I agree with the general point that idea generators are often overconfident and poorly calibrated, but I'm not super excited about conflating that with being disagreeable. There's something there, in that being impatient with the failings of others can be a good impetus to create something better, but I think there are also plenty of generators who aren't highly disagreeable, and if that's the case it seems bad for the dichotomy as formulated here to catch on.

One example archetype of a generator who isn't disagreeable is someone who is both very intelligent and very enthusiastic/excitable. This person will generate lots of cool new ideas they're super excited about, and will likely still be overconfident and poorly calibrated, without feeling particularly motivated to yell about other people's bad ideas.

Disclaimer: I have disagreeable tendencies, working on it but biased. I think you're getting at something useful, even if most people are somewhere in the middle. I think we should care most about the outliers on both sides because they could be extremely powerful when working together.

I want to add some **speculations** on these roles in the context of the level at which we're trying to achieve something: individual or collective.

When no single agent can understand reality well enough to be a good principal, it seems most beneficial for the collective to consist of modestly polarized agents (this seems true from most of the literature on group decision-making and policy processes, e.g. Adaptive Rationality, Garbage Cans, and the Policy Process | Emerald Insight).

This means that the EA network should want people who are confident enough in their own world views to explore them properly, who are happy to generate new ideas through epistemic trespassing, and to explore outside of the Overton window etc. Unless your social environment productively reframes what is currently perceived as "failure", overconfidence seems basically required to keep going as a disagreeable.

By nature, overconfidence gets punished in communities that value calibration and clear metrics of success. Disagreeables become poisonous as they feel misunderstood and good assessors become increasingly conservative. The succesful ones of the two characters build up different communities in which they are high status and extremize one another.

To succeed altogether, we need to walk the very fine line between productive epistemic trespassing and conserving what we have.

Disagreeables can quickly lose status with assessors because they seem insufficiently epistemically humble or outright nuts. Making your case against a local consensus costs you points. Not being well calibrated on what reality looks like costs you points.

If we are in a sub-optimal reality, however, effort needs to be put into defying the odds and change reality. To have the chutzpah to change a system, it helps to ignore parts of reality at times. It helps to believe that you can have sufficient power to change it. If you're convinced enough of those beliefs, they often confer power on you in and of themselves.

Incrementally assessing baseline and then betting on the most plausible outcomes also deepens the tracks we find ourselves on. It is the safe thing to do and stabilizes society. Stability is needed if you want to make sure coordination happens. Thus, assessors rightly gain status for predicting correctly. Yet, they also reinforce existing narratives and create consensus about what the future could be like.

Consensus about the median outcome can make it harder to break out of existing dynamics because the barrier to coordinating such a break-out is even higher when everyone knows the expected outcome (e.g. odds of success of major change are low).

In a world where ground truth doesn't matter much, the power of disagreeables is to create a mob that isn't anchored in reality but that achieves the coordination to break out of local realities.

Unfortunately, to us who have insufficient capabilities to achieve their aims - to change not just our local social reality but the human condition - creating a cult just isn't helpful. None of us have sufficient data or compute to do it alone.

To achieve our mission, we will need constant error correction. Plus, the universe is so large that information won't always travel fast enough, even if there was a sufficiently swift processor. So we need to compute decentrally and somehow still coordinate.

It seems hard for single brains to be both explorers and stabilizers simultaneously, however. So as a collective, we need to appropriately value both and insure one another. Maybe we can help each other switch roles to make it easier to understand both. Instead of drawing conclusions for action at our individual levels, we need to aggregate our insights and decide on action as a collective.

As of right now, only very high status or privileged people really say what they think and most others defer to the authorities to ensure their social survival. At an individual level, that's the right thing to do. But as a collective, we would all benefit if we enabled more value-aligned people to explore, fail and yet survive comfortably enough to be able to feed their learnings back into the collective.

This is of course not just a norms questions, but also a question of infrastructure and psychology.

I was writing a LessWrong post on what I called "generators" and "evaluators" of ideas a few years ago, but never finished it. "Evaluators"  and "assessors" seem similar. The main difference was that I didn't postulate that "generators" are disagreeable, unlike your "disagreeables".

This post by the cognitive scientist Hugo Mercier may also be of relevane: "Why assholes are more likely to be wrong".

Great post! 

I love Taleb but he embarrasses me, so it's nice to find a dialectical role for him in my head.

I think I've met a few people who can do both modes, and they're all deeply impressive people. What in late Jobs do you see as sane assessment? Always seemed like a wild half-charlatan half-prophet to me.

Thanks for the post - I can see what you're getting at, but this doesn't feel like two clearly distinct categories to me. The first person I thought to try and apply this to had strong traits from both columns, for example. As a similar but more available example, where would you fit Bryan Caplan here? He's disagreeable without being angry, and is trying hard not to be wrong while happily telling others why they are. 

I'm not sure whether my intuition here is that these can both be strong/weak in the same person, that there's more of a spectrum, or that they're a set of characteristics that may or may not cluster the way you've described. I'm not really sure what shape you meant for this to take, or how well it applies in these intermediate cases.

The distinction reminds me of the foxes vs hedgehogs model from Superforecasting / Tetlock. Hedgehogs being "great idea thinkers" seeing everything in the light of that one great idea they're following, whereas foxes are more nuanced, taking in many viewpoints and trying to converge on the most accurate beliefs. I think he mentioned in  the book that while foxes tend to make much better forecasters, hedgehogs are not only more entertaining but also good in coming up with good questions to forecast in the first place.

An entirely different thought: The Laws of Human Nature by Robert Greene was the first audible book I returned without finishing. It was packed with endless "human archetypes" described in great detail, making some rather bold claims about what "this type" will do in some given situation. You mention in the footnotes already that people who dislike e.g. personality profiling tools might not like this post. And it did indeed somewhat remind me of that book, but maybe your "assessor" way of describing the model, as opposed to Greene's very overconfident seeming way of writing, made this seem much more reasonable. There seems to be a fine line between actually useful models of this kind which have some predictive power (or at least allow thoughts to be a bit tidier), and those that are merely peculiarly entertaining, like Myers-Briggs. And I find it hard to tell from the outside on which side of that line any given model falls. 

Scott Garrabrant has discussed this (or some very similar distinction) in some LessWrong comments. There's also been a lot of discussion about babble and prune, which is basically the same distinction, except happening inside a single mind instead of across multiple minds.