Some neural networks can exist, but cannot be trained to be reliable, according to mathematicians at the Universities of Cambridge and Oslo. “The paradox identified by Turing and Gödel has now been brought forward into the world of AI” Matthew Colbrook (right) “Many AI systems are unstable, and it’s becoming a major liability, especially as they ...

This story continues at Some neural networks will always be unreliable

Or just read more coverage at Electronics Weekly