What is Artificial Superintelligence?
Defining the basics: part III
Note: I’m writing a number of small definitional posts to refer to later. Readers of this publication might already be familiar with the basic terms.
In two previous posts, I have offered definitions for intelligence, Artificial Intelligence (AI) and Artificial General Intelligence (AGI). Now it’s time to define the (even) more powerful kind of AI: Artificial Superintelligence (ASI).
It goes without saying that an ASI must be more intelligent than any AGI, but the question remains: how much more?
It seems to me that for any given intellectual task, an ASI should be able to learn to do that task at at least the level of the best humans for that task, and if possible, at a higher level. (If the best human is actually perfect at a certain task, it’s obviously impossible to do better.)
So, an ASI should be able to learn to do physics at least as well as, say, Albert Einstein, to learn to make a movie at least as popular as The Shawshank Redemption, to learn to win (virtually) any Formula 1 race, etc.
An ASI is an AI that can do or learn to do any intellectual task to at least the level of the best humans for that task.
