X
Business

Singularity Summit 2007: Google co-evolving with the Web

Google Director of Research Peter Norvig opened the second day of the Singularity Summit 2007 with his take on the state artificial general intelligence (AGI).Norvig was asked during a Q&A after his talk if Google has seen any "emergent" property or behavior (the way complex systems and patterns arise out of a multiplicity of relatively simple interactions) as it amasses data and manages massive infrastructure that has been a surprise.
Written by Dan Farber, Inactive

Google Director of Research Peter Norvig opened the second day of the Singularity Summit 2007 with his take on the state artificial general intelligence (AGI).

Norvig was asked during a Q&A after his talk if Google has seen any "emergent" property or behavior (the way complex systems and patterns arise out of a multiplicity of relatively simple interactions) as it amasses data and manages massive infrastructure that has been a surprise.

"We were all surprised at how game theoretic it is," Norvig. "We made a copy of the Web and indexed it and we thought it reflected of Web. Now we understand that we are in co-evolution. When we make a move, the Web changes and when Web changes we do. Optimizers look at what we do and we look at what they do and Web moves in different directions because of the interaction between them. We hadn't expected that."

File photo

He also cited communications between computers at low level, such unexpected behavior in balancing bandwidth between switches.

Norvig doesn't expect AGIs to suddenly appear overnight. The U.S. Gross Domestic Product index, for example, shows constant exponential progress but it doesn't seem like any great acceleration can be seen or predicted due to technology, Norvig said. The same could be said for percentage annual growth worldwide.

For AGIs to appear, and we only recognize them with some hindsight, several components need to in place, he said:

  • Probablistics first-order logic
  • Hierarchical representation and problem solving--to go from pixels to lines to faces to people in a crowd in a hierarchical fashion
  • Learning over the above
  • With lots of data
  • Online and continuously updated
  • Efficiency

Editorial standards