On Intelligence, by Jeff Hawkins (2004). Some thoughts.

P. 104 “To know something means that you can make predictions about it.”

Assume knowing something comes with the fact we can make predictions about it, but does it work the other way around? Does making predictions about something mean that one knows it? Also, knowing something does not have to come with the ability to make predictions. Can this “knowledge about something” be measured differently?

I know I have a cup of tea on my table. I can make a prediction that I will either drink the tea or more likely forget about it, as always. Does it tell you much about me knowing something about this cup of tea? What about knowing someone as a person?

Defining knowing via predictions only is limiting and does not tell us what actually knowing means.

P. 97 “Intelligence is measured by the capacity to remember and predict patterns in the world, including language, mathematics, physical properties of objects, and social situations.”

I like to follow the definition of Rich Sutton [more here] that starts with John McCarthy’s one: Intelligence is the computational part of the ability to achieve goals. A goal achieving system is one that is more usefully understood in terms of outcomes than in terms of mechanisms.

This definition comes with other questions: what do we mean by goals? If we can align on the definition of goals, then perhaps, we could align on how we could test intelligence. Rich’s well known example is about thermostat: would we still deem intelligent a thermostat that is graciously achieving the goal of keeping the right temperature in a room?