sal@COLUMBIA-20.ARPA@sri-unix.UUCP (03/22/84)
From: Sal Stolfo <sal@COLUMBIA-20.ARPA> "The Numericists Meet the Symbolicists and Ask Why?" With the recent interest in Fifth Generation Computing and Artificial Intelligence, many scientists with backgrounds in other disparate fields are beginning to study symbolic computation in a serious manner. The ``parallel architectures community'' has mostly been interested in novel computer architectures to accelerate numeric computation (usually represented as Fortran codes). Similarly, the ``data base machine community" has been interested in more conventional data processing (for example, large-scale data bases). Now that the interest of these communities and others are focusing on Artificial Intelligence computing, a question that is often asked is ``What are the fundamental characteristics of AI computation that distinguish it from more conventional computation"? Indeed, are there really any differences at all? These questions have no simple answers; they can be viewed from many different perspectives. This note is a solicitation of the AI community for cogent discussion of this issue. We hope that all facets will be addressed including: - Differences between the kinds of problems encountered in AI and those considered more conventional. (A simple answer in terms of ``ill-defined'' and ``well-defined'' problems is viewed as a copout.) - Methodological differences between AI computing and conventional computing. - Computer resource requirements and programming environments with technical substantiations of the differences rather than aesthetic preferences. I expect to collect responses from the AI community and produce a final report which will be made available to any interested parties. Thank you in advance. Salvatore J. Stolfo Assistant Professor Computer Science Department Columbia University
dsn%umcp-cs.csnet@csnet-relay.arpa (03/30/84)
From: Dana S. Nau <dsn%umcp-cs.csnet@csnet-relay.arpa> From: Sal Stolfo <sal@COLUMBIA-20.ARPA> This note is a solicitation of the AI community for cogent discussion ... We hope that all facets will be addressed including: - Differences between the kinds of problems encountered in AI and those considered more conventional. (A simple answer in terms of ``ill-defined'' and ``well-defined'' problems is viewed as a copout.) ... One of the biggest differences involves how well we can explain how we solve a problem. The problems that humans can solve can be divided roughly into the following two classes: 1. Problems which we can solve which we can also explain HOW to solve. Examples include sorting a deck of cards, adding a column of numbers, and payroll accounting. Any time we can explain how to solve a problem, we can write a conventional computer procedure to solve it. 2. Problems which we can solve but cannot explain how to solve (for a discussion of some related issues, see Polanyi's "The Tacit Dimension"). Examples include recognizing a face, making good moves in a chess game, and diagnosing a medical case. We can't solve such problems using conventional programming techniques, because we don't know what algorithms to use. Instead, we use various heuristic approaches. The latter class of problems corresponds roughly to what I would call AI problems.
dsn@umcp-cs.UUCP (03/30/84)
- - From: Sal Stolfo <sal@COLUMBIA-20.ARPA> This note is a solicitation of the AI community for cogent discussion ... We hope that all facets will be addressed including: - Differences between the kinds of problems encountered in AI and those considered more conventional. (A simple answer in terms of ``ill-defined'' and ``well-defined'' problems is viewed as a copout.) ... One of the biggest differences involves how well we can explain how we solve a problem. The problems that humans can solve can be divided roughly into the following two classes: 1. Problems which we can solve which we can also explain HOW to solve. Examples include sorting a deck of cards, adding a column of numbers, and payroll accounting. Any time we can explain how to solve a problem, we can write a conventional computer procedure to solve it. 2. Problems which we can solve but cannot explain how to solve (for a discussion of some related issues, see Polanyi's "The Tacit Dimension"). Examples include recognizing a face, making good moves in a chess game, and diagnosing a medical case. We can't solve such problems using conventional programming techniques, because we don't know what algorithms to use. Instead, we use various heuristic approaches. The latter class of problems corresponds roughly to what I would call AI problems. -- Dana S. Nau ...!seismo!umcp-cs!dsn (Usenet) dsn.umcp-cs@CSNet-Relay (Arpanet)