The problem of other minds

post by Elo · 2018-07-24T01:04:06.396Z · score: 4 (7 votes) · LW · GW · 9 comments

Originally posted at Bearlamp.com.au: http://bearlamp.com.au/the-problem-of-other-minds/ (also Greaterwrong [LW · GW])

I'm trapped in here and you are trapped in there.

It's an odd problem. How do we communicate when we are each locked in our own Chinese room.

In computing terms - a handshake allows for an affirmation of knowledge to be returned. Unfortunately a completed handshake does not confirm shared knowledge. RandomAgent can sometimes confirm a handshake with YesAgent by getting randomly the right 1’s and 0’s to fulfil YesAgent’s criteria for confirmation.

The problem of other minds is that we can't necessarily trust both agents to be using the same communication method. We could have a bit-flip, we could have a different code. We can't assume the method of communication is sound.

This is seen in the Two Generals’ Problem. Where each general cannot trust the other general’s communication.

So what can we do? One solution is a three pass encryption protocol. With this system, each agent self-verifies the message.

But for all that I do, No matter how hard I try to pass the message, I will never know if the same subjective qualia of the colour blue is being seen by you.

Next: The experimental apparatus (greaterwrong)

9 comments

Comments sorted by top scores.

comment by habryka (habryka4) · 2018-07-25T00:43:45.907Z · score: 2 (1 votes) · LW · GW

(Your greaterwrong link goes to LW)

comment by Elo · 2018-07-27T02:32:49.218Z · score: 2 (1 votes) · LW · GW

The only reason it does that is because the lesswrong editor is doing something funky with the link. I have fixed it by opening the post on lesswrong and setting the link straight then saving it via the lesswrong editor.

I don't understand how that happens. Do you have insight into what the editor is doing? I'm already struggling to create several links backwards and forwards. It's just really painful to cross post.

comment by habryka (habryka4) · 2018-07-27T15:39:39.423Z · score: 2 (1 votes) · LW · GW

The greaterwrong editor is changing all links to greaterwrong to go to LessWrong, if I remember correctly. (And on the display side makes it so that all links to LW are displayed as links to greaterwrong, to prevent unnecessary site transitions)

comment by Elo · 2018-07-27T18:15:51.554Z · score: 2 (1 votes) · LW · GW

Yes. That's exhaustingly frustrating when trying to standardise links. I just have to do it the hard way.

comment by habryka (habryka4) · 2018-07-27T20:03:35.485Z · score: 4 (2 votes) · LW · GW

It might make sense to create an issue about this on the greaterwrong github.

comment by Elo · 2018-07-27T23:20:06.666Z · score: 2 (1 votes) · LW · GW

You don't want to work on it. GW's Saturn doesn't want to work on it (as far as I understand)... Why am I making a github issue? So far as I can tell it's my problem and I can't do anything about it unless I learn to code it myself.

comment by habryka (habryka4) · 2018-07-28T00:53:15.309Z · score: 2 (1 votes) · LW · GW

Ah, I didn't know Saturn doesn't want to work on it. This happening has nothing to do with the LessWrong code, so there is nothing I can do about it. Would be happy to fix were it in my control, but our server just receives a batch of HTML that no longer has any greaterwrong links in it.

comment by Gurkenglas · 2018-07-25T00:05:47.860Z · score: 1 (1 votes) · LW · GW

Bit-flipped models are indistinguishable, so don't call them different. "My red is your green" shouldn't mean anything.

Are you okay? This sounds like the cliché train of thought someone would go through on their first drug trip, so if you are just now thinking of it your psychology might be changing in a way to be checked.

comment by Elo · 2018-07-25T00:32:20.253Z · score: 2 (1 votes) · LW · GW

Indistinguishable from the outside, different on the inside potentially. That's a paradox that doesn't collapse.

Yes I'm fine. It's a meditative experience not a drug one, but again, I'm fine and I have it under control.