City Research Online

Abstract

As society contemplates AI’s impact on everyday life and work, the opacity surrounding AI development since the release of ChatGPT con tributes to fears of existential risk and fuels claims of an upcoming AI bubble burst. In this article, I argue that the emerging field of neurosymbolic AI can address the lack of reliability of current AI. Instead of ever increasing compute power, use of chain-of-thought prompting and performing alignment via reinforcement learning, neurosymbolic AI promotes model compression, symbolic knowledge reuse and alignment via knowledge sharing. I discuss how the persisting problem of reliability can be addressed by neurosymbolic AI with the use of formal reasoning, causal inference and extrapolation towards artificial general intelligence.

Publication Type:	Other (Preprint)
Subjects:	Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Departments:	School of Science & Technology School of Science & Technology > Department of Computer Science
SWORD Depositor:	Symplectic Administrator

Export

Downloads