Hard to Vary or Hardly Usable?
Showing only #3716 and its comments.
See full discussionLog in or sign up to participate in this discussion.
With an account, you can revise, criticize, and comment on ideas.Deutsch’s stance in my own words:
The distinguishing characteristic between rationality and irrationality is that rationality is the search for good explanations. All progress comes from the search for good explanations. So the distinction between good vs explanations is epistemologically fundamental.
A good explanation is hard to vary “while still accounting for what it purports to account for.” (BoI chapter 1 glossary.) A bad explanation is easy to vary.
For example, the Persephone myth as an explanation of the seasons is easy to change without impacting its ability to explain the seasons. You could arbitrarily replace Persephone and other characters and the explanation would still ‘work’. The axis-tilt explanation of the earth, on the other hand, is hard to change without breaking it. You can’t just replace the axis with something else, say.
The quality of a theory is a matter of degrees. The harder it is to change a theory, the better that theory is. When deciding which explanation to adopt, we should “choose between [explanations] according to how good they are…: how hard to vary.” (BoI chatper 9; see similar remark in chapter 8.)
Deutsch leaves open how we find out how hard to vary an explanation is. We need more details. In some cases it’s obvious, but we need a general description for less-obvious cases.
Deutsch contradicts his yardstick for understanding a computational task. He says that you haven’t understood a computational task if you can’t program it. His method of decision-making based on finding good explanations is a computational task. He can’t program it, so he hasn’t understood it.
Isn’t this basically asking for a specification of the creative program? Isn’t this effectively an AGI project?
No, see #3706. I’m open to user input (within reason). That covers creative parts. The non-creative parts can be automated by definition.
Isn’t this asking for a formalization of creativity, which is impossible?
No, it’s asking for a formalization of rational decision-making, which is a related but separate issue. Given a set of explanations (after they’ve already been created), what non-creative sorting algorithm do we use to find the best one?
Popper formalized much of his epistemology, such as the notions of empirical content and degrees of falsifiability. Why hold Deutsch to a different standard? Why couldn’t he formalize the steps for finding the quality of a given explanation?
Maybe Deutsch just means hard to vary as a heuristic, not as a full-fledged decision-making algorithm.
A heuristic or heuristic technique (problem solving, mental shortcut, rule of thumb) is any approach to problem solving that employs a pragmatic method that is not fully optimized, perfected, or rationalized, but is nevertheless "good enough" as an approximation or attribute substitution.
None of this means a heuristic couldn’t be programmed. On the contrary, heuristics sound easier to program than full-fledged, ‘proper’ algorithms.
I’d be happy to see some pseudo-code that uses workarounds/heuristics. That’d be a fine starting point.
@dirk-meulenbelt suggested in a space (at 21:30) that a bunch of epistemology is underspecified. There are many epistemological concepts (like criterion of democracy, falsifiability, etc.) that we don’t know enough about to express in code.
Yes, many ideas fail Deutsch’s yardstick. But so what? That doesn’t make things better.
Deutsch’s yardstick applies to computational tasks. It’s not meant for other things. It’s not clear to me that the criterion of democracy is a computational task.
Yes, the criterion for democracy is not a computational task, but an abstraction that constrains computational tasks. In the same way: the criterion for HTV is also not a computational task, it constrains the possible computational tasks that attempt to quantify HTV.
We understand computational tasks by being able to program them (as per Deutsch' criterion). But we understand criteria/ principles/ axioms/ theories ... (non computational tasks) in another way: by varying them and eliminating the variants that do not solve the problem the principle purported to solve.
For example:
a+b=b+a (in arithmetic) is a principle/ axiom that we understand by elimination of possible variants (a+b =/= a ... a+b =/=b ... etc)
but 3+5=5+3 is a specific transformation that should be understood via a computational task: adding 5 to 3 and then 3 to 5 and comparing both outcomes, via a program.
So my criticism is that the HTV criterion is not a computational task (but a principle, universal statement) and Deutsch's criterion of understanding (you need a program) only applies to computational tasks.
With principle/ universal statement/ theory, I mean for example: for all masses, there is a force proportional to the inverse square of their distances/ for all integers, addition is commutative/ for all species, their evolution is governed by variation and selection, for all interpretations of moral actions, they are moral relativism when ... applies to that interpretation/ ....
- Principles/ universal statements/ theories are not computable because they speak about sets of (possible) transformations (not 1 in particular which would be a computation) and they offer a constraining criterion to those transformations in the set.
- Whereas a computer program is an abstraction capable of causing 1 particular transformation (between sets of inputs and sets of outputs)
There may be a way to quantify HTV, and thus deal with specific evaluations of how HTV of one theory is higher than another. That would be a computational task. But that is different from the criterion for HTV (which is by definition not computable). And having no program for that computational task does not imply that the criterion for HTV is irrelevant or not usable, or even fluff.
Compare for example to the theory of evolution: the theory of "variation and selection" is the criterion for a set of allowable transformations (of species), but not having a specific program (e.g. for how a particular species can evolve in some particular niche) does not imply that the criterion is useless or fluff.
I think the usefulness of the HTV criterion becomes clear when you link it to Constructor Theory, then one can argue that HTV criterion adds more than criticisms alone can do. But that's a whole other story we could get into.
You criticized your own idea. Presumably that’s not what you meant to do.
HTV isn’t a principle even by your own definition. What on earth are you talking about man.
Even if HTV itself is not a computational task, the decision-making method Deutsch proposes is one, and it depends on HTV. But even if we sidestep that issue and outsource HTV completely to the user, we still run into all kinds of issues. This has all been addressed. No fancy talk about sets or constraints is going to change that.
You previously claimed you’re an engineer. I don’t think you are. You just pasted some code that was clearly written by AI and didn’t even compile, twice.
You talk about ‘sets’ and ‘constraints’ and ‘computations’ but I don’t think you understand any of them. No offense but I think those concepts are all distractions so you don’t need to actually address HTV. That’s why you need to use those big words.
Discussing with you is a waste of time. Again, no offense but I don’t think you’re qualified to weigh in on this discussion. Prove me wrong and submit working, handwritten code for HTV or Deutsch’s decision-making method. I’ll delete any further comments from you in this discussion that don’t contain working code. If you keep commenting anyway, I’ll lock your account.
As I write in my article:
… Popper did formalize/specify much of his epistemology, such as the notions of empirical content and degrees of falsifiability. So why couldn’t Deutsch formalize the steps for finding the quality of a given explanation?
Liberty said (at 1:38:39) hard to vary isn’t a method of decision-making. It’s a factor people take into account when they make decisions, but decision-making itself is a creative process.
I’m not saying hard to vary is a decision-making method. I’m saying it’s an integral part of Deutsch’s decision-making method. As I write in my article:
He argues that “we should choose between [explanations] according to how good they are…: how hard to vary.”
Liberty responded (1:39:46) that that quote is misleading because it makes it sound like hard to vary is the only criterion people use when making decisions, which can’t be true. There are other criteria, like “consistency with data”, “logical consistency”, “fitting in with existing theories”, etc.
The quote may be false, but I don’t see how it’s misleading. I’m not quoting Deutsch in isolation or cherry-picking information or anything like that.
It does sound like Deutsch thinks all these different criteria boil down to being about hard vs easy to vary, see #3814.
I’m fine allowing user input to sidestep the creativity problem, see #3802.
Deutsch says to choose between explanations “according to how good they are” – note the plural.
What if I can only come up with one explanation? Can I just go with that one? What if it’s bad but still the best I could do? He leaves such questions open.
Deutsch says rationality means seeking good explanations, so without a step-by-step guide on how to seek good explanations, we cannot know when we are being irrational. That’s bad for error correction.
From my article:
[D]epending on context, being hard to change can be a bad thing. For example, ‘tight coupling’ is a reason software can be hard to change, and it’s considered bad because it reduces maintainability.
@edwin-de-wit recently asked Deutsch about this:
… hard to vary can also be a property that you don’t want. Sometimes they say that complicated systems are just so entangled that they’re no longer changeable. You could call that hard to vary but it’s not a property you pursue.
Deutsch’s reply:
Quite so. So hard to vary is about explanations, not about end products. So the end product should be easy to vary, if you want to. But it should be hard to vary in the sense that you don’t want to.
Explanations are functions. In other words, explanations are software. There’s no sharp distinction between explanations and end products here. So the concern about tight coupling still applies.
But HTV is also about art (BoI chapter 14).
A piece of art is a kind of end product. And we could write code to produce art.
Naval, who’s inspired by Deutsch, says good products are hard to vary, too: https://nav.al/good-products
Gives iPhone as example.
From my article:
[I]sn’t the difficulty of changing an explanation at least partly a property not of the explanation itself but of whoever is trying to change it? If I’m having difficulty changing it, maybe that’s because I lack imagination. Or maybe I’m just new to that field and an expert could easily change it. In which case the difficulty of changing an explanation is, again, not an objective property of that explanation but a subjective property of its critics. How could subjective properties be epistemologically fundamental?
fundamental
@zelalem-mekonnen suggested during a space (37:36) that hard to vary is just one mode of criticism.
Not according to Deutsch. He says hard to vary is epistemologically fundamental, that all progress is based on it. For example, he phrases testability in terms of hard to vary (BoI chapter 1):
When a formerly good explanation has been falsified by new observations, it is no longer a good explanation, because the problem has expanded to include those observations. Thus the standard scientific methodology of dropping theories when refuted by experiment is implied by the requirement for good explanations.
He also says that “good explanations [are] essential to science…” (thanks @tom-nassis for finding this quote). Recall that a good explanation is one that is hard to vary.
For Deutsch, hard to vary is the key mode of criticism, not just one of many.
From my article:
Isn’t the assignment of positive scores, of positive reasons to prefer one theory over another, a kind of justificationism? Deutsch criticizes justificationism throughout The Beginning of Infinity, but isn’t an endorsement of a theory as ‘good’ a kind of justification?
@lola-trimble suggested during a space that a theory is hard to vary if it’s not easy to vary. So the maximum score would be 0, not +1,000 or whatever. In which case ‘hard to vary’ isn’t an endorsement.
Even so, if a criticism gets score -10, that will push the parent theory’s score above 0.
But then the ease with which a criticism could be varied might have no effect on its parent. So why even bother having a notion of ‘easiness to vary’ at that point?
But calling a theory ‘good’ sounds like an endorsement. Deutsch also writes (BoI chapter 10) that a “superb” theory is “exceedingly hard to vary”. Ultimately we’d have to ask him, but for now, given the strength and positivity of those terms, I think it’s fair to conclude that he means ‘hard to vary’ as an endorsement.
The criterion for HTV applied to 2 explanation is not justificationism I think. It allows to say explanation A is better than explanation B, which is equivalent to: explanation B is worse than explanation A. So it is a relative claim about an explanation, relative to another, not versus some absolute criterion of goodness. Similar to a crucial test (e.g. Eddington): we refute Newton's theory and keep Einsteins, that is not a claim about the goodness of Einsteins theory, that theory merely has survived, it has not gotten "goodness points". It could be refuted always later on by any better theory, in which case we would drop it too.
Similar to a crucial test …
But that’s exactly where HTV differs from Popper. Popper doesn’t give a theory points when it survives a crucial test. HTV does. From BoI chapter 1:
… testable explanations that have passed stringent tests become extremely good explanations …
That's because a good explanation for Deutsch is not an explanation with good points, but an explanation that is harder to vary compared to any other explanation. So again relative to other explanations.
The word "good" is indeed misleading in that sense, but he clearly qualifies it as performing better, relative to other explanations, on his HTV criterion, and not as: the explanation having scored high points.
with good points
I didn’t say the explanation doesn’t make good points, I said the explanation doesn’t get points.
So it is a relative claim about an explanation, relative to another, not versus some absolute criterion of goodness.
So what? I didn’t mention an absolute criterion. My original criticism already applies to both relative and absolute criteria of quality (what you call “goodness”).
Because relative criteria are fine to posit and not justificationist. We can propose criteria that claim that explanation A is better than explanation B without that being justificationism
From BoI chapter 1 glossary:
The misconception that knowledge can be genuine or reliable only if it is justified by some source or criterion.
That says nothing about absolute vs relative. Stop making up stuff.
@liberty-fitz-claridge says (#3885) it’d be implausible for HTV to be justificationist since that would contradict the rest of Deutsch’s anti-justificationist philosophy.
I don’t think that alone means my interpretation of HTV is implausible. We’re bound to find contradictions eventually. In a good book like BoI, they’re just rare, so when we do find them, they go against the bulk of the philosophy.
Our explanations do get better the more criticisms we address, but Deutsch has it backwards: the increasing quality of an explanation is the result of critical activity, not its means.
From my article:
[T]he assignment of positive values enables self-coercion: if I have a ‘good’ explanation worth 500 points, and a criticism worth only 100 points, Deutsch’s epistemology (presumably) says to adopt the explanation even though it has a pending criticism. After all, we’re still 400 in the black! But according to the epistemology of Taking Children Seriously, a parenting philosophy Deutsch cofounded before writing The Beginning of Infinity, acting on an idea that has pending criticisms is the definition of self-coercion. Such an act is irrational and incompatible with his view that rationality is fun in the sense that rationality means unanimous consent between explicit, inexplicit, unconscious, and any other type of idea in one’s mind.
In short, does the search for good explanations enable self-coercion and contradict TCS?
Even if we allow creative user input, eg a score for the quality of an explanation, we run into all kinds of open questions, such as what upper and lower limits to use for the score, and unexpected behavior, such as criticisms pushing an explanation’s score beyond those limits.