TITLE Information content of human intelligence and life AUTHOR Warren D. Smith DATE May 2006 ABSTRACT Regard various standard approximate biochemical assumptions as correct. Then we show that the proteome of human cells constitutes an ``information bottleneck.'' We use this to estimate the information content of a ``blueprint'' sufficient for efficiently building a human intelligence (as opposed to some unintelligent animal). We find it is only about 7 or 95 Kbytes (in two oppositely-extreme models), with 4 times these figures being proposed as a high confidence upper bound. This seems surprisingly small compared to many software attempts to create artificial intelligences. However, in the companion paper we create a ``mathematical definition of intelligence,'' use it to prove a theorem that a ``universally asymptotically competitive intelligence'' (UACI) exists, and show how to implement it as a remarkably short program. (Although mathematically a success, this UACI is not a success in any engineering sense because it will take a very long time before doing anything interesting.) Also Fong's PAPPI human language principles package and Baum \& Durdanovic's ``Hayek'' AI-via-artificial-economy code both lie between these two numbers. Hence this may not really be a paradox despite initial appearances. The tasks of creating an AI and of creating artificial life may share the similarity of requiring a very large search over configurations to find a good one.