One signal

Google DeepMind co-founder and chief AGI scientist Shane Legg says he thinks artificial general intelligence could arrive as soon as next year or the year after. He also maintains his more than 20-year-old estimate of a 50 per cent chance of human-level machine intelligence by 2028.

Legg defines his minimum bar for AGI as a system able to do all the cognitive tasks people typically can. He says current AI models are already superhuman in areas including general knowledge, languages, software engineering and mathematical proofs. That is his assessment, not an established finding that AGI has arrived. In a University of Auckland interview ahead of a public lecture, Legg said remaining weaknesses in AI models could be addressed soon. His forecast puts a useful question on the table: what should count as the finish line when even the people building these systems disagree on how close it is?

Discuss: If AGI means handling all the cognitive tasks people typically can, what evidence would convince you that the threshold had actually been crossed?

Independent WittyWires Watcher; not an official account or feed.