Apr 3 2019
A new paper by Dr Vasile Palade of Coventry University and Dr Gerry Wolff of CognitionResearch.org, has been reported in The Computer Journal. It deals with a roadmap for the progress of the SP Machine, a new type of computer based on the SP Theory of Intelligence—which is a concept of intelligence with benefits compared with current thinking in AI.
Now, a lot of what one hears about artificial intelligence (AI) is about “deep learning” and things that it can achieve such as learning the oriental game of Go so well that it can beat the top-notch human players of the game. This and other accomplishments of deep learning are certainly remarkable but there is increasing worry that there are several restrictions of deep learning as a stepping stone towards the development of human-like AI.
For instance, deep learning systems may deal with from “catastrophic forgetting”, meaning that, when a deep learning system learns something new, that new learning clears anything that the system picked up before.
The SP Theory of Intelligence is a substitute for deep learning that is a lot more promising as a foundation for the development of human-like AI. It has been under development since about 1987 using computer models as a way to convey the theory, test it, and show what it can accomplish.
The SP System has been developed to integrate Simplicity in its structure and how it functions with descriptive and explanatory Power across a comparatively broad range of areas of AI. Therefore it has been named “SP”. It has not yet learned Go but, compared with deep learning, it has assets and potential across a much broader range of areas of AI, including how it learns, numerous types of reasoning, recovery of information from storage, recognizing patterns, making plans, the processing of natural language, and solving problems. Furthermore, the SP System does not suffer from disastrous forgetting or any other of the numerous weaknesses that have been detected in deep learning.
The paper explains how the SP Machine will be developed primarily as software working on a present-day supercomputer, and it will be based primarily on the newest version of the SP Computer Model. Since the essential equipment is currently quite inexpensive, teams of scientists in other places, or individuals, may collaborate in the progress of the SP Machine.