Crossposted from the AI Alignment Forum. May contain more technical jargon than usual.

I've recently thought of a possibly simpler way of expressing the argument from the Occam's razor paper. Namely:

  • Human biases and human preferences contain more combined information than human behaviour does. And more than the full human policy does.

Thus, in order to deduce human biases and preferences, we need more information than the human policy caries.

This extra information is contained in the "normative assumptions": the assumptions we need to add, so that an AI can learn human preferences from human behaviour.

We'd ideally want to do this with as few extra assumptions as possible. If the AI is well-grounded and understands what human concepts mean, we might be able to get away with a simple reference: "look through this collection of psychology research and take it as roughly true" could be enough assumptions to point the AI to all the assumptions it would need.

New Comment
5 comments, sorted by Click to highlight new comments since: Today at 11:39 PM

But is that true? Human behavior has a lot of information. We normally say that this extra information is irrelevant to the human's beliefs and preferences (i.e. the agential model of humans is a simplification), but it's still there.

Look at the paper linked for more details ( ).

Basically "humans are always fully rational and always take the action they want to" is a full explanation of all of human behaviour, that is strictly simpler than any explanation which includes human biases and bounded rationality.

"look through this collection of psychology research and take it as roughly true"

Well, you are an intelligence that "is well-grounded and understands what human concepts mean", do you think that the above approach would lead you to distill the right assumptions?

No. But I expect that it would be much more in the right ballpark than other approaches, and I think it might be refined to be correct.

I suppose the question is whether we can predict the "hidden inner mind" through some purely statistical model, as opposed to requiring some deeper understanding of human psychology of an AI. I'm not sure that a typical psychologist would claim to be able to predict behaviour through their training, whereas we have seen cases where even simple, statistical predictive systems can know more about you than you know about yourself - [1].

There's also the idea that social intelligence is the ability to simulate other people, so perhaps that is something that an AI would need to do in order to understand other consciousnesses - running shallow simulations of those other minds.