Provably Safe AGI is Potentially a Very Dangerous Concept

Published on May 23, 2024

The article reviews the paper "Toward Guaranteed Safe AI" by Davidad and others, including notable figures like Yoshua Bengio and Max Tegmark. The paper advocates for integrating symbolic and subsymbolic AI to enhance intelligence, ethics, transparency, and efficiency. While acknowledging the value of pursuing provably safe AI, the article criticizes the paper for rehashing old ideas and underestimating practical challenges. It warns that the push for formal safety guarantees might lead to stringent regulations, restricting AGI development to large corporations and governments. The author argues for a decentralized, democratic approach to AGI development, cautioning against allowing the pursuit of provable safety to justify hierarchical control, which could stifle innovation and favour corporate and governmental dominance over AGI research.

Read the full article here.