Alan Turing's Biggest AI Assumptions May Be Wrong
Could foundational assumptions be leading AI research astray?
1Alan Turing's Foundational Assumptions
In 1950, Alan Turing proposed two foundational assumptions for AI: intelligence could exist independently of a physical body, and a machine could demonstrate intelligence by conversing like a human. These ideas, known as the Turing test, have shaped AI research.
2The Importance of Tacit Knowledge
Peter J. Denning argues that tacit knowledge, the unspoken understanding humans possess, is a major hurdle for AI. This includes emotions, practical skills, and cultural knowledge, which machines cannot capture.
3Machine Learning and Its Limits
Denning highlights that machine learning cannot capture tacit knowledge. For example, a virtuoso musician cannot explain their performance. Such skills require a biological body and emotional connection, making them impossible for machines to replicate.
4The Representation Problem
Denning explains that computers can only process encoded data and instructions, but tacit knowledge cannot be represented this way. Words are symbolic representations of meanings, not the meanings themselves, posing a significant challenge for AI.
5The Role of Context and Culture
Denning emphasizes that intelligence is shaped by context and culture. Humans understand social cues and emotions through context, but AI systems cannot capture this knowledge, preventing them from achieving human-like thought.
6AI Safety and the Future
Denning concludes that AI systems cannot understand human tacit knowledge, creating serious risks for AI safety. He argues that there is an insurmountable knowledge gap between humans and machines.