Branches of Artificial Intelligence (AI)

In the last post I talked about Basics of Artificial Intelligence  https://panthimanshu17.wordpress.com/2013/07/25/artificial-intelligence-and-machine-learning-fundamentals-2/

In this post I will discuss about branches of Artificial Intelligence. As AI is continuously evolving it is very difficult to classify branches of Artificial Intelligence and the problem aggravates due to the fact that many branches of AI are overlapping in nature. Still I will try to classify them as neatly as possible.

Primarily there are three core branches  Artificial Intelligence.

  1. Symbolic AI:  the work in the field of symbolic AI is nearly abandoned, and reduced to mainly course books. In early stages of developments (1960s) symbolic AI  tasted tremendous success with expert systems and game playing problems, but in 1980s the research nearly exhausted due to the lack of some implicit design issues in the formulation itself and it was assumed symbolic AI would never be able to achieve human cognition level. One major implicit design issue was “General Knowledge problem” or “common sense problem”. It means designers were able to mimic explicit human behavior into the machine but implicit common sense which we never say and take for granted which forms the base of explicit behavior was not conceptualized at the design stages, for example if I say that “she is my mother” it implicitly means that “I am her son” etc. CYC (CyCorp) project aims to beat this common sense problem.For more information on CYC project you can follow the link :    http://www.cyc.com/time-digital.html
  2. Statistical AI: Statistical AI advocates deterministic approach in Artificial Intelligence taking inspiration from mathematics and operation research. critics argue that this approach looses capability of generalization and hence ultimate aim of Artificial Intelligence.
  3. Computational Intelligence: Computational Intelligence aims to solve real world problems that are computationally expensive or not at all possible to solve by traditional means (mathematical models). The guiding principle of soft computing  is exploit the tolerance for imprecision, uncertainty and partial truth to achieve tractability, robustness and low cost solution with improved adaptability.Common branches of Computational Intelligence  are but not limited to.
    1. Artificial Neural Networks
    2. Fuzzy Logic
    3. Heuristic Search
    4. Pattern Recognition

Well thats all for now. in coming posts I will be discussing more about Computational Intelligence and its branches.

Advertisements

4 thoughts on “Branches of Artificial Intelligence (AI)

  1. From the brief descriptions of the 3 categories given above, “Computational Intelligence” looks very much like a subsection of “Statistical AI”. The most popular form of “expert system” today is probably the Bayesian Network, which is more “Statistical” than “Symbolic”, even though the nodes in the network mostly have high level descriptions.

    Insofar as symbolic AI uses rules, and rules can be hard or soft, there is also no hard boundary between Symbolic and Statistical AI. Ultimately all digital computation is symbolic because it deals only with discrete bits, but those bits can be used to represent high level symbols like words, or sense data like pixels or audio samples, and in the later case the number of rules required would be too large to be useful.

    The accuracy of automatically generated statistical grammars for natural language generally increases as you allow the number of non terminals and production rules to increase. This suggests that manually generated grammars from pre-computational linguistics were an oversimplification of what is an inherently deeper problem.

    The problem with oversimplified statistical models is that their generalisation power is very limited, even with access to astronomical amounts of training data. The key is therefore to uncover as many “prior” constraints on model structure for the problem in hand as possible, preferably without overly restricting the domain in which the statistical model can be useful.

  2. Hi Andrew, Thanks for the expert comments on the post. As I have already mentioned in the beginning of the post that it is very hard to classify Branches of AI due to its overlapping nature. Well its true that Statistical AI more or less forms the basis of computational intelligence.
    The basic aim of Expert systems is to mimic the reasoning of human expert.
    Other than Bayesian Belief Networks , Fuzzy Inference System (FIS) is also strong candidate for implementation of Expert systems which is a branch of computational intelligence.So contradictions are bound to emerge when we try to classify Branches of Artificial Intelligence.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s