doesn't seem to be aware of computational intelligence; it shouldn't be connected to this article at all.
The International Journal of Computational Intelligence Research treats CI as a subfield. So there's also that.
And finally, CIRG (which has the nicest website) gives it's definition in complete line with the idea of this as subfield.
So the answer (for Wikipedia) must be: the subject is controversial, and this article should take a neutral point of view. There seems to be (at least) two uses of the term. (1) as a new name for AI as it is practiced today, in contrast to the old (and failed) ways of doing AI. (2) as a new sub-field of AI, gathering together some approaches and ignoring others. The article should mention that different researchers, journals and research groups use the word differently.
As an editorial choice, I would not expand this article beyond making this point, since any additional material should be covered in the article on AI, since many people still consider this material as part of AI. (Unfortunately, the article on AI does need a lot of tuning up). ---- CharlesGillingham 03:28, 5 September 2007 (UTC)

I would add a third use of the term: (3) a synonym for AI. I'd say that Computational Intelligence uses CI in such a way.

Regards, --zeno 11:48, 3 October 2007 (UTC)

[edit] Is CI "scuffy" or "neat"?

Answer: It can be either.

First a source: Russell & Norvig, describing modern approaches to AI, write "recent years have seen a revolution in the content and the methodology of work in AI" and "some have characterized this change as a victory of the 'neats'" and later that "neural networks also fit into this trend." They are implicitly saying: neural networks are neat.

The precise mathematics used in genetic algorithms and neural networks can be very 'neat'. And these are clearly in CI's area.

Also, the sloppy, ad hoc symbolic AI epitomized by Doug Lenat's Cyc is 'scruffy', but it's good old fashioned symbolic AI, and is not in CI's area.

The neat/scruffy distinction is orthogonal to the GOFAI/CI distinction: there are cases of neat GOFAI (logic programming), scruffy GOFAI (semantic webs), neat CI (pattern matching neural networks), and scruffy CI.

If there are sources that disagree with this analysis, I'd love to see them. ---- CharlesGillingham 03:28, 5 September 2007 (UTC)