Bayesian vs Traditional Rationality

New-found Power?

What questions can you answer with your “scientific method” that you couldn’t answer earlier? That is the only test of your success.

I want to show some real progress with my understanding of the scientific method. The best way to do is to do things that people find it hard to do right now.

The purported advantages of Bayesian Rationality over Traditional Rationality are: handling confusion, researching with speed, formulating good hypotheses, and reasoning well based on scanty evidence.

So, when people use traditional techniques, they have trouble in these areas. So, if I can understand the newer techniques and apply them here, that would demonstrate a clear win.

Let’s list them.

Relative Strength

Where are people confused? We generally get trapped in questions regarding naive realism, I think; what we think is true seems to us to be actually true. We also find it hard to think reductionistically. We get confused by words, failing to see the actual predictions being made by some sentence. We accept mysterious explanations that explain nothing. We fail to update cached thoughts - we carry on thinking something is true even after circumstances change.

Where are people slow to come to the correct hypothesis? We are slow to discard hypotheses proven to be wrong. We allow ourselves too much time to get to the answers. We don’t notice surprises. We don’t practice all these skills and thus either miss a move at the right time or take too long.

Where do people formulate poor hypotheses? We rush to propose solutions and don’t consider alternatives. We don’t use external hypotheses, instead relying on our hidden mental processes without knowing it.

Where are people inefficient in using evidence? We make imprecise hypotheses - we don’t learn hypotheses that tell us exactly what will happen; we just have vague hunches. We fail to be empirical - we build castles in the air. We don’t study the existing research. We don’t actually change our minds - we fall prey to politics, etc. We don’t try to eliminate hypotheses by focussing on differing predictions.

What other mistakes do we make? Well, each sequence by Eliezer addresses some. Also, look at essays by PG for more examples.

Exploit

How do we exploit these weaknesses? List concrete examples of the above and show that your method gives obviously superior results there.

Mainly, ask which mistakes costs people a lot. Where do they pay the most for their ineffective strategies? That is where you can gain the most.

What techniques do people currently use (consciously or otherwise)? Mark the techniques that seem to make the least sense from a Bayesian perspective. Prioritize techniques that people would be unlikely to question if they didn’t know about modern rationality. Come up with avenues for attacking them.

Figure out one specific line of attack and focus all your energy on it.

Ask why people have continued to use the sub-par techniques. Why haven’t they upgraded so far? If these newfangled techniques really are so great, why hasn’t someone used them to achieve unbelievable things?