KDM2520@TAMSIGMA.BITNET (08/28/88)
Date: Thu, 25 Aug 88 22:21 EDT From: KDM2520%TAMSIGMA.BITNET@MITVMA.MIT.EDU Subject: Expert Systems for Statistical Analysis To: AILIST@AI.AI.MIT.EDU X-Original-To: AILIST@AI.AI.MIT.EDU, KDM2520 If anyone on the list is aware of any commercial expert systems which do statistical analysis, I sure would appreciate the information. I am looking for expert statistical analysis packages which process data, analyze correlation etc., forecast trends, detect regeneration cycles, and so on. I have looked through AI Magazine, IEEE Expert and several Computer journals and I couldn't see any such product advertisements. Could someone who is aware of such expert statistical analysis packages send me the info please? Thank you. MURALI@TAMLSR (bitnet)
cik@L.CC.PURDUE.EDU (Herman Rubin) (08/30/88)
Date: Mon, 29 Aug 88 12:19 EDT From: Herman Rubin <cik@l.cc.purdue.edu> To: ailist-request@kl.sri.com Path: l.cc.purdue.edu!cik From: Herman Rubin <cik@l.cc.purdue.edu> Newsgroups: comp.ai.digest Subject: Re: Expert Systems for Statistical Analysis Summary: Do not ask for the impossible. Date: Mon, 29 Aug 88 08:24 EDT References: <19880828002614.0.NICK@HOWARD-JOHNSONS.LCS.MIT.EDU> Organization: Purdue University Statistics Department Lines: 35 In a previous article, KDM2520@TAMSIGMA.BITNET writes: > If anyone on the list is aware of any commercial expert systems which do > statistical analysis, I sure would appreciate the information. I am looking > for expert statistical analysis packages which process data, analyze > correlation etc., forecast trends, detect regeneration cycles, and so on. > I have looked through AI Magazine, IEEE Expert and several Computer journals > and I couldn't see any such product advertisements. Could someone who is > aware of such expert statistical analysis packages send me the info please? > > Thank you. MURALI@TAMLSR (bitnet) There are things that a computer is capable of doing, but this is not one of them. Statistics is not a black box into which one can put data and come out with the state of the universe. To analyze a problem, it is necessary for the user to input a model, or better, a collection of models. The user must realize that many assumptions must be made. It is advisable to have a good mathematical statistician available to point out the consequences of the model which the user does not realize have been inserted. Analyze correlation indeed! It is extremely rare that correlation has anything to do with the real problem. In addition, the user must have an evaluation of the consequences of an incorrect action. Massive statistical uncertainty may be irrelevant if the resulting action is unaffected, and small uncertainty may be very important if the effects of a wrong action are sufficiently great. I personally have worked on this problem, and the difficulties are major. Statistical packages of the type to use this input, when well formulated, are still in the development stage. -- Herman Rubin, Dept. of Statistics, Purdue Univ., West Lafayette IN47907 Phone: (317)494-6054 hrubin@l.cc.purdue.edu (Internet, bitnet, UUCP)