This might feel obvious, but I think it’s under-appreciated how much disagreement on AI progress just comes down to priors (in a pretty specific way) rather than object-level reasoning.
I was recently arguing the case for shorter timelines to a friend who leans longer. We kept disagreeing on a surprising number of object-level claims, which was weird because we usually agree more on the kinda stuff we were arguing about.
Then I basically realized what I think was going on: she had a pretty strong prior against what I was saying, and that prior is abstract enough that there’s no clear mechanism by which I can push against it. So whenever I made a good object-level case, she’d just take the other side — not necessarily because her reasons were better all else equal, but because the prior was doing the work underneath without either of us really knowing it.
There’s something clearly rational here that’s kinda unintuitive to get a grip on. If you have a strong prior, and someone makes a persuasive argument against it, but you can’t identify the specific mechanism by which their argument defeats it, you should probably update that the arguments against their case are better than they appear, even if you can’t articulate them yet. From the outside, this totally just looks like motivated reasoning (and often is), but I think it can be pretty importantly different.
The reason this is so hard to disentangle is that (unless your belief web is extremely clear to you, which seems practically impossible) it’s just enormously complicated. Your prior on timelines isn’t an isolate thing — it’s load-bearing for a bunch of downstream beliefs all at once. So the resistance isn’t obviously irrational, it’s more like… the system protecting its own coherence.
I think this means that people should try their best to disentangle whether some object level argument they’re having comes from real object level beliefs or pretty abstract priors (in which case, it seems less worthwhile to press on them).
[Graham’s hierarchy of disagreements] is useful for its intended purpose, but it isn’t really a hierarchy of disagreements. It’s a hierarchy of types of response, within a disagreement. Sometimes things are refutations of other people’s points, but the points should never have been made at all, and refuting them doesn’t help. Sometimes it’s unclear how the argument even connects to the sorts of things that in principle could be proven or refuted.
If we were to classify disagreements themselves – talk about what people are doing when they’re even having an argument – I think it would look something like this:
Most people are either meta-debating – debating whether some parties in the debate are violating norms – or they’re just shaming, trying to push one side of the debate outside the bounds of respectability.
If you can get past that level, you end up discussing facts (blue column on the left) and/or philosophizing about how the argument has to fit together before one side is “right” or “wrong” (red column on the right). Either of these can be anywhere from throwing out a one-line claim and adding “Checkmate, atheists” at the end of it, to cooperating with the other person to try to figure out exactly what considerations are relevant and which sources best resolve them.
If you can get past that level, you run into really high-level disagreements about overall moral systems, or which goods are more valuable than others, or what “freedom” means, or stuff like that. These are basically unresolvable with anything less than a lifetime of philosophical work, but they usually allow mutual understanding and respect.
More on the high-level generators of disagreement (emphasis mine, other than 1st sentence):
High-level generators of disagreement are what remains when everyone understands exactly what’s being argued, and agrees on what all the evidence says, but have vague and hard-to-define reasons for disagreeing anyway. In retrospect, these are probably why the disagreement arose in the first place, with a lot of the more specific points being downstream of them and kind of made-up justifications. These are almost impossible to resolve even in principle. …
Some of these involve what social signal an action might send; for example, even a just war might have the subtle effect of legitimizing war in people’s minds. Others involve cases where we expect our information to be biased or our analysis to be inaccurate; for example, if past regulations that seemed good have gone wrong, we might expect the next one to go wrong even if we can’t think of arguments against it. Others involve differences in very vague and long-term predictions, like whether it’s reasonable to worry about the government descending into tyranny or anarchy. Others involve fundamentally different moral systems, like if it’s okay to kill someone for a greater good. And the most frustrating involve chaotic and uncomputable situations that have to be solved by metis or phronesis or similar-sounding Greek words, where different people’s Greek words give them different opinions.
You can always try debating these points further. But these sorts of high-level generators are usually formed from hundreds of different cases and can’t easily be simplified or disproven. Maybe the best you can do is share the situations that led to you having the generators you do. Sometimes good art can help.
The high-level generators of disagreement can sound a lot like really bad and stupid arguments from previous levels. “We just have fundamentally different values” can sound a lot like “You’re just an evil person”. “I’ve got a heuristic here based on a lot of other cases I’ve seen” can sound a lot like “I prefer anecdotal evidence to facts”. And “I don’t think we can trust explicit reasoning in an area as fraught as this” can sound a lot like “I hate logic and am going to do whatever my biases say”. If there’s a difference, I think it comes from having gone through all the previous steps – having confirmed that the other person knows as much as you might be intellectual equals who are both equally concerned about doing the moral thing – and realizing that both of you alike are controlled by high-level generators. High-level generators aren’t biases in the sense of mistakes. They’re the strategies everyone uses to guide themselves in uncertain situations.
Regarding your “something clearly rational here that’s kinda unintuitive to get a grip on”, I think of it as epistemic learned helplessness as a “social safety valve” to the downside risk of believing persuasive arguments that can (potentially catastrophically) harm the believer, cf. Reason as memetic immune disorder.
There’s a lot more to the study of disagreement if you’re keen, shame it’s mostly just one person working on it and they’re busy writing a book nowadays.
This might feel obvious, but I think it’s under-appreciated how much disagreement on AI progress just comes down to priors (in a pretty specific way) rather than object-level reasoning.
I was recently arguing the case for shorter timelines to a friend who leans longer. We kept disagreeing on a surprising number of object-level claims, which was weird because we usually agree more on the kinda stuff we were arguing about.
Then I basically realized what I think was going on: she had a pretty strong prior against what I was saying, and that prior is abstract enough that there’s no clear mechanism by which I can push against it. So whenever I made a good object-level case, she’d just take the other side — not necessarily because her reasons were better all else equal, but because the prior was doing the work underneath without either of us really knowing it.
There’s something clearly rational here that’s kinda unintuitive to get a grip on. If you have a strong prior, and someone makes a persuasive argument against it, but you can’t identify the specific mechanism by which their argument defeats it, you should probably update that the arguments against their case are better than they appear, even if you can’t articulate them yet. From the outside, this totally just looks like motivated reasoning (and often is), but I think it can be pretty importantly different.
The reason this is so hard to disentangle is that (unless your belief web is extremely clear to you, which seems practically impossible) it’s just enormously complicated. Your prior on timelines isn’t an isolate thing — it’s load-bearing for a bunch of downstream beliefs all at once. So the resistance isn’t obviously irrational, it’s more like… the system protecting its own coherence.
I think this means that people should try their best to disentangle whether some object level argument they’re having comes from real object level beliefs or pretty abstract priors (in which case, it seems less worthwhile to press on them).
My go-to diagram for illustrating your point, from (who else?) Scott Alexander’s varieties of argumentative experience:
More on the high-level generators of disagreement (emphasis mine, other than 1st sentence):
(also related: Value Differences As Differently Crystallized Metaphysical Heuristics and the previous essays in that series)
Regarding your “something clearly rational here that’s kinda unintuitive to get a grip on”, I think of it as epistemic learned helplessness as a “social safety valve” to the downside risk of believing persuasive arguments that can (potentially catastrophically) harm the believer, cf. Reason as memetic immune disorder.
There’s a lot more to the study of disagreement if you’re keen, shame it’s mostly just one person working on it and they’re busy writing a book nowadays.