Today's post, Archimedes's Chronophone was originally published on March 23, 2007. A summary (taken from the LW wiki):

Imagine that Archimedes of Syracuse invented a device that allows you to talk to him. Imagine the possibilities for improving history! Unfortunately, the device will not literally transmit your words - it transmits cognitive strategies. If you advise giving women the vote, it comes out as advising finding a wise tyrant, the Greek ideal of political discourse. Under such restrictions, what do you say to Archimedes? In other words, how can you communicate general thinking patterns which will lead to right answers, as opposed to cached content?

This post is part of the Rerunning the Sequences series, where we'll be going through Eliezer Yudkowsky's old posts in order so that people who are interested can (re-)read and discuss them. The previous post was Useless Medical Disclaimers, and you can use the sequence_reruns tag or rss feed to follow the rest of the series.

Sequence reruns are a community-driven effort. You can participate by re-reading the sequence post, discussing it here, posting the next day's sequence reruns post, or summarizing forthcoming articles on the wiki. Go here for more details, or to have meta discussions about the Rerunning the Sequences series.

New Comment
21 comments, sorted by Click to highlight new comments since:

What surprised me is that many took for granted that the abolition of slavery in the Greek world was economically and socially feasible without making the Greeks less an influence on later civilization than they where.

Modern morals simply don't work very well in the ancient world. This goes for a variety of practices and laws that modern Westerners find abhorrent. To borrow Robin Hanson's vocabulary if you succeed in convincing ancient farmers to be more like foragers morality wise, this would not result in history being prettier in our eyes, but in them being displaced by other farmers with farmer values, since they will be, you know, better farmers.

Edit: Changed everyone to many in the first sentence.

I didn't get this impression from the article. I'm sure that some people read it as taking the moral superiority of our age for granted simply because they can't imagine that someone smart and respectable could do otherwise. However, this seems like a very superficial and careless reading -- the article clearly takes a dig at this feeling of superiority, claiming that it's due to the same cognitive strategy that would have lead people in past ages to align with the prevailing values of their time. It doesn't seem to me that the majority, let alone everyone, could have read it so negligently; the OB/LW audience is normally better than that.

read it as taking the moral superiority of our age for granted ... this seems like a very superficial and careless reading

I think it's an accurate reading. I think based on Eliezer's other writings, he believes that, while modern morality is wrong on many points, on those issues where modern and ancient morality differ, it's generally because moderns understand things that ancients did not.

I haven't read all that he's written on the topic, so it may be true for all I know. But I'd still be surprised and disappointed if he and other prominent participants here take for granted, for example, that one-person-one-vote democracy is a good idea for all places and times, which is given in the original article as one of the ideas that a proponent of modern values might want to transmit. (Of course, this is a widespread and high-status delusion nowadays, but the amount of evidence against it beats almost anything that's normally considered superstitious.)


Thanks for the comment, I've edited the original statement based on the feedback.

I didn't imply Eliezer thought that way in fact he implicitly lays the groundwork to question our currently held moral sensibilities by emphasising the need to look at the cognitive strategies that produce results.

"Everyone" was an exaggeration, not meant literally, put there in the spur of the moment to communicate my surprise that many did. Much more than I would have expected. Eliezer asked them to come up with cognitive strategies that would help Archimedes "win", implicitly thus helping modern humanities odds of "winning" and serving as a good lesson about how the right answer can't be conflated with the method of obtaining the right answer.

Well, the obvious problem is that I can't convince Archimedes of anything I currently believe by providing arguments for it. Once you realize that, the next logical conclusion is to simply try to improve Archimedes' beliefs. For instance, if I were to critique modern science and its failures relative to bayesianism, it would come out as a critique of Archimedes' system of obeying authority. Additionally, I've recently been wondering about the line between sentience and non-sentience, and more specifically if animals like dolphins might actually deserve to be classified as "people" (as a side note, if anyone knows of any research on just how sentient dolphins are, I would be really interested in seeing it). If I were to talk about that, I suspect it would come out as a series of musings about whether women, slaves, etc, might deserve personhood rights.

If I try to be clever and argue for something I don't personally believe, there will be flaws that I can see in the argument, which will translate to flaws that Archimedes can see easily. So for instance, I can't put together a series of arguments in favor of slavery, because I will see flaws in those arguments. One obvious way I could get around this is to put someone else on the phone. I'm sure I can find a very well educated priest or rabbi who believes that following authority is the best system of epistemology, and despises science. That individuals arguments (of course, I wouldn't explain how the chronophone worked) would come out as arguments that sound convincing to Archimedes that following authority is bad. I could do something similar with my dolphin idea by putting the most eloquent speaker for PETA I could find.

I think RobinZ's comment is great:

I think my major problem with this article is that the perfectly reasonable conclusion - that you can't do better in the future [of] today by thinking today's cached thoughts about how people in the past could have done better in the future of the past - is obscured by the utterly ridiculous device of the eponymous chronophone.

I think EY's story might still be useful if it were prefaced with RobinZ's comment and the chronophone as an example of how you have to think about your present problem.

If you want to do better in today's future you have to figure out what you should be doing differently than you have been. The answer can't be 'apply the scientific method!' because we already do that. Useful answers might be 'think about where we should apply the scientific method where we haven't been applying it already' or 'think about how to improve the process of science'. If we give these answers to the chronophone, archimedes should hear that he should think about his own process of discovery and how to apply it to novel situations or how to improve his process. Of course this advice is vague so perhaps they won't be of much use, but thats the kind of questions we need to work on.


Clearly the chronophone thought experiment failed as a pedagogic device, since most of the comments to the article were completely confused about what it was getting at.

But I think part of what makes it confusing is that the distinction between cached thoughts and cognitive strategies is not as clear-cut as you might think. For example, consider Polya's How to Solve It. This book is just a collection of cached thoughts -- "Hm, I seem to be stuck, I guess I should... think of a related, more accessible problem". But at the same time, it is also a cognitive strategy -- in fact, it's like a high-level description of an algorithm for solving math problems.

Similarly, the premise of the chronophone experiment is that we cannot transmit the Scientific Method back in time, because it's merely an idea, not a strategy for coming up with an idea. But consider how we will strive to do better in the future than today -- the Scientific Method (and related ideas, like "try to work quantitatively") will be absolutely crucial. They are the best way we know to make our reasoning approximate the Bayesian ideal. On the other hand, it is not clear a priori that the process that invented the Scientific Method is going to be very important -- for all we know that process could just have been "try any random thing".

I guess what my complaint boils down to is that the thought experiment undervalues ideas. It makes them seem like an epiphenomenon, the noises we make after we have already arrived at a conclusion via some other (unspecified) cognitive strategy. But ideas is the way our culture implements cognitive strategies.

But I think part of what makes it confusing is that the distinction between cached thoughts and cognitive strategies is not as clear-cut as you might think.

Clarifies some vague feelings I had while reading the article. Good work.

You might think of Cached Thoughts (CTs) as organized into a rough hierarchy:

  • level-0 CTs (object level, i.e. 'apples are good')
  • level-1 CTs which are about how to work with/create level-0 CTs ('try to be quantitative', 'think of related more accessible problems')
  • level-2 CTs which are about how to work with/create level-1 CTs ('try to visualize what the world would be like if you were wrong' )
  • level-n CTs are about how to work with/create level-(n-1) CTs

The original post was focused on level-2 CTs of course. Obviously this hierarchy isn't strict and the boundaries are fuzzy, but I do think its meaningful to try to develop level-2 CTs.

Can you clarify what you mean by:

But ideas is the way our culture implements cognitive strategies.


(Sorry for the slow reply!)

I just meant something like this: the article talks about "cognitive strategies" as something mysterious and unspecified, which the cronophone can somehow read off and transmit. But lacking a cronophone, how can we convey strategies? Well, by expressing them as ideas, e.g. "try to work at several levels of abstractions at the same time", or "pick a good system of notation". (At least this is true if you try to improve the thinking patterns of an entire culture -- a single individual can probably get better at some things through practice without conceptualizing why).

As for the CT-level, as you say the boundaries are fuzzy. It's hard to find examples which fit squarely at the third level: "Be quantitative" seems to work when thinking about problem-solving strategies as when think about object level problems, and "visualize the world if you were wrong" works equally well when doubting souls as when doubting the scientific method.

So perhaps it's enough to try to improve our thinking, and our thinking about thinking will automatically follow? The articles final paragraph notes that "to get nonobvious output, you need nonobvious input". That's true for the cronophone, but is it true for thinking in general? If so, that's slightly discouraging. Compare with "civilization advances by extending the number of important operations which we can perform without thinking about them" -- I would hope that just by gradually adding more and more ideas to the pool of things we consider obvious, we will get saner and saner, even if the mental operations we do on those ideas are not particularly novel.

Yes. I didn't understand what the hell this thing was EY was describing, or why that was its name, until I understood why he was asking the question, at which point the device becomes superfluous and he should have just asked what he was asking.

My first thought is to just discuss methods of learning things. It will get Archimedes to discuss methods of learning things, and could result in him thinking of the scientific method.

Also, does he know about how this works? If so, I'd talk about the scientific method, he'd hear about his epistemological approach, realize that I feel it is important to talk about an epistemological approach, and figure that I must believe his is wrong. He'll then think about it, and may come up with the scientific method.

I wonder what would happen if I just construct an argument that slavery is good.

I think we are getting into meatier posts now, hopefully there will be more interesting discussions in the comments.

I noticed that the post and the comments were a little bit vague on what exactly does or doesn't belong in my 'culture'. Would arguing that nudity is good translate into something obvious and true (like for most Europeans, as I understand it), or would it turn into something absurd and possibly evil (like how I think most Americans see it)? In other words, whenever the examples talk about zeitgeists, am I to translate that into generalizing from the example of me?

Arguing for evolution is super-obvious to me, but a large fraction of Alaskans aren't too keen on the idea. So if I told the chronophone about natural selection while I'm in Alaska, would it tell Archimedes that humans are featherless chickens... but if I said it in the midst of a Biology symposium, it would tell him that humans are mortal?

My guess is that you can't change your affiliations so easily, because it avoids the point of the hypothetical. But I would be interested to see what comes out if I tell him about all the things that I think other cultures do better than my own.

I'd say whatever culture's beliefs you most identify with, i.e., assume things are maximally inconvinient.

And so in that case, I've been wondering how the Chronophone would handle the fact that my culture has orders of magnitude more information available to it than Archimedes' did. It seems the simplest thing to do would be to drop packets, but that would complicate all the individual transmissions sent by all the people prior to this question. Avoiding that issue, I think there's a difference between the ancient culture and my own that the Chronophone couldn't easily handle: back then, the sum of human knowledge was probably knowable to a brilliant mind like his. So if I used a computer to send back all of wikipedia, it would have to lose so much detail that Archimedes would be able to understand all of it in his own lifetime, whereas I couldn't even read the new content added to wikipedia each month.

I'm not trying to say that I'd read wikipedia to him, or that I think a data dump would be a good way to try and get the right insights into his mind to get his culture going down the path that my culture would most approve. What I'm trying to say is that since he was a big fish in a tiny pond, and I'm an exceedingly below-average fish in a freaking colossal sea (which in the spirit of maximal inconvenience, I will take to be a community of Feynmans), and the Chronophone scales down the bits of my sea that I'm trying to import to his pond, there is a tiny chance that anything I say will be 1) true, 2) surprising, and 3) able to sway him in the direction I want.

Boiling down the requirements of 1 and 2, this Chronophone doesn't seem to map true statements to true statements. It maps the obviousness of statements. So things that are both true, and obvious to my maximally inconvenient community, won't map to true insights to Archimedes' world. As I understand the question posed by Eliezer about bringing science to them, my task is tantamount to teaching Feynman that he is wrong about science: that there is a better way to come to know the world. Unless Feynman knew less about Bayes Theorem than me (which would be an accident of history more than anything), this could not be both true and obvious.

Would arguing that nudity is good translate into something obvious and true (like for most Europeans, as I understand it)

Seriously? How many people have you seen walking around casually nude in an average European city? What does it even mean to say "nudity is good"?

A better example would be homosexuality as seen in Iran (they hang you for it) vs as seen in most of the Western world (it's okay)

I've seen more boobs on German television during the daytime than I've ever seen on television in America. If you count beaches next to cities, then I have seen quite a few casually nude people walking around, actually. Walking around in inner cities isn't the only metric to go by.

Obviously there are good reasons to wear clothes, and you are right that I should have chosen a word that isn't stupidity reversed. Perhaps I should have said" nudity isn't superawfulsinsinsin that harms children and topples empires".

If I weren't looking for an example where my beliefs are at odds with the people around me, and that I thought were done better elsewhere, homosexuality in the Arabic world would be a great example. But more apropros to my main point would be comparing America's stance on homosexuality with, say, Spain's.

I liked that article! The main takeaway is separating "knowing how to think" from various trivia, but the exercise is fun too:

I'd talk about how difficult it is to actually figure things out, to get correct beliefs in science, history, politics etc. About the times I've changed my minds, about the things I must admit I don't know enough about to have a much of an opinion (like economics), while still knowing that there probably is a right answer, and that I have to avoid the toxic ambient anti-epistemology (things like "everyone can hold the opinion he wants", or "changing your mind is a sign of weakness", or "you have to listen to your heart"), or the cheap way out of taking a middle ground position on everything.

I'd talk about various subjects where I suspect the conventional wisdom is wrong, about what the alternatives are, and about how likely those are to be wrong too.

I'd talk about the unfortunate fact that even scientists don't tend to update their beliefs enough, and that we have to wait for them to die out before new views can be accepted. I'd talk about various ways of correcting that (betting markets, open-access journals, various schemes of debate tools I've come up with back in the days when I was interested in that), and caution about whether those would really improve things.

I'd talk about the failure modes of smart people, and how to try correcting for those.

Isn't the straight forward answer to not communicate in cached thoughts, build everything form shared fundamentals and over time build a new shared language to explain modern concepts to Archimedes?

The rules do not seem to stop this and as long as Archimedes is dedicated to the task you could communicate arbitrarily complex ideas.

What fundamentals do we share with Archimedes?

I'd probably start by expressing frustration with the chronophone's limitations, as it sounds like that would get through. (Done!) Then I might try telling him to convince rulers to act in ways that would benefit him personally, since figuring out what policy he wants would require empirical reasoning if he does it right. (Sadly, 'rationalist' people in the present do not agree about which national policies would help me, and I've had no apparent success convincing politicians of my own views.)