Hassabis says AGI can come in two to three years
Earlier this year he gave three to five years, and he gives no test result for the new date.
Claimed, not confirmed
Demis Hassabis founded Google DeepMind and is a Nobel laureate. He says AGI can come in two to three years. He told the Wall Street Journal that his estimate gets closer to today. He gave no test result for the new date. AGI, or artificial general intelligence, is a system that matches the full range of the human brain.
His test for AGI
His test is the Einstein test. Can a system that has only knowledge from before 1901 reproduce the 1905 papers of Einstein? He compares the effect of AGI to 10 times the industrial revolution at 10 times the speed.
The proof
The date is his own estimate. The source gives no data and no test result for it. It does not say that a system passes the Einstein test.
Safety and rules
He proposed an oversight board that industry pays for and the government controls. The board can vet models before release. He warns that bad rules are worse than no rules. The source does not name the members of the board.
Insight: Labs ask for controls that they help pay for
Hassabis wants an industry-funded board, controlled by the government, to vet models before release. Persons at Anthropic and OpenAI want a mandatory kill switch. Some think that a large AI incident can occur in 6 to 12 months. This looks like a race to write the rules before an incident makes others write them. The labs can shape the rules. Plan for checks before release and a stop rule.
Sources
Posted
