Posts
Comments
Some people (me included) value a certain level of non-manipulation. I'm trying to cash out that instinct. And it's also needed for some ideas like corrigibility. Manipulation also combines poorly with value learning, see eg our paper here https://arxiv.org/abs/2004.13654
I do agree that saving the world is a clearly positive case of that ^_^
I have an article on "Anthropic decision theory". with the video version here.
Basically, it's not that the presumptuous philosopher is more likely to be right in a given universe, its that there are far more presumptuous philosophers in the large universe. So if we count "how many presumptuous philosophers are correct", we get a different answer to "in how many universes is the presumptuous philosopher correct". These things only come apart in anthropic situations.
Suart, by " is complex" are you referring to...
I mean that that defining can be done in many different ways, and hence has a lot of contingent structure. In contrast, in , the $\rho is a complex distribution on , conditional on ; hence itself is trivial and just encodes "apply to and in the obvious way.
This is a link to "An Increasingly Manipulative Newsfeed" about potential social media manipulation incentives (eg FaceBook).
I'm putting the link here because I keep losing the original post (since it wasn't published by me, but I co-wrote it).
A boundedly-rational agent is assumed to be mostly rational, failing to be fully rational because of a failure to figure things out in enough detail.
Humans are occasionally rational, often biased, often inconsistent, sometimes consciously act against their best interests, often follow heuristics without thinking, sometimes do think things through. This doesn't seem to correspond to what is normally understood as "boundedly-rational".
that paper was about fitting observations of humans to a mathematical model of "boundedly-rational agent pursuing a utility function"
It was "any sort of agent pursuing a reward function".
We don't need a special module to get an everyday definition of doorknobs, and likewise I don't think we don't need a special module to get an everyday definition of human motivation.
I disagree. Doornobs exist in the world (even if the category is loosely defined, and has lots of edge cases), whereas goals/motivations are interpretations that we put upon agents. The main result of the Occam's razor paper is that there the goals of an agent are not something that you can know without putting your own interpretation on it - even if you know every physical fact about the universe. And two very different interpretations can be equally valid, with no way of distinguishing between them.
(I like the anthropomorphising/dehumanising symmetry, but I'm focusing on the aspects of dehumanising that cause you to make errors of interpretation. For example, out-groups are perceived as being coherent, acting in concert without disagreements, and often being explicitly evil. This is an error, not just a reduction in social emotions)
For instance throughout history people have been able to model and interact with traders from neighbouring or distant civilizations, even though they might think very differently.
Humans think very very similarly to each other, compared with random minds from the space of possible minds. For example, we recognise anger, aggression, fear, and so on, and share a lot of cultural universals https://en.wikipedia.org/wiki/Cultural_universal
There haven’t been as many big accomplishments.
I think we should look at the demand side, not the supply side. We are producing lots of technological innovations, but there aren't so many major problems left for them to solve. The flush toilet was revolutionary; a super-flush ecological toilet with integrated sensors that can transform into a table... is much more advanced from the supply side, but barely more from the demand side: it doesn't fulfil many more needs than the standard flush toilet.
Cool, good summary.
Humans have a theory of mind, that makes certain types of modularizations easier. That doesn't mean that the same modularization is simple for an agent that doesn't share that theory of mind.
Then again, it might be. This is worth digging into empirically. See my post on the optimistic and pessimistic scenarios; in the optimistic scenario, preferences, human theory of mind, and all the other elements, are easy to deduce (there's an informal equivalence result; if one of those is easy to deduce, all the others are).
So we need to figure out if we're in the optimistic or the pessimistic scenario.
My understanding of the OP was that there is a robot [...]
That understanding is correct.
Then my question was: what if none of the variables, functions, etc. corresponds to "preferences"? What if "preferences" is a way that we try to interpret the robot, but not a natural subsystem or abstraction or function or anything else that would be useful for the robot's programmer?
I agree that preferences is a way we try to interpret the robot (and how we humans try to interpret each other). The programmer themselves could label the variables; but its also possible that another labelling would be clearer or more useful for our purposes. It might be a "natural" abstraction, once we've put some effort into defining what preferences "naturally" are.
but "white box" is any source code that produces the same input-output behavior
What that section is saying is that there are multiple white boxes that produce the same black box behaviour (hence we cannot read the white box simply from the black box).
modularization is super helpful for simplifying things.
The best modularization for simplification will not likely correspond to the best modularization for distinguishing preferences from other parts of the agent's algorithm (that's the "Occam's razor" result).
but the function f is not part of the algorithm, it's only implemented by us onlookers. Right?
Then isn't that just a model at another level, a (labelled) model in the heads of the onlookers?
Thanks! Useful insights in your post, to mull over.
An imminent incoming post on this very issue ^_^
Yes, things like honour and anger serve important signalling and game-theoretic functions. But they also come to be valued intrinsically (the same way people like sex, rather than just wanting to spread their genes), and strongly valued. This makes it hard to agree that "oh, your sacred core value is only in the service of this hidden objective, so we can focus on that instead".
Cool, neat summary.
Sorry, had a few terrible few days, and missed your message. How about Friday, 12pm UK time?
Stuart, I'm writing a review of all the work done on corrigibility. Would you mind if I asked you some questions on your contributions?
No prob. Email or Zoom/Hangouts/Skype?
Very good. A lot of potential there, I feel.
The information to distinguish between these interpretations is not within the request to travel west.
Yes, but I'd argue that most of moral preferences are similarly underdefined when the various interpretations behind them come apart (eg purity).
There are computer programs that can print their own code: https://en.wikipedia.org/wiki/Quine_(computing)
There are also programs which can print their own code and add something to it. Isn't that a way in which the program fully knows itself?
Thanks! It's cool to see his approach.
Wiles proved the presence of a very rigid structure - not the absence - and the presence of this structure implied FLT via the work of other mathematicians.
If you say that "Wiles proved the Taniyama–Shimura conjecture" (for semistable elliptic curves), then I agree: he's proved a very important structural result in mathematics.
If you say he proved Fermat's last theorem, then I'd say he's proved an important-but-probable lack of structure in mathematics.
So yeah, he proved the existence of structure in one area, and (hence) the absence of structure in another area.
And "to prove Fermat's last theorem, you have to go via proving the Taniyama–Shimura conjecture", is, to my mind, strong evidence for "proving lack of structure is hard".
You can see this as sampling times sorta-independently, or as sampling times with less independence (ie most sums are sampled twice).
Either view works, and as you said, it doesn't change the outcome.
Yes, I got that result too. The problem is that the prime number theorem isn't a very good approximation for small numbers. So we'd need a slightly more sophisticated model that has more low numbers.
I suspect that moving from "sampling with replacement" to "sampling without replacement" might be enough for low numbers, though.
Note that the probabilistic argument fails for n=3 for Fermat's last theorem; call this (3,2) (power=3, number of summands is 2).
So we know (3,2) is impossible; Euler's conjecture is the equivalent of saying that (n+1,n) is also impossible for all n. However, the probabilistic argument fails for (n+1,n) the same way as it fails for (3,2). So we'd expect Euler's conjecture to fail, on probabilistic grounds.
In fact, the surprising thing on probabilistic grounds is that Fermat's last theorem is true for n=3.
Good, cheers!
Another key reason for time-inconsistent preferences: bounded rationality.
Why do the absolute values cancel?
Because , so you can remove the absolute values.
Cheers, interesting read.
I also think the pedestrian example illustrates why we need more semantic structure: "pedestrian alive" -> "pedestrian dead" is bad, but "pigeon on road" -> "pigeon in flight" is fine.
Nope! Part of my own research has made more optimistic about the possibilities of understanding and creating intelligence.
I think this shows that the step-wise inaction penalty is time-inconsistent: https://www.lesswrong.com/posts/w8QBmgQwb83vDMXoz/dynamic-inconsistency-of-the-stepwise-inaction-baseline
https://www.futuretimeline.net/forum/topic/17903-kurzweils-2009-is-our-2019/ , forwarded to me by Daniel Kokotajlo (I added a link in the post as well).
"How to think about features of models and about consistency", in a relatively fun way as an intro to a big post I'm working on.
Then it has a wrong view of wings and fur (as well as a wrong view of pigs). The more features it has to get right, the harder the adversarial model is to construct - it's not just moving linearly in a single direction.
Thanks! Good insights there. Am reproducing the comment here for people less willing to click through:
I haven't read the literature on "how counterfactuals ought to work in ideal reasoners" and have no opinion there. But the part where you suggest an empirical description of counterfactual reasoning in humans, I think I basically agree with what you wrote.
I think the neocortex has a zoo of generative models, and a fast way of detecting when two are compatible, and if they are, snapping them together like Legos into a larger model.
For example, the model of "falling" is incompatible with the model of "stationary"—they make contradictory predictions about the same boolean variables—and therefore I can't imagine a "falling stationary rock". On the other hand, I can imagine "a rubber wine glass spinning" because my rubber model is about texture etc., my wine glass model is about shape and function, and my spinning model is about motion. All 3 of those models make non-contradictory predictions (mostly because they're issuing predictions about non-overlapping sets of variables), so the three can snap together into a larger generative model.
So for counterfactuals, I suppose that we start by hypothesizing some core of a model ("a bird the size of an adult blue whale") and then searching out more little generative model pieces that can snap onto that core, growing it out as much as possible in different ways, until you hit the limits where you can't snap on any more details without making it unacceptably self-contradictory. Something like that...
(this is, obviously, very speculative ^_^ )
...which also means that they didn't have an empire to back them up?
Thanks for your research, especially the Afonso stuff. One question for that: were these empires used to gaining/losing small pieces of territory? ie did they really dedicate all their might to getting these ports back, or did they eventually write them off as minor losses not worth the cost of fighting (given Portuguese naval advantages)?
Based on what I recall reading about Pizzaro's conquest, I feel you might be underestimating the importance of horses. It took centuries for European powers to figure out how to break a heavy cavalry charge with infantry; the amerindians didn't have the time to figure it out (see various battles where small cavalry forces routed thousands of troops). Once they had got more used to horses, later Inca forces (though much diminished) were more able to win open battles against the Spanish.
Maybe this was the problem for these empires: they were used to winning open battles, but were presented with a situation where only irregular warfare or siege defences could win. They reacted as an empire, when they should have been reacting as a recalcitrant province.
Also, giving something more points for killing people than making cake sounds like a bad incentive scheme.
In the original cake-or-death example, it wasn't that killing got more points, it's that killing is easier (and hence gets more points over time). This is a reflection of the fact that "true" human values are complex and difficult to maximise, but many other values are much easier to maximise.
My main note is that my comment was just about the concept of rigging a learning process given a fixed prior over rewards. I certainly agree that the general strategy of "update a distribution over reward functions" has lots of as-yet-unsolved problems.
Ah, ok, I see ^_^ Thanks for making me write this post, though, as it has useful things for other people to see, that I had been meaning to write up for some time.
On your main point: if the prior and updating process are over things that are truly beyond the AI's influence, then there will be no rigging (or, in my terms: uninfluenceable->unriggable). But there are many things that look like this, that are entirely riggable. For example, "have a prior 50-50 on cake and death, and update according to what the programmer says". This seems to be a prior-and-update combination, but it's entirely riggable.
So, another way of seeing my paper is "this thing looks like a prior-and-update process. If it's also unriggable, then (given certain assumptions) it's truly beyond the AI's influence".
Thanks! Responded here: https://www.lesswrong.com/posts/EYEkYX6vijL7zsKEt/reward-functions-and-updating-assumptions-can-hide-a
Thanks! Responded here: https://www.lesswrong.com/posts/EYEkYX6vijL7zsKEt/reward-functions-and-updating-assumptions-can-hide-a
thanks! Changed title (and corrected badly formatted footnote)
I agree that for such a system, the optimal policy of the actor is to rig the estimator, and to "intentionally" bias it towards easy-to-satisfy rewards like "the human loves heroin".
The part that confuses me is why we're having two separate systems with different objectives where one system is dumb and the other system is smart.
We don't need to have two separate systems. There's two meaning to your "bias it towards" phrase: the first one is the informal human one, where "the human loves heroin" is clearly a bias. The second is some formal definition of what is biasing and what isn't. And the system doesn't have that. The "estimator" doesn't "know" that "the human loves heroin" is a bias; instead, it sees this as a perfectly satisfactory way of accomplishing its goals, according to the bridging function it's been given. There is no conflict between estimator and actor.
Imagine that you have a complex CIRL game that models the real world well but assumes that the human is Boltzmann-rational. [...] Such a policy is going to "try" to learn preferences, learn incorrectly, and then act according to those incorrect learned preferences, but it is not going to "intentionally" rig the learning process.
The AI would not see any of these actions as "rigging", even if we would.
It might think "hey, I should check whether the human likes heroin by giving them some", and then think "oh they really do love heroin, I should pump them full of it".
It will do this if it can't already predict the effect of giving them heroin.
It won't think "aha, if I give the human heroin, then they'll ask for more heroin, causing my Boltzmann-rationality estimator module to predict they like heroin, and then I can get easy points by giving humans heroin".
If it can predict the effect of giving humans heroin, it will think something like that. It think: "if I give the humans heroin, they'll ask for more heroin; my Boltzmann-rationality estimator module confirms that this means they like heroin, so I can efficiently satisfy their preferences by giving humans heroin".
Strong upvote for writing your own predictions before seeing the 2019 graph.