More problems in AI research and how the SP System may help to solve them
Year: 2020
Type: report
Abstract: This technical report, an adjunct to the paper "Problems in AI research ...", describes some problems in AI research and how the SP System (meaning the SP Theory of Intelligence and its realisation in the SP Computer Model) may help to solve them. It also contains a fairly detailed outline of the SP System. Most of the problems considered in this report are described by leading researchers in AI in interviews with science writer Martin Ford, and presented in his book "Architects of Intelligence". Problems and their potential solutions that are described in this report are: the need for more emphasis in research on the use of top-down strategies is met by the way SP has been developed entirely within a top-down framework; the risk of accidents with self-driving vehicles may be minimised via the theory of generalisation within the SP System; the need for strong compositionality in the structure of knowledge is met by processes within the SP Computer Model for unsupervised learning and the organisation of knowledge; although commonsense reasoning and commonsense knowledge are challenges for all theories of AI, the SP System has some promising features; the SP programme of research is one of very few working to establishing the key importance of information compression in AI research; Likewise, the SP programme of research is one of relatively few AI-related research programmes attaching much importance to the biological foundations of intelligence; the SP System lends weight to 'localist' (as compared with 'distributed') views of how knowledge is stored in the brain; compared with deep neural networks, the SP System offers much more scope for adaptation and the representation of knowledge; reasons are given for why the important subjects of motivations and emotions have not so far been considered in the SP programme of research. Evidence in this report, and "Problems in AI research ...", suggests that the SP System provides a relatively promising foundation for the development of artificial general intelligence.
Source: HAL
Author: Wolff, James Gerard, Gerry
Institution: University of Freiburg
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W3044620863