LessWrong 2.0 Reader

I'm afraid in a situation of power imbalance these interpersonal differences won't matter much. I'm thinking of examples like enclosures in England, where basically the entire elite of the country decided to make poor people even poorer, in order to enrich themselves. You can't point to individual personalities responsible for this. Or colonialism, which lasted for centuries and genocided or enslaved entire continents. Lots and lots of people participated, and the good people among the dominant group didn't stop it. What should we think of someone like Leopold II, who was a benevolent ruler in Belgium and at the same time a horrible tyrannical ruler of the Congo Free State? You can't say he was inherently good or evil, you can't blame ideology or fanaticism, he simply acted one way toward his people and a different way toward outsiders. This attitude was shared by most Europeans for centuries, and just seems normal for people overall.

To be clear, I'm not saying there are no interpersonal differences. But I am saying that if we find ourselves at the bottom of a power imbalance, the biggest danger to us will be systemic, not due to interpersonal differences.

The brain needs to observe something (sense, interoception) from which it can infer this. The pattern in what observations would that be? 

(partly copying from my other comment [LW(p) · GW(p)]) For example, consider the following fact.

FACT: Sometimes, I’m thinking about pencils. Other times, I’m not thinking about pencils.

Now imagine that there’s a predictive (a.k.a. self-supervised) learning algorithm which is tasked with predicting upcoming sensory inputs, by building generative models. The above fact is very important! If the predictive learning algorithm does not somehow incorporate that fact into its generative models, then those generative models will be worse at making predictions. For example, if I’m thinking about pencils, then I’m likelier to talk about pencils, and look at pencils, and grab a pencil, etc., compared to if I’m not thinking about pencils. So the predictive learning algorithm is incentivized (by its predictive loss function) to build a generative model that can represent the fact that any given concept might be active in the cortex at a certain time, or might not be.

See also §1.4 [LW · GW].

That's why I have long thought that there has to be a feedback from the current thought back as input signal (thoughts as observations). Such a connection is not present in the brain-like model [? · GW], but it might not be the only way. Another way would be via memory.

I mean yeah obviously the cortex has various types of memory, and this fact is important for all kinds of things. :)

Clearly, action A can happen without S(A) being present. In fact, actions are often more effectively executed if you don't think too hard about them[citation needed]. An S(A) is not required. Maybe S(A) and A cooccur often, but that doesn't imply causality.

These sentences seem to suggest that either A’s are either always, or never, caused by a preceding S(A), and out of those two options, “never” is more plausible. But that’s a false dichotomy. I propose that sometimes they are and sometimes they aren’t caused by S(A).

By analogy, sometimes doors open because somebody pushed on them, and sometimes doors open without anyone pushing on them. Also, it’s possible for there to be a very windy day where the door would open with 30% probability in the absence of a person pushing on it, but opens with 85% probability if somebody does push on it. In that case, did the person “cause” the door to open? I would say yeah, they “partially caused it” to open, or “causally contributed to” the door opening, or “often cause the door to open”, or something like that. I stand by my claim that the self-reflective S(standing up), if sufficiently motivating, can “cause” me to then stand up, in that sense.

The fact that “actions are often more effectively executed if you don't think too hard about them” is referring to the fact that if you have a learned skill, in the form of some optimized context-dependent temporal sequence of motor-control and attention-control commands, then self-reflective thoughts can interrupt and thus mess up that temporal sequence, just as people shouting random numbers can disrupt someone trying to count, or how you can’t sing two songs in your head simultaneously. A.k.a. the limited capacity of cortex processing. Whereas that section is more about whether some course-of-action (like saying something, wiggling your fingers, standing up, etc.) starts or not.

Flow states (post 4) are a great example of A’s happening without any S(A).

So what's missing?

I haven't looked at any of the studies and also don't know much about genomics so my guess might be completely wrong, but a different hypothesis that seems pretty plausible to me is:

Most of the variance of intelligence comes from how well different genes/hyperparamets-of-the-brain can work together, rather than them having individually independent effects on intelligence. Aka e.g. as made-up specifc implausible example (I don't know that much neuroscience), there could be different genes controlling the size, the snapse-density, and the learning/placticity-rate of cortical columns in some region and there are combinations of those hyperparameters which happen to work well and some that don't fit quite as well.

So this hypothesis would predict that we didn't find the remaining genetic component for intelligence yet because we didn't have enough data to see what clusters of genes together have good effects and we also didn't know in what places to look for clusters.

I've been thinking about this in the back of my mind for a while now. I think it lines up with points Cory Doctorow has made in talks about enshittification. 

I'd like to see recommendation algorithms which are user-editable and preferably platform-agnostic, to allow low switching costs. A situation where people can build their own social media platform and install a recommendation algorithm which works for them, pulling in posts from other users across platforms who they follow. I've heard that the fediverse is trying to do something like this, but I've not been able to get engaged with it yet. 

It's cool to see efforts like Tournesol, though it's a shame they don't have a mobile extension yet.

Personally, I read approximately everything you (Elizabeth) write on the Forum and LW, and occasionally cite it to others in EA leadership world. That's why I'm pretty sure your work has had nontrivial impact. I am not too surprised that its impact hasn't become apparent to you though.

[...]

I don't see solutions or great ways forward yet, and I sense that nobody really does

That does sound like learned helplessness and that the EA leadership filters people out who would see ways forward.

Let me give you one:

If people in EA would consider her critiques to have real value, then the obvious step is to give Elizabeth money to write more. Given that she has a Patreon the way to give her money is pretty straightforward. If the writing influences what happens in EV board discussions, paying Elizabeth for the value she provides for the board would be straightforward. 

If she would get paid decently, I would expect she would feel she's making an impact. 

Paying Elizabeth might not be the solution to all of EA's problems, but it's a way to signal priorities. Estimate the value she provides to EA and then pay her for that value and publically publish as EV a writeup that EV thinks that this is the amount of value she provides to EA and was paid by EV. 

I like this post. I made similar observations and arguments in The Translucent Thoughts Hypotheses and Their Implications [LW · GW] but these were less clear and buried at the end of the post.

Your definition of the Heaviside step function has H(0) = 1.
Your definition of L has L(0) = 1/2, so you're not really taking the derivative of the same function.

I don't really believe nonstandard analysis helps us differentiate the Heaviside step function. You have found a function that is quite a lot like the step function and shown that it has a derivative (maybe), but I would need to be convinced that all functions have the same derivative to be convinced that something meaningful is going on.  (And since all your derivatives have different values, this seems like a not useful definition of a derivative)

Alternatively

Still . . .

The early signs have been promising.

What concrete things did he change at CEA that are promising signs?

If I say that other psychiatrists at the conference are engaging in an ethical lapse when they charge late fees to poor people then I'm engaging in an uncomfortable interpersonal conflict. It's about personal incentives that actually matter a lot to the day-to-day practice of psychiatry. 

While the psychiatrists are certainly aware of them charging poor people, they are likely thinking about it normally as business as usual instead of considering it as an ethical issue. 

If we take Scott's example of psychiatrists talking about racism being a problem in psychiatry I don't think the problem is that that racism is unimportant. The problem is rather that you can get points by virtue signaling talking about the problem and find common ground around the virtue signaling if you are willing to burn a few scapegoats while talking about the issues of charging poor people late fees is divisive. 

Washington DC is one of the most liberal places in the US with people who are good at virtue signaling and pretending they care about "solving systematic racism" yet, they passed a bill to require college degrees for childcare services. If you apply the textbook definition of systematic racism, requiring college degrees for childcare services is about creating a system that prevents poor Black people to look after children. 

Systematic racism that prevents poor Black people from offering childcare services is bad but the people in Washington DC are good at rationalising. The whole discourse about racism is of a nature where people score their points by virtue signaling about how they care about fighting racism. They practice steelmanning racism all the time and steelmanning the concept of systematic racism and yet they pass systematic racist laws because they don't like poor Black people looking after their children. 

If you tell White people in Washington DC who are already steelmanning systematic racism to the best of their ability that they should steelman it more because they are still inherently racist, they might even agree with you, but it's not what's going to make them change the laws so that more poor Black people will look after their children. 

That tactic helps reduce ignorance of the "other side" on the issues that get the steelmanning discussion

If you want to reduce ignorance of the "other side", listening to the other side is better than trying to steelman the other side. Eliezer explained problems with steelmanning well in his interview with Lex Friedmann.

Also, in judging a strategy, we should know what resources we assume we have (e.g. "the meetup leader is following the practice we've specified and is willing to follow 'reasonable' requests or suggestions from us"), and know what threats we're modeling.

Yes, as far as resources go, you have to keep in mind that all people involved have their interests.

When it comes to thread modelling reading through Ben Hoffman's critique of GiveWell based on his employment at it, give you a good idea of what you want to model.