Wednesday, September 9, 2015

Good news, bad news

My perspective on intelligence is a simple one - it is an automatic, conditioned response based on subjective beliefs rather than objective input. This is relevant to AI since it challenges the mainstream belief about the nature of intelligence. Most people believe that intelligence is had and utilized by the intelligent agent, and that intelligence requires a correct, unbiased processing of information.

This means we must defend the idea of free will, of an intelligent entity that chooses to believe and act in a certain way. The problem AI research then faces is one about safety. If AI has free will, how can it be 100% safe? We might solve that by eliminating its emotions so it has no intrinsic motivation or desires. But without that, why would it be of any help to us? Ok then, let’s build in some motivation to help humans, to be at our service. Now AI is free from emotions and only wants to obey its masters. To me that sounds a lot like weak AI.

But if we challenge the illusion of an agent with free will and agency, we get another perspective on this. If intelligence is automatic, it can be engineered and conditioned. If it is based on subjective, unconscious patterns of belief, there must be a certain configuration to generate such patterns. And if so, there is no need to program AI to recognize all possible novel input as to execute the correct response. It is enough to make it respond to bigger classes/sets of input as if they were a single representation of the whole class/set. We call that prejudice and humans do it a lot.

The following paper is one of many obituaries on free will:
When we consciously experience the making of a decision, it is just the final firing in a neural process that started several seconds earlier. Like it or not, that goes for all of your conscious experience. Consciousness is a dashboard displaying what has just happened. It is not a crucial component of intelligent action.

The next paper explains why you will probably reject the first one:
You are likely to believe that intelligent people are the ones able to correct their misconceptions and flawed conclusions, while stupid people ignore obvious facts and hold onto their “fantasies”. Again, not knowing the nature of intelligence leads us astray. In reality, the more intelligent you are, the less likely will you process input objectively and unbiased. The more you have invested in a certain belief system/theory, the more biased your processing of new data will be. This is why arguing with experts will never lead to progress.

Bottom line is this: when we finally give ourselves up, the designing of intelligence is not about millions of elaborate details and infinite storage. It is about idiosyncratic processing of all data as if it had a certain, predetermined meaning. It is about classifying, valuing and interpreting input so that none of it is perceived as novel. That’s how intelligence avoids the horror of not-knowing novel input while maintaining the ability to create novelty from old input. We re-create input as to create output, and it is all done before we know it.


No comments:

Post a Comment

Respond as it happens here