Friendship, Evolution And “Evil AI” Cliches

It seems logical that mutual friendship could have been produced by evolution – the “I help you, you help me”, contract-style relationship can be seen as beneficial even from the “selfish” PoV or evolution. What caught me by surprise was the idea that true friendship – perseverant, everlasting despite circumstances – is also a product of “survival of the fittest”.

I came accross this theory when re-reading Creating friendly AI

Since humans live in a world full of unknowns and unexpected reversals, the most adaptive friendship instinct was probably that one which urged a friendship with a certain amount of built-in context insensitivity. […] Another, more widely recognized evolutionary advantage of context-insensitive friendship is that, all else being equal, a prospective ally will choose to become friends with a context-insensitive friend rather than a “fair weather” friend.

I could agree with that.

So this post is basically a heartfelt recommendation and a fumbling compliment to the excellence of the CFAI website. Being the armchair AI enthusiast that I am, I felt the need to spread the word and hope that some of the people who read this will enjoy the site like I did. There’s some interesting discussion about human cognition, evolution and artificial intelligence.

I would especially like to draw your attention to the second chapter (PDF) that does a good job of dispelling some common Hollywood-style Evil AI myths. In particular, it explains how selfishness is an evolved trait, produced by specific factors inherent in the evolution of living organisms. This leads to an important consequence – an artificial intellect would not be inherently selfish/observer-centric (note : it would not be “altruistic” in the typical sense either). Well, one could program it to be selfish, but the trait would not arise spontaneously.

Why is this important? Because lots of common arguments against AI are based on the implicit assumption that the AI would be selfish and observer-biased. Two classical examples –

  • “The AI will come to hate us if we make it work for us and enslave it.” – The unselfish AI would not be resentful about being made to work for others, as it would have no inherent, personal, selfish desires that would be slighted by this. An AI would simply not consider it’s “self” inherently important.
  • “If some fool decides to punch the AI in the nose, the AI will respond by exterminating all humankind.” – retaliation is a product of evolution, something that’s hardwired into the selfish human brain. CFAI argues that it would be very hard for the AI to independently invent retaliation. Even if it did, it would have no “punch him back!” instinct like people do.

The CFAI paper does a better job of explaining this stuff, go read skim it! 🙂

Related posts :

Leave a Reply