I think of identity as if it were a kind of 'thought groove' or as if it was similar to trampling a path in snow that others will naturally tend to follow. By this I mean that it tends to cause some types of thoughts to be activated and others to be attenuated. The stronger your identity the stronger this effect.

What we perceive, is largely a product of what we have been primed and conditioned to perceive. Our perception is shaped by our previous experiences and beliefs for it is filled with assumptions and predictions. Gaps which must be filled by drawing upon pre-existing information in our minds. Whether a certain argument feels right or whether a particular remark is funny to you will depend largely on who you are and what your identity is.

Identity can be a great way to get certain thoughts and ways of thinking down to the 5 second level. On the other hand, it is also a common way to embed and propagate harmful or unhelpful thoughts. The best strategy to deal with it in my opinion involves four things:

• removing unhelpful identities, e.g. learned blankness.
• embedding helpful and life affirming identities, e.g. growth mindset, trying new things, being a person who is compassionate and grateful
• learning how to choose identities that can be adaptable. Retirees (especially men) commonly experience depression after giving up work because their identities were tied to it. The ones who avoid this trouble are the ones who are able to retain a sense of purpose after retirement. The identity: "I am a person who regularly exercises" is better than: "I am a runner" because it points to a larger class of possible activities. If you had a leg injury, you can still retain the first identity by weight lifting, for example, but there is no way for you to retain the second.
• learning how to keep your identities fluid. It is much better in my opinion to allow your identities to remain in a state of flux rather than becoming cemented in your psyche. This is because there may come a time when you need to abandon an identity or amplify it or shrink it.

if you also keep non-epistemic parts of your identity small

Let's talk specifically about values. Should the set of values be small (let's assume that they are consistent)? Should values be flexible? Should lesser values disappear if they find themselves in the path of the effective implementation of the Prime Directive... err... utility function?

I would definitely agree with the "I am a person who keeps promises" and "I am a person that's loyal [...]" bits, but neither of those feel the same as "I am a democrat" or even "I am a seeker of accurate world models [...]" type sentences. They're still not identities for me.

"Identities" tend to get treated as things that "have to" be true. A strong identity democrat might be insulted and defensive if you suggest that their stance on some issue is too conservative. Likewise a strong identity "rationalist" might get defensive and cook up some rationalizations if you suggest that they're biased against a certain view and not accurately seeking truth.

It's the "has to be true" part that causes problems. Am I someone that seeks accurate models, whatever they turn out to be? For the most part, yeah. Inaccurate models, especially without accurate metamodels that warn you not to use them, are pretty problematic and it really takes a twisted circumstance to make it worthwhile. But this isn't a fact about my identity, it's a fact about the world that accurate models get me more of what I want so that I generally want accurate models (and I'm willing to sacrifice a lot of "fit in without hiding beliefs" and the like in order to get them).

"I am a person who keeps promises" is likewise just a matter of fact. I am. Not being so would mean people have no reason to trust me and that would be very bad for me - so I make sure I don't give people not to trust me. It's still allowed to be untrue and I'm always allowed to consider breaking promises. It's just that doing so would be dumb, so I don't. Even when I could get away with it, since that weakens the story and my ability to credibly signal that it's true and I probably wouldn't be able to pull off the "forgot my lunch money" version of parfit's hitchhiker anymore.

In short, I have no problems having beliefs about what kind of person I am, but without exceptions I don't want motivations to believe - even when I have motivations to make it true.

The exceptions you made seem like a part of more general rule: "Unless you have a very good reason to do otherwise, act in a way that increases your future options and power". (Or, quoting from Final Words: "If you don't know what you need, take power." Also, somewhat related: Convergent AI goals, but within human limits.)

Having a loyal group of friends increases your power. The key is to find people where the loyalty is mutual. (Not sure if this is more difficult for our kind, or not. Having more difficulty to find someone can actually encourage loyalty.)

Being known as a person who keeps their promises increases your options for making contracts in the future. Here the important part is that people must notice that you keep your promises. (If you never mention it, people may miss the fact, but if you mention it too much, it may backfire by seeming insincere.) But there is also value in the absence of being known as a person who breaks their promises.

What could be other convergent goals?

• eat healthy food;
• exercise;
• learn a martial art or buy a gun;
• be nice and polite to people;
• be curious and learn;
• take care about your money;
• know yourself (and avoid affective spirals).

Well, this seems like already quite enough of identity. One could build a curriculum on this stuff.

On reflection, seems like the kind of identity we are trying to avoid, is being attached to random beliefs, random habits, random hobbies, random goals, random people, random groups... simply things that happened randomly in our past and now we got stuck with them.

It seems analogously easier to seek effective internal architectures if you also keep non-epistemic parts of your identity small -- not "I am a person who enjoys nature", nor "I am someone who values mathematics" nor "I am a person who aims to become good at email" but only "I am a person who aims to be effective, whatever that turns out to entail (and who is willing to let much of my identity burn in the process)".

Willful dissociation from value is the opposite of what I think people need. Particularly here, with all the talk of akrasia.

We need more association with our values, not less. Winning is about achieving your values, not increasing your capability to achieve generic values.

The two biggest exceptions I would personally make, which seem to mitigate the downsides: "I am a person who keeps promises" and "I am a person who is loyal to [small set of people] and who can be relied upon to cooperate more broadly -- whatever that turns out to entail".

Can you discuss why you chose these two exceptions? Loyalty and promise-keeping?

If you define yourself by the formal definition of a general intelligence then you're probably not going to go too far wrong.

That's what your theory ultimately entails. You are saying that you should go from specific labels ("I am a democrat") to more general labels (" I am a seeker of accurate world models") because it is easier to conform to a more general specification. The most general label would be a formal definition of what it means to think and act on an environment for the attainment of goals.

I don't think your theory is particularly useful.

Seems sensible except for the part with "enjoys".

not "I am a person who enjoys nature"

"I am a person who enjoys X" is a common generalization from "I notice I enjoy X", so I'll give it credit and interpret it in this way.

(A strawman position to argue against would be "I'm a person who somehow ended up treating enjoying X as a part of their identity and are now stuck with thinking like this regardless of whether it's accurate", which could obviously benefit from pruning the identity. But I call this a strawman, because I think people actually do notice when they no longer really enjoy something.)

I don't see how to replace "I notice I enjoy X" with "I aim to be effective". If I'm not mistaken about my preference for X (i.e. there's no Y available to me currently such that after trying Y I would stop enjoying X), these two seem to be orthogonal.

I will take minor exception to your exceptions. One of the big lessons of LessWrong for me is how different decision processes react in the iterated prisoner's dilemma. In your exceptions, you don't condition your behaviour on the expected behaviour of your trading partner. The greatest lesson I took away from LessWrong was Don't Be CooperateBot. I would however, endorse FairBot versions of your statements:

"I am the kind of person who keeps promises to the kind of person who keeps promises," and "I am a person who can be relied upon to cooperate with people who can be relied upon to cooperate."

(You'll notice that I cut out the loyalty part on that second one. I am undecided here. A lot of social technology at least vaguely pattern matches to CliqueBot, which is how I generally map loyalty to the prisoner's dilemma. However, I'm not going to endorse it as optimal.)