Last we left off, we listed the shortcomings of Eliezer's level of rationality that the next level of rationality would have to deal with, we focused on "how to collaborate as rationalists or be more rational as a pair or a group", you suggested that Socratic Dialogues are an important tool for this purpose, and we planned to start with "a Socratic dialogue about how to do a Socratic dialogue". 

I'll let you start :)

What do you know about Socratic dialogues?

That it is a dialogue where one side asks the other side questions that challenge their claim to knowledge. It is basically asking someone the Fundamental Question of Rationality over and over again.

That's not a bad place to start. If you're new to Socratic dialogue, then following the rule "just ask questions" can very well get the job done.

Performed artfully, questions don't always end in a question mark. They often take the form of a joke, a parable, or even a statement of fact.

Right, but the goal is always to sow some doubt or challenge the claim the other side makes, so they have to reconsider it or make their position more sophisticated/nuanced.

When encountering over-confidence, a Rationalist's duty is to sow doubt. When encountering under-confidence, a Rationalist's duty is to sow confidence.

Yep, I agree with that, and I think the second part was missing in the original Socratic Dialogues.

Indeed. Plato's works are a curated sample. It could be that those dialogues just weren't written down. Or perhaps Socrates just didn't encounter any under-confidence.

Probably a combination of the two, and also because they had so little capacity to know at the time relative to today, levels of confidence that we would consider low today would have probably been considered high by Socrates.

And I think Nassim Taleb's critique is also correct: that just because you don't know how to justify your knowledge doesn't mean you don't know, but Socrates equivocated the two. 

They knew the Sun rose in the east. They knew how to grow wheat. I don't know how to grow wheat.

The inverse is true too: just because you can justify your claims doesn't mean they're correct.

Right, I mean as a society. I don't know how to grow wheat as well, but surely there's many people today that know how to grow wheat at least as good as the ancient Greeks, if not better.

And indeed, your justifications may always be wrong, you can never have absolute certainty.

How certain are you that $1+1=2$?

Principally, not 100%, practically, basically 100% to the point where the difference doesn't really matter.

Why is Socratic dialogue so important to the art of rationality?

Because it asks the fundamental question of rationality. But the importance of the dialogue, is having outside input.

What is the fundamental question of rationality? Why is outside input important?

"Why do you believe what you believe?", or alternatively, "What do you think you know, and how do you think you know it?".

Outside input is important because you can't always see your the flaws in your reasoning on your own. There are some directions your mind doesn't go in, some moves it doesn't make, that you don't even know it doesn't. Conversing with another person can show you these blind spots.

What are the flaws in your reasoning? What are the directions your mind doesn't go, the moves it doesn't make?

I don't know :)

What are the flaws in my reasoning? What are the directions my mind doesn't go, the moves it doesn't make?

I don't know. Maybe if I knew you better I would.

Then let's get to know each other better. :)

What's asymmetric about this is (I think) I do see flaws in your reasoning. (I think) I do see directions your mind doesn't go, moves it doesn't make.

Do share.

You said the purpose of Socratic dialogue was to sow doubt. I pointed out that the reverse is also true. You pointed out that that just because you don't know how to justify your knowledge doesn't mean you don't know. I pointed out that the inverse is also true. This is a pattern.

What is the pattern?

Good question. It seems obvious to me, yet I'm having trouble putting it into words.

Socrates would have said that it means you don't really know ����������������

Socrates was wrong about many things.

Indeed.

Many things are about balance. Confidence and doubt. Selfishness and generosity. Courage and discretion.

When someone praises yin, you fail to ask about yang. When someone praises yang, you do not immediately think of yin.

I think I am pretty good at doing that actually. I very often give considerations from the other side in a debate, and I often think about the idea of reversing every piece of advice. To use your language, I don't think I tend to focus just on one part of the yin-yang. I do seek to calibrate and find the sweet spot (or dynamically move along the scale).

I think in this case I was giving the classic conception of Socratic Dialogue I'm familiar with, and then I was pointing out the thing that seemed less obvious / more insightful to me.

[Chuckles.]

How do you know you're good at reversing advice? Why do you think you know it?

"Good" is relative, and it's not clear which scale we're using, but to me it seems that I do it often.

I used to tutor physics. Students would come in to my table because they needed help with their homework.

"Did you attend class?" I'd ask.

"Yes," they'd say.

"Did you do the reading?" I'd ask.

"Yes," they'd say.

"Do you understand how to do this problem?" I'd ask.

"Yes," they'd say.

"Can you do this problem?" I'd ask.

"Yes," they'd say.

"Great! Now do this problem," I'd say.

They'd pick up their pencil, confident, and then freeze. "I don't know how," they'd say.

This works in the domain of physics, where it is easy to test whether someone can perform a skill. How does one perform the test in the domain of rationality?

Ah, that's the (LessWrong) age old question. How do we verify rationality? It's an important problem, but we haven't made much progress on it. (It should be added to the list from last time).

There are many ways to verify rationality. But some cannot be done directly. I can't just say "you're bad at reversing advice". Why? Because…

• When you encounter advice and reverse it, you notice.
• When you encounter advice and fail to reverse it, you don't notice.

Counting only the situations you notice, it feels to you like you reverse advice all the time. I don't think you actually have a problem with reversing advice. I'm just using this as an example.

This is why we use Socratic dialogue, instead.

It's correct that I don't notice when I encounter advice and fail to reverse it, but if I encounter about the same amount of advice as others, and reverse more total advice than others, then I also reverse a higher ratio than others.

Have you figured out what advantage Socratic dialogue has in solving this problem, compared to regular dialogue?

Yes, asking me how I know instead of telling me why I'm wrong is more productive (Especially If I'm not wrong!).

Why?

Because when asking questions we can both understand my position and the reasons I believe it better. If you just tell me the reasons you think I'm wrong, then we never actually engage the same question together. 

That's one reason. There are many layers to this game.

But let's go back to question "How do we verify rationality?" How do you show that someone's rationality is broken? One way is to hack them. You ask them leading questions designed to produce nonsensical output. You do it until the person notices they're emitting nonsensical output. Then they fix themselves.

• If I say "You tend to err in $x$ direction," then a normal person will say "No I don't."
• But if that person has a predictable error, then I can trick them into making a noticeable mistake, and they will self-correct.

One way is to hack them

You could also say... Troll them :)