| Summary: | This paper highlights 20 significant problems in AI research, with potential solutions via the <i>SP Theory of Intelligence</i> (SPTI) and its realisation in the <i>SP Computer Model</i>. With other evidence referenced in the paper, <i>this is strong evidence in support of the SPTI as a promising foundation for the development of human-level broad AI, aka artificial general intelligence</i>. The 20 problems include: the tendency of deep neural networks to make major errors in recognition; the need for a coherent account of generalisation, over- and under-generalisation, and minimising the corrupting effect of ‘dirty data’; how to achieve one-trial learning; how to achieve transfer learning; the need for transparency in the representation and processing of knowledge; and how to eliminate the problem of catastrophic forgetting. In addition to its promise as a foundation for the development of AGI, the SPTI has potential as a foundation for the study of human learning, perception, and cognition. And it has potential as a foundation for mathematics, logic, and computing.
|
|---|