Peter Norvig


So far we know of exactly one system in which trillions of pieces of information can be intelligently transmitted to billions of learners: the system of publishing the written word. No other system, artificial or otherwise, can come within a factor of a million for successful communication of information. This is despite the fact that the written word is notoriously ambiguous, ill-structured, and prone to logical inconsistency or even fallacy.

In the early days of AI, most work was on creating a new system of transmission—a new representation language, and/or a new axiomatization of a domain. This was necessary because of the difficulty of extracting useful information from available written words. One problem was that people tend not to write down the obvious; as Lenat and Feigenbaum put it “each of us has a vast storehouse of general knowledge, though we rarely talk about any of it ... Some examples are: ‘water flows downhill’ ...”. This is undeniably true; if you look at a 1000-page encyclopedia, there is no mention of “water flows downhill”. But if you look at a 8 billion page web corpus, you find about one in a million pages mention the phrase, including some quite good Kindergarten lesson plans.

This suggests that one future for AI is “in the middle” between author and reader. It will remain expensive to create knowledge in any formal language (Project Halo suggests $10,000/page) but AI can leverage the work of millions of authors of the written word by understanding, classifying, prioritizing, translating, summarizing and presenting the written word in an intelligent just-in-time basis to billions of potential readers.