# Aubrey de Grey Discussion, 15

I discussed epistemology and cryonics with Aubrey de Grey via email. Click here to find the rest of the discussion. Yellow quotes are from Aubrey de Grey, with permission. Bluegreen is me, red is other.
A reason "strong refutation" seems to make sense is because of something else. Often what we care about is a set of similar ideas, not a single idea. A refutation can binary refute some ideas in a set, and not others. In other words: criticisms that refute many variants of an idea along with it seem "strong”.
That’s basically what I do. I agree with all you go on to say about closeness of variants etc, but I see exploration of variants (and choice of how much to explore variants) as coming down to a sequence of dice-rolls (or, well, coin-flips, since we’re discussing binary choices).
I don't know what this means. I don't think you mean you judge which variants are true, individually, by coin flip.

Maybe the context is only variants you don't have a criticism of. But if several won their coin flips, but are incompatible, then what? So I'm not clear on what you're saying to do.

Also, are you saying that amount of sureness, or claims criticisms are strong or weak (you quote me explaining how what matters is which set of ideas a criticism does or doesn't refute), play no role in what you do? Only CR + randomness?
The coin flips are not to decide whether a given individual idea is true or false, they are to decide between pairs of ideas. So let’s say (for simplicity) that there are 2^N ideas, of which 90% are in one group of close variants and the other 10% are in a separate group of close variants. “Close”, here, simply means differing only in ways I don’t care about. Then I can do a knockout tournament to end up choosing a winning variant, and 90% of the time it will be in the first group. Since I don’t actually care about the features that distinguish the variants within either group, only the features that distinguish the groups. I’m done. In other words, the solidity of an idea is measured by how many close variants it has - let’s call it the “variant density” in its neighbourhood. In practice, there will typically be numerical quantities involved in the ideas, so there will be an infinite number of close variants in each group - but if I have a sense of the variant densities in the two regions then that’s no problem, because I don’t need to do the actual tournament.
OK, I get the rough idea, though I disagree with a lot of things here.

You are proposing a complex procedure, involving some tricky math. It looks to me like the kind of thing requiring, minimum, tens of thousands of words to explain how it works. And a lot of exposure to public criticism to fix some problems and refine, even if the main points are correct.

Perhaps, with a fuller explanation, I could see why Aubreyism is correct about this and change my mind. I have some reasons not to think so, but I do try to keep an open mind about explanations I haven't read yet, and I'd be willing to look at a longer version. Does one exist?

Some sample issues where I'd want more detail include (no need to answer these now):

• Is the score the total variants anywhere, ignoring density, regions and neighborhoods? If so, why are those other things mentioned? If not, how is the score calculated?
• Why are ideas with more variants better, more likely to be true, or something like that? And what is the Aubreyism thing to say there, and how does that concept work in detail?
• The "regions" discussed are not regions of space. What are they, how are they defined, what are they made out of, how is distance defined in them, how do different regions connect together?
• The coin flipping procedure wouldn't halt. So what good is it?
• I can imagine skipping the coin flipping procedure because the probabilities will be equally distributed among the infinite ideas. But then the probabilities will all be infinitesimal. Dealing with those infinitesimals requires explanation.
• I'm guessing the approach involves grouping together infinitesimals by region. This maybe relies on there being a finite number of regions of ideas involved, which is a premise requiring discussion. It's not obvious because we're looking at all ideas in some kind of idea-space, rather than only looking at the finite set of ideas people actually propose (as Elliotism and CR do normally do).
• When an idea has infinite variants, what infinity are we talking about? Is it in one-to-one correspondence with the integers, the reals, or what? Do all ideas with infinite variants have the same sort of infinity variants? Infinity is really tricky, and gets a lot worse when you're doing math or measurement, or trying to be precise in a way that depends on the detailed properties of infinity.
• There are other ways to get infinite variants other than by varying numerical quantities. One of these approaches uses conjunctions – modify an idea by adding "and X". Does it matter if there are non-numerical ways to get infinite variants? Do they make a difference? Perhaps they are important to understanding the number and density of variants in a region?
• Are there any cases where there's only finite variants of an idea? Does that matter?
• You can't actually have 90% or 10% of 2^N and get a whole number. This won't harm the main ideas, but I think it's important to fix detail errors in one's epistemology (which I think you agree with: it's why you specified 2^N ideas, instead saying even or leaving unspecified).
• Do ideas actually have different numbers of variants? Both for total number, and density. How does one know? How does one figure out total variant count, and density, for a particular idea?
• How is the distance between two ideas determined? Or whatever is used for judging density.
• What counts as a variant? In common discussion, we can make do with a loose idea of this. If I start with an idea and then think about a way to change it, that's a variant. This is especially fine when nothing much depends on what is a variant of what. But for measuring solidity, using a method which depends on what is a variant of what, we'll need a more precise meaning. One reason is that some variant construction methods will eventually construct ALL ideas, so everything will be regarded as a variant of everything else. (Example method: take ideas in English, vary by adding, removing or modifying one letter.) Addressing issues like this requires discussion.
• Where does criticism factor into things?
• What happens with ideas which we don't know about? Do we just proceed as if none of those exist, or is anything done about them?
• Does one check his work to make sure he calculated his solidity measurements right? If so, for how long?
• Is this procedure truth-seeking? Why or why not? Does it create knowledge? If so, how? Is it somehow equivalent to evolution, or not?
• Why do people have disagreements? Is it exclusively because some people don't know how to measure idea solidity like this, because of calculation errors, and because of different ideas about what they care about?
• One problem about closeness in terms of what people care about is circularity. Because this method is itself supposed to help people decide things like what to care about.
• How does this fit with DD's arguments for ideas that are harder to vary? Your approach seems to favor ideas that are easier to vary, resulting in more variants.
• I suspect there may be lots of variants of "a wizard did it". Is that a good idea? Am I counting its variants wrong? I admit I'm not really counting but just sorta wildly guessing because I don't think you or I know how to count variants.
That is only an offhand sampling of questions and issues. I could add more. And then create new lists questioning some of the answers as they were provided. Regarding what it takes to persuade me, this gives some indication of what kind of level of detail and completeness it takes. (Actually a lot of precision is lost in communication.)

Does this assessment of the situation make sense to you? That you're proposing a complex answer to a major epistemology problem, and there's dozens of questions about it that I'd want answers to. Note: not necessarily freshly written answers from you personally, if there is anything written by you or others at any time.

Do you think you know answers to every issue I listed? And if so, what do you think is the best way for me to learn those full answers? (Note: If for some answers you know where to look them up as needed, instead of always saving them in memory, that's fine.)

Or perhaps you'll explain to me there's a way to live with a bunch of unanswered questions – and a reason to want to. Or maybe something else I haven't thought of.
To try to get at one of the important issues, when and why would you assign X a higher percent (aka strength, plausibility, justification, etc) than Y or than ruminating more? Why would the percents ever be unequal? I say either you have a criticism of an option (so don't do that option), or you don't (so don't raise or lower any percents from neutral). What specifically is it that you think lets you usefully and correctly raise and lower percents for ideas in your decision making process?

I think your answer is you judge positive arguments (and criticisms) in a non-binary way by how "solid" arguments are. These solidity judgments are made arbitrarily, and combined into an overall score arbitrarily.
I think my clarification above of the role of “variant density” as a measure of solidity answers this, but let me know if it doesn’t.
I agree with linking issues. Measuring solidity (aka support aka justification) is a key issue that other things depend on.

It's also a good example issue for the discussion below about how I might be persuaded. If I was persuaded of a working measure of solidity, I'd have a great deal to reconsider.
Sure - and that’s what I claim I do (and also what I claim you in fact do, even though you don’t think you do).
I do claim to do this [quoted below]. Do you think it's somehow incompatible with CR?
On reflection, and especially given your further points below, I’d prefer to stick with Aubreyism and Elliotism rather than justificationism and CR, because I’m new to this field and inadequately clear as to precisely how the latter terms are defined, and because I think the positions we’re debating between are our own rather than other people’s.
OK, switching terminology.

Do you think
doing your best with your current knowledge (nothing special), and also specifically having methods of thinking which are designed to be very good at finding and correcting mistakes.
is incompatible with Elliotism? How?
OK - as above, let’s forget unmodified CR and also unmodified justificationism. I think we’ve established that my approach is not unmodified justificationism, but instead it is (something like) CR triaged by justificationism. I’m still getting the impression that your stated approach, whether or not it’s reeeeally close to CR, is unable to make decisions adequately rapidly for real life, and thus is not what you actually do in real life.
I don't know what to do with that impression.

Do you believe you have a reason Elliotism could not be timely in theory no matter what? Or only a reason Elliotism is not timely today because it's not developed enough and the current approach is flawed, but one day there might be a breakthrough insight so that it can be timely?

I think the timeliness thing is a second key issue. If I was persuaded Elliotism isn't or can't be timely, I'd have a lot to reconsider. But I'm pretty unclear on the specifics of your counter-arguments regarding timeliness.
What's the problem for CR with consensus-low fields?
Speed of decision-making. The faster CR leads to consensus in a given field, the less it needs to be triaged.
OK, I have a rough idea of what you mean. I don't think this is important to our main disagreements.
This is a general CR approach: do something with no proof it will work, no solidity, no feeling of confidence (or if you do feel confidence, it doesn't matter, ignore it). Instead, watch out for problems, and deal with them as they are found.
Again, I can’t discern any difference in practice between that and what I already do.
Can you discern a difference between it and what most people do or say they do?
Oh, sure - I think most people are a good deal more content than me to hold pairs of views that they recognise to be mutually incompatible.
What I was talking about above was an innocent-until-proven-guilty approach to ideas, which is found in both CR and Elliotism (without requiring infallible proof). You indicated agreement, but now bring up the issue of holding contradictory ideas, which I consider a different issue. I am unclear on whether you misunderstood what I was saying, consider these part of the same issue, or what.

Regarding holding contradictory ideas, do you have a clear limit? If I were to adopt Aubreyism, how would I decide which mutually incompatible views to keep or change? If the answer involves degrees of contentness, how do I calculate them?

Part of the Elliotism answer to this issue involves context. Whether ideas relevantly contradict each other is context dependent. Out of context contradictions aren't important. The important thing is to deal with relevant contradictions in one's current context. Put another way: deal with contradictions relevant to choices one makes.

Consider the contradicting ideas of quantum mechanics and general relativity. In a typical dinner-choosing context, neither of those ideas offers a meal suggestion. They both say essentially "no comment" in this context, which doesn't contradict. They aren't taking different sides in the dinner arbitration. I can get pizza for dinner without coming into conflict with either of those ideas.

On the other hand if there was a contradiction in context – basically meaning they are on disagreeing sides in an arbitration – then I'd address that with a win/win solution. Without such a solution, I could only proceed in a win/lose way and the loser would be part of me. And the loser would be chosen arbitrarily or irrationally (because if it weren't, then what was done would be a rational solution and we're back to win/win).

Understanding of context is one of the things which allows Elliotism to be timely. (A refutation of my understanding of context is another thing which would lead to me reconsidering a ton.)

If I were to change my mind and live by Aubreyism, I would require a detailed understanding of how to handle context under Aubreyism (for meals, contradictions, and everything else).
I don’t think our disparate conclusions with regard to the merits of signing up with Alcor arise from you doing the above and me doing something different; I think they arise from our having different criteria for what constitutes a problem. And I don’t think this method allows a determination of which criterion for what constitutes a problem is correct, because each justifies itself: by your criteria, your criteria are correct, and by mine, mine are. (I mentioned this bistability before; I’ve gone back to your answer - Sept 27 - and I don’t understand why it’s an answer.)
Criteria for what is a problem are themselves ideas which can be critically discussed.

Self-justifying ideas which block criticism from all routes are a general category of idea which can be (easily) criticized. They're bad because they block critical discussion, progress, and the possibility of correction if they're mistaken.
OK then: what theoretical sequence of events would conclude with you changing your mind about how you think decisions should be made, in favour of my view?
Starting at the end, I'd have to understand Aubreyism to my satisfaction, think it was right, think Elliotism and (unmodified) CR were both wrong. The exact details are hard to specify in advance because in the sequence of events I would change my mind about what criteria to use when deciding what ideas to favor. So I would not think Aubreyism has no known criticism, rather I'd understand and use Aubreyism's own criteria. And similarly I wouldn't be rejecting Elliotism or CR for having one outstanding criticism (taking into account context), but rather because of some reasons I learned from Aubreyism.

For that matter, I might not have to understand Aubreyism to my satisfaction. Maybe it'd teach me how to adopt ideas without understanding them to my current criteria of satisfaction. It could offer different criteria of satisfaction, but it could also offer a different approach.

So, disclaimer: the below discussion of persuasion contains Elliotist ideas. But if Elliotism is false, then I guess persuasion works some other way, which I don't know and can't speak to.

Starting more at the beginning, my ideas about Elliotism are broadly integrated into my thinking (meaning connected to other ideas). An example area where they are particularly tightly integrated is parenting and education. For ease of reference, my views are called TCS (Taking Children Seriously).

So I'd have to find out things like, if I rejected Elliotism, what views am I to adopt about parenting and education? Is Aubreyism somehow fully compatible with TCS (I don't think so)? Even if it was, I'd have to find out things like how to argue TCS in new ways using Aubreyism instead of Elliotism, there'd be changes.

To give you a sense of the integration, TCS has many essays which explicitly discuss Popper, (unmodified) CR, and Elliotism. A large part of the way TCS was created was applying CR ideas to parenting and education. And also, some TCS concepts played a significant role in creating Elliotism. In addition to TCS learning things from CR, CR can learn from TCS, resulting in a lot of the unmodified-CR/Elliotism differences.

If I'm to change my views on Elliotism and also on TCS, I'll also have to find out why the new views are moral, not immoral (or learn a new approach to morality). I'll have to find out why thousands of written TCS arguments are mistaken, and how far the mistakes go. (Small change in perspective and way of arguing basically saves all the old conclusions? Old conclusions have to be thrown out and recreated with Aubreyism? Somewhere in between?)

And when I try to change my thinking about TCS, I'll run into that fact that it's integrated with many other ideas, so will they have to change to? And they connect to yet more ideas.

So there's this tangled web of ideas. And this is just one area of integration, Elliotism and TCS. Elliotism is also integrated with my politics. And with my opinions of philosophy books. And with my approach to social life. All this could require reevaluation in light of changes to my epistemology.

How can something like this be approached?

It takes a lot of work (which I have willingness to do). One of the general facts of persuasion is, the person being persuaded has to do the large majority of the work. I'd have to persuade myself, with hints and help from you. That is the only way. You cannot make me change my mind, or do most of the work for me.

Though, again, this is an Elliotist view which might not be applicable if you refuted Elliotism. Maybe you can tell me a different way.

(Tangentially, you may note here some incompatibilities with this perspective and how school teachers approach education.)

Another consequence of this integration is that if you persuaded me I was wrong about politics, that could pose a problem for Elliotism. I'd have to figure out where the mistakes were and their full consequences, and that process might involve rejecting Elliotism. If I decide a political idea is false, and there's a chain of ideas from it to an Elliotism idea (which there is), then I'll have to find a mistake in that chain or else rethink part of Elliotism (which is itself linked with the rest of Elliotism and more, posing similar problems). So it could be possible to change my mind about Elliotism without ever discussing it.

Integration of ideas is stabilizing in some ways. If you say I'm wrong about X, I may know a dozen implications of X which I want to figure out how to deal with. This can make it more challenging to provide a satisfactory new view. But integration is also destabilizing because if I do change my mind about X, the implications spread more easily. Persuasion about one point can cause a chain reaction. Especially if I don't block off that chain reaction with a bunch of rationalizations, irrational evasions, refusals to think about implications of ideas, willful disconnections of ideas into more isolated pieces to prevent chain reaction, and so on.

The consequences of a refutation aren't predictable in advance. Maybe it turns out that idea was more isolated than you thought – or less. Maybe you can find mistaken connections near it, maybe not. Until you work out new non-refuted positions, you don't know if it will be a tiny fix or require a whole new philosophy.

Getting back to your question: The sequence of events to change my mind would be large, and largely outside of your control. The majority of it would be outside your view, even if I tried hard to share the process. My integrity would be required.

Ayn Rand says you can't "force a mind". Persuasion has to be voluntary. It's why the person to be persuaded must actively want to learn, and take initiative in the process, not be passive.

However, you could play a critically important role. If you told me one idea (e.g. how to measure solidity), and I worked out the rest from there, you would have had a major role.

More normally, I'd work out a bit from that idea, then ask you a question or argue a point, get your answer, work out a bit more, and so on. And some of your answers would refer me to books and webpages, rather than be written fresh.

It hasn't gone like this so far because I'm experiencing the epistemology discussion as you saying things I've already considered. And frequently already had several debates about. Not exactly identical ideas, but similar in the relevant ways so my previous analysis still applies. Rather than needing to rethink something, I've been using ideas I already know and making minor adjustments to fit the details of our conversation.

I'm also using the discussion to work on ongoing projects like trying to understand Elliotism more clearly, invent better ways to explain it, and better understand where and why people misunderstand it or disagree. I also have more tangential projects like trying to write better.

It's also being used by others who want to understand Elliotism better. People write comments and use things you or I said as a jumping off point for discussions. If you wanted, you could read those discussions and comments.

Those people are also relevant to the issue of a sequence of events in which I'd be persuaded of Aubreyism. If you managed to inspire any doubts about Elliotism, or raise any problems I didn't think I had an answer to, I would raise those issues with others and see what they said. So, via me (both writing and forwarding things), you'd have to end up persuading those people of Aubreyism too. And on the other hand, they could play a big role in persuading me of Aubreyism if they understood one of your correct points before me, and then translated it to my current way of thinking well. (The Aubreyism issue could also create a split and failure to agree, but I wouldn't expect it and I see no signs of that so far.)

I also want to differentiate between full persuasion and superficial persuasion. Sometimes people are persuaded about X pretty easily. But they haven't changed their mind about anything else, so now X contradicts a bunch of their other ideas. A common result is the persuasion doesn't last. Whereas if one is persuaded about X and then makes changes to other ideas until X is compatible with all their thinking, and there's various connections, that'd be a more full kind of persuasion that does a better job of lasting.

One reason superficial persuasion seems to work and last, sometimes, is because of selective attention. People will use idea X if and only if dealing with one particular topic, and not think about other stuff. Then for other topics, they only think about other stuff and not X. So the contradictions between their other ideas and X don't get noticed, because they only think about one or the other at a time.

This further speaks to the complexity and difficulty of rational persuasion.

Getting back to a sequence of events, I don't know a specific one in detail or I'd be persuaded now. What I know is more like the categories of events that would matter and what sorts of things have to happen. (The sequencing, to a substantial extent, is flexible. Like I could learn an epistemology idea and adjust my politics, or vice versa, the sequence can go either way. At least that's the Elliotism view.)

Trying to be more specific, here's an example. You say something I don't have an answer to. It could be about measuring solidity, but it could be about pretty much any of my views I've been explaining because I take them all seriously and they're all integrated. I investigate. I find problems with several of my related ideas. I also consider some related ideas which I don't see any problem with, so I ask you about the issue. My first question is whether you think those ideas are false and I'm missing it, or you think I'm mistaken that they are related.

Trying to fix some of these problems, I run into more problems. Some of them I don't see, but you tell them to me. I start arguing some Aubreyism ideas to others who agree with Elliotism, and learn Aubreyism well enough to win those arguments (although I have to relay back to you a few of their anti-Aubreyism arguments which I'm unable to answer myself. But the more more I do that, the more I pick up on how things work myself, eventually reaching full autonomy regarding Aubreyism). Others then help me with the task of reconciling various things with Aubreyism, such as the material in Popper's books. We do things like decide some parts can be rescued and figuring out how. Other parts have to be rejected, and we work through the implications of that and figure out where and why those implications stop. To do this well involves things like rereading books while keeping in mind some Aubreyism arguments and watching out for contradictions, and thus seeing the book material in a new way compared to prior readings with a different perspective. And it involves going back through thousands of things I and others wrote and using new Aubreyism knowledge to find errors, retract things, write new things about new positions, etc. The more Aubreyism has general principles, the better this will work – so I can find patterns in what has to change instead of dealing with individual cases.

OK, there's a story. Want to tell me a story where you change your mind?
I don’t think anyone does CR, and I also don’t think anyone does the slightly modified CR that you think you do. I think people do a triaged version of CR, and some people do the triaging better than others.