The test for an unbiased definition of a philosophical position is (surely) that it is equally acceptable to critics and defenders of the position. I think the definition of reductionism in the wiki blatantly fails this test. The same bias is apparent in the old Sequence posting dealing with reductionism. (Some comments called it a “straw man” without spelling out why.)

Consider the definition:-

Reductionism is a disbelief that the higher levels of simplified multilevel models are out there in the territory, that concepts constructed by mind in themselves play a role in the behaviour of reality. This doesn't contradict the notion that the concepts used in simplified multilevel models refer to the actual clusters of configurations of reality.

 

The unavoidable implication is that critics of reductionism believe that the higher levels of simplified multilevel models are out there in the territory.

Certainly, nobody but the flimsiest of straw men could possibly believe this, since all parts of the models are by definition not part of the territory: they are part of the map. It might be possible to believe, by contrast, that the territory actually has built into it things that correspond in some sense to higher levels of a hierarchical map, whether simplified or no; or that whether there are or are not such things is not decidable or meaningful; or one could believe that there are definitely no such things, that hierarchical higher levels of organisation (clusters) are meaningful only as mental artefacts.

The second disbelief, that concepts constructed by mind in themselves play a role in the behaviour of reality, also appears to be unimpeachable. However, by attaching this claim to the first claim, it is implied (without examination of the implication) that this statement only applies to the higher levels of simplified models. Yet logically, it cannot be confined to the higher levels but equally applies to the lowest level of the map: it is not the pixels of even the best possible map that themselves play a role in the behaviour of reality, but only something in some way corresponding to them.

In the Sequence discussion of reductionism we read:

So is the 747 made of something other than quarks?  No...

And bigjeff5 later comments

The territory is only quarks (or whatever quarks may be made of).

But baryons, which are part of the map, are made of quarks, which are therefore also part of the map. In fact they are the pixels of the best available map today. The map is not the territory. Therefore quarks are not part of the territory. So nothing is made of quarks except in our current map. To say that reality is made of quarks is an acceptable shorthand in many contexts, but in a discussion whose whole point is to emphasize the need not to confuse the map with the territory, disregarding the distinction at quark level is at least prima facie evidence of a biased approach!

TWO KINDS OF SIMPLIFIED MAPS

The wiki definition I began with includes the word simplified for reasons that are not clear. The Sequence discussion seems to me to confuse two different senses of the term: simplification by approximation and simplification by selection. Newton’s theory is now regarded as a simplified version of Special Relativity that serves as an excellent approximation in certain contexts. In this case the key is that the simplification is an approximation. Treating the forces on an aircraft wing at the aggregate level is leaving out internal details that per se do not affect the result. There will certainly be approximations involved, of course, but they don’t stem from the actual process of aggregation, which is essentially a matter of combining all the relevant force equations algebraically, eliminating internal forces, before solving them; rather than combining the calculated forces numerically. So is the definition addressing approximate maps or selective maps?

WHICH TERRITORY?

A question raised in the discussion but never answered as far as I can see is whether the belief referred to applies to our particular universe or to any universe one could conceive (so is more like a belief about the nature of explanation). While much of the discussion is focussed on our universe, various analogies advanced by commenters suggest that applicability to all universes is intended. I will assume the latter. A further confusion is that whereas the usual definition of reductionism refers to reduction of any system to its elements, and thus, for example, covers the reduction of lifeforms to their genetic recipes, the focus on “the territory” seems to confine the definition to physics: genes, being “made of quarks”, are just a level of the map.

DEFINITIONS

A definition that consists of disbelieving a contradiction in terms and then disguising a selective application of a truism, is clearly biased, and leads to apparently biased thinking, so I will attempt an unbiased definition.

Of course, we could simply import a definition from Wikipedia or elsewhere, but I am trying to capture the particular approach and terminology of this site (from initial impressions) in an unbiased way.

If I understand it correctly, reductionists on this site believe that, for the purposes of causal explanation, any “territory” in the sense of physical reality is best characterised as corresponding only to the lowest hierarchical level of our best map of it, higher levels of organisation existing only in the map. Is that right?

The summary in the SEQ_RERUN is also worth repeating:

We build models of the universe that have many different levels of description. But so far as anyone has been able to determine, the universe itself has only the single level of fundamental physics - reality doesn't explicitly compute protons, only quarks.

Which seems to mean much the same as my definition (if it means anything to say that “the universe computes”).

Psy-Kosh implicitly criticises the reference to quarks in claiming:

Reductionism does _NOT_ mean "reduction to particles", just "reduction to simple principles that are the basic thing that give rise to everything else".

But that doesn’t exclude “simple principles” that emerge from higher levels of organization, so doesn’t really fit the bill either.

The most relevant corresponding wording in Wikipedia is interesting because it makes no reference to the model/reality (map/territory) distinction which Eliezer seems to think makes ontological reductionism intelligible:

In a reductionist framework, [a phenomenon] that can be explained completely in terms of relations between other more fundamental phenomena  exerts no causal agency on the fundamental phenomena that explain it.

This assumes that the term “more fundamental” is defined, and (like my definition) doesn’t distinguish sequential causation from structural causation. Hmm, maybe this needs a separate post sometime.

FOOTNOTE: WHAT ABOUT THE HEURISTIC SENSE?

As Perplexed pointed out, the discussion of reductionism in the Sequences clearly refers to ontological reductionism by contrast with methodological reductionism. The same applies to the wiki definition and my proposed correction.

I entirely support this distinction. I am focussing here on ontological reductionism because so many contributors define themselves as “reductionist materialists” as a matter of belief. One would no more define oneself as a heuristic reductionist than one would define oneself as a hammer user (rather than a screwdriver user, say) as a matter of conviction.

TEASER

This quarrel about definition is not mere pedantry, since it hints at an unconscious bias. Moreover, if there are signs of agreement with this definition (or improved versions of it) and my newbie’s karma doesn’t suffer, I hope to build on this beginning to suggest that the real difference between proponents and supporters of ontological reductionism is that they are using two subtly different conceptions of what we mean by “the territory” and “the map”, both consistent with the definitions.  (Both conceptions have been implied by different contributors but without considering that the difference may be one of convention).

New to LessWrong?

New Comment
48 comments, sorted by Click to highlight new comments since: Today at 1:27 AM

We might do well to observe Dennett's distinction between "reductionism" and "greedy reductionism".

Reductionism says that higher-level phenomena can be explained in terms of their parts and the interactions between them. Greedy reductionism is when we notice a particular lower level (e.g. atoms) and attempt to explain a much higher level (e.g. minds) directly in terms of it, underestimating complexities which may be better explained with intermediate layers.

Reductionism about minds and atoms would say that minds are made of a (fantastically complex) arrangement of atoms; that you don't need to posit extra mind-stuff in order to explain the behavior of minds. But it would be greedy reductionism to say that particular features of minds can be explained directly in terms of atomic behavior — for instance the view of Epicurus (and some later philosophers...) that atom-level indeterminacy explains human freedom of choice.

I agree that's a good distinction, though direct explanation of a higher level obviously works in some cases (e.g. the weight of the brain is a simple aggregate of the weight of the constituent atoms).

"Can be explained in terms of.." seems a much less biased way of framing the definition to me than the one in the wiki. I'll add it to my list of starting points for discussion!

"Can be explained in terms of.." seems a much less biased way of framing the definition to me

It's also much more standard.

If I understand it correctly, reductionists on this site believe that, for the purposes of causal explanation, any “territory” in the sense of physical reality is best characterised as corresponding only to the lowest hierarchical level of our best map of it, higher levels of organisation existing only in the map. Is that right?

You see how you've basically characterized "reductionists on this site" as greedy reductionists?

I wasn't intending to unless that's what the Wiki definition characterizes it as, because I simply tried to re-express that definition without using the terms map and territory in ways that their definitions exclude.

I think perhaps I can see the problem. My phrase "for the purposes of causal explanation" is ambiguous. I wasn't meaning "as a way of explaining any particular behaviour" but rather "as a way of establishing the root causes underlying any behaviour". Does that make it more acceptable/less "greedy"?

Another possible cause of misunderstanding is that I have never seen the point of essentialism, in other words I think the questions that matter are always "how can something be most usefully described (for some stated purpose)" rather than "what is the essence of that something", so I instinctively avoided an essentialist definition. I'll think about rewording my version in a way that essentialists will recognize, as the sort of reductionism we are talking about does seem to hinge on reality being a sort of Platonic essence...

What made your characterization one of greedy reduction in my eyse was this

is best characterised as corresponding only to the lowest hierarchical level

Describe it at whatever level is most convenient. All levels are real to the extent that they model accurately.

I'm still puzzled, as you seem to be both defending and contradicting EY's view that:

the reductionist thesis is that we use multi-level models for computational reasons, but physical reality has only a single level. (Italics added).

I'm not actually attacking this view so much as regarding it as a particular convention or definition of reality rather than a "thesis".

Perhaps you are reading "best characterized as" as "best modelled as"? I'm not saying that, just that this is the sense of "reality" that EY/the wiki writer prefers to adopt.

Well, no, according to Dennett, it's more the disregard of complexity. Which woukd be an epistemological matter...you seem to be grasping for an ontological version.

The test for an unbiased definition of a philosophical position is (surely) that it is equally acceptable to critics and defenders of the position.

Disagree. That test is biased in favor of positions held by people who are better at strategically taking offense to definitions that are not biased in their favor.

I am trying to capture the particular approach and terminology of this site (from initial impressions) in an unbiased way.

If I understand it correctly, reductionists on this site believe that, for the purposes of causal explanation, any “territory” in the sense of physical reality is best characterised as corresponding only to the lowest hierarchical level of our best map of it, higher levels of organisation existing only in the map. Is that right?

You're strawmanning; we're not idiots. Saying that the universe directly computes quarks is the conjecture that our current map is fine enough that it contains objects referenced by the simple principles that give rise to everything else. I agree with Psy-Kosh that it was a mistake to mix that conjecture with an explanation of reductionism.

Sorry, I don't understand. Are you saying that you don't agree with my definition of reductionism (which was intended as a point of agreement, not a straw man at all)? I agree that an opinion about the likelihood that the standard model will continue to serve is a separate question.

Are you saying that you don't agree with my definition of reductionism?

Yes. Reductionism has nothing to do with how detailed our map is.

I find it hard to square that with the Sequence item referred to, but then you imply you also found it confused. So, what do you use the word to mean?

I have no objection to the definition given in the LW wiki.

That's using reductionism to mean materialism.

The unavoidable implication is that critics of reductionism believe that the higher levels of simplified multilevel models are out there in the territory.

Certainly, nobody but the flimsiest of straw men could possibly believe this

Have you checked to make sure this is the case? As far as I can tell, this is a common dualist position.

The answer is in the rest of the sentence that you truncated! I imagine a dualist would say that there is something out there in the territory which you consider to be a manifestation of a model higher level, but they don't. That isn't the same.

I don't find the monist/dualist distinction helpful. Computers have hardware and information: that's a dualist model, and it has served very well as a model. At any given instant, the information is a state of the hardware, which requires a monist model, but it's information that is downloaded etc. So is information "stuff"? Depends what you mean by "stuff". In short it's an argument about definitions in the bad sense of insisting about the "true" meaning.

Computers have hardware and information: that's a dualist model,

Not really, since the information/computation is always predicable from a sufficiently detailed description of the hardware physics.

The answer is in the rest of the sentence that you truncated!

It's not clear to me that's a complete response. It seems to be assuming that all instances of disagreement between reductionists and non-reductionists are linguistic, rather than causal. It looks to me like Bryan thinks that his qualia, like pain, are actually out there, and are not just very complicated ensembles of things reductionists like building models out of, like quarks and wavefunctions and so on.

It seems like attempting to resolve Bryan's disagreement with Robin about where Bryan's pain is located (B's "in my mind, not my brain" vs. R's "in your brain, which is also your mind") by claiming "well, of course Bryan's mental model of his pain doesn't exist in reality by definition" seems to be assuming definitions of 'mental model,' 'pain,' and 'reality' that I don't think Bryan would agree with.

Computers have hardware and information: that's a dualist model, and it has served very well as a model. At any given instant, the information is a state of the hardware, which requires a monist model, but it's information that is downloaded etc.

What do you mean when you say "but it's information that is downloaded"? That the monist model does not completely describe reality? That computer programming is easier with the dualist model than the monist model? That information lives in a nonphysical universe that communicates with the physical universe, such that only having the physical universe would be insufficient for computers to run?

Sorry no time for a full answer, but roughly, yes, in a sense I do think that many of these disagreements turn out to be linguistic if you dig far enough. But if they are causal, the definition needs to compare two intelligible models of causality, not define one in a self-contradictory way. My reply to buybuydandavis may also help clarify.

That computer programming is easier with the dualist model than the monist model?

Yes, and anything else that requires an intelligible account of what is going on. You start with a monist model and then you have to define something called or synonymous with information. In my understanding that makes it a dualist model. (I hope my draft next discussion, Karma permitting(!), will elucidate further.)

I don't see any evidence that information is "extra", ontologically.

From:

Sorry no time for a full answer, but roughly, yes, in a sense I do think that many of these disagreements turn out to be linguistic if you dig far enough.

and

Yes, and anything else that requires an intelligible account of what is going on. You start with a monist model and then you have to define something called or synonymous with information

it looks to me like you don't actually disagree with the definition of reductionism quoted in your post, and it seems like your primary concern is a combination of not being fair to critics of reductionism and a definition that doesn't distinguish between kinds of reductionism.

Those concerns are worth considering, but I think you're wrong on at least the first. The critics of reductionism that this post is targeted at are the Navy Gunners who think that GR and Newtonian Mechanics are different parts of the territory, not different maps that describe the territory at different levels of detail and completeness.

You mention someone calling the post attacking straw men, but I think that comment tree is worth rereading fully. Basically, this is what a worldview feels like from the inside- of course every sane person sees things this way, how could they not? But other people do sometimes have radically different worldviews, and sometimes they are radically confused about things.

Have you read EY's more recent post on reductionism (which may be clearer after reading the preceding posts)? I'm curious if that would help clarify where precisely you disagree.

Well, if the definition said that "reductionists disagree that 2 & 2 make 5" I wouldn't disagree with that either. What worries me is the apparent refusal to engage with the rational critics of reductionism. But I am mainly thinking of critics in fields other than physics - politics "there is no such thing as society", Skinner's psychology, "there are no thoughts, only stimuli and responses", not to mention developmental biology, weather forecasting & even mechanical engineering analysis, none of which actually get near "the territory" of quarks and leptons. So I am beginning to suspect that reductionism is used in a special sense by EY, more or less as a synonym for monism. And it's true, I wouldn't want to defend "substance dualists".

As for the Naval Gunner, the point is that he would be right in other fields than fundamental physics. In weather forecasting long term forecasts using coarser models are actually more accurate than those using fine meshes, because of the chaotic behaviour at smaller scales. So I would say the gunner was just misinformed! The fact that one of the two theories happens to be one of the very few theories that are exact as far as we currently know, and the other an approximation, makes it a special case - though possibly one of special relevance if monism/dualism is really the issue in question.

Thanks for pointing to the more recent EY post, which I look forward to reading. No time tonight.

What worries me is the apparent refusal to engage with the rational critics of reductionism. But I am mainly thinking of critics in fields other than physics - politics "there is no such thing as society", Skinner's psychology, "there are no thoughts, only stimuli and responses", not to mention developmental biology, weather forecasting & even mechanical engineering analysis, none of which actually get near "the territory" of quarks and leptons. So I am beginning to suspect that reductionism is used in a special sense by EY, more or less as a synonym for monism. And it's true, I wouldn't want to defend "substance dualists".

What you are calling reductionism here is the refusal to countenance some higher-level properties. But, in fact, most reductionists do countenance most h-l properties. What makes them reductionists (which of course is not brought out by the broken LW wiki definition) is that they think all the h-l properties they countenance can be explained at a lower level. BTW, people who don't countenance any h-l properties, states, or entities are called mereological nihiists, not reductionists.

As for the Naval Gunner, the point is that he would be right in other fields than fundamental physics. In weather forecasting long term forecasts using coarser models are actually more accurate than those using fine meshes, because of the chaotic behaviour at smaller scales.

I don't quite agree here. It's true that chaotic interactions and floating point multiplication errors mean that long-running fine-grained maps are less accurate than long-running coarse-grained maps, but it seems cleaner to consider that a fact about computer science, not meteorology.

Thanks for pointing to the more recent EY post, which I look forward to reading. No time tonight.

I would actually recommend Hands vs. Fingers first if you haven't read it yet. It's shorter and may be more directly relevant to your interests.

Re Hands vs. Fingers. What worries me about this is the lack of any attention to the different contexts/purposes of different statements about hands & fingers. I have added a comment to the original post to amplify this.

Thanks again.

it seems cleaner to consider that a fact about computer science, not meteorology.

I'd call it a fact about any system whose trajectories diverge at a smaller scale and converge at a larger scale (roughly), but that's a radical view that needs a new discussion some time.

I think I can see a useful way of taking the reductionism question further, but will do more reading first...

I mentioned back in April that the point about chaos and computer science needed a proper discussion. It is here.

I also mentioned another way of taking the reductionism question further. I was referring to this.

Re my claim:

"well, of course Bryan's mental model of his pain doesn't exist in reality by definition"

On reflection I suspect the disagreement here is that I am doubting that Bryan could consciously deny this, and you & EY & others are suspecting that he is unconsciously denying it. Well, that's a theory. I have added an edit to my post recognizing this. This seems to boil down to the LW-wiki "definition" not really defining what reductionists believe, but rather defining why they believe certain criticisms of reductionism are wrong. That at least would explain why it sounds biased!

What do you mean when you say "but it's information that is downloaded"? That the monist model does not completely describe reality? That computer programming is easier with the dualist model than the monist model? That information lives in a nonphysical universe that communicates with the physical universe, such that only having the physical universe would be insufficient for computers to run?

To answer more fully: The 'monist' model without information as a category describes reality at any instant but does not describe what is conserved from one instant to the next. Any activity that requires an intelligible account of what is going on is easier with the concept of information as a separate "thing". Of course, information doesn't belong in a nonphysical universe, since it obeys physical laws. Nevertheless the fact that it has a life of its own, with laws distinct to the laws specific to the materials which embody it an any given time, give it part (but not all) of the character of a separate physical but intangible substance.

The point of my analogy was to emphasise that all categories are man-made, including "substance", so that "substance counting" has an element of arbitrariness. Actually I don't find that treating "mind" as a separate substance is helpful!

To answer more fully: The 'monist' model without information as a category describes reality at any instant but does not describe what is conserved from one instant to the next.

If you mean infromation, it is not clear that that is conserved. And I don't see how a sufficiently detailed description of reality at the quark level could fail to describe all the information.

OK, not strictly "conserved", except that I understand quantum mechanics requires that the information in the universe must be conserved. But what I meant is that if you download a file to a different medium and then delete the original, the information is still the same although the descriptions at quark level are utterly different. Thus there is a sense in which a quark level description of reality fails to capture an important fact about it (the identity of the two files in information terms).

I don't think this has anything to do with dualism in the Cartesian sense, it's just an example of my general preference for not taking metaphysical positions without reference to the context. I'm afraid I don't know the label for that!

OK, not strictly "conserved", except that I understand quantum mechanics requires that the information in the universe must be conserved

..absent collapse..

But what I meant is that if you download a file to a different medium and then delete the original, the information is still the same although the descriptions at quark level are utterly different.

But a 4D descriptions of al the changes involved in the copy-and-delete process would be sufficient to show that the information in the first medium is equivalent to the information in the second. In fact, your problem would be false positives, since determinism will always show that subsequent state contains the same information as a previous one.

..absent collapse..

Ah, is that so.

But a 4D descriptions of all the changes involved in the copy-and-delete process would be sufficient..

Yes, I can see that that's one way of looking at it.

In fact, your problem would be false positives

I don't think so, since the information I would be comparing in this case (the "file contents") would be just a reduction of the information in two regions of space-time.

I don't think so, since the information I would be comparing in this case (the "file contents") would be just a reduction of the information in two regions of space-time.

And under determinsim, all the information in any spatial slice will be reproduced throughout time. Hence the false positives.

I'm not clear what you are meaning by "spatial slice". That sounds like all of space at a particular moment in time. In speaking of a space-time region I am speaking of a small amount of space (e.g. that occupied by one file on a hard drive) at a particular moment in time.

Your can prove conservation of information over small space times volumes without positing information as an ontological extra ingredient. You will also get false positives over larger space time volumes.

Computers have hardware and information: that's a dualist model

Actually, it's a trialist model, or worse. Something or someone had to create and program the computers and push the start button.

;) ... but that's still only matter and information, just that we're now just information....

Dualists are not necessarily emergentists.

Emergentists typically view emergent higher level properties as unpredictable from the smallest constituents of a system and some minimal set of their relations. Which is more or less to say that the h-l property contains more information than its base. So there is little evidence that emergentists are engaging in map-territory confusion, or trying to reify simple h-l properties.

Recommendation: Don't put titles or second headings in all caps. Use the style formatter (the leftmost button on the post editor- it will read 'paragraph' at first) and use Heading 2 for your section headings.

(I've uncapslocked the title.)

Thanks! (Both of you)

Why do I care about having "unbiased" definitions?

In the same way that it's a very good exercise when having a rational debate to start by each side paraphrasing the view they oppose in a manner that is acceptable to holders of that view, otherwise the chances are you haven't understood what is being said. Is "a dialogue of the deaf" really what you want?

[-][anonymous]11y00

I like to say that a reductionist is one who believes simpler explanations are superior to more complex explanations, all other things being equal.

[This comment is no longer endorsed by its author]Reply

An "emergentist" would probably define their view in the same way. The question is, simpler for what purposes?

[-][anonymous]11y20

For that very reason, I also tend not to use the word "reductionist".

So my objections aren't aimed at you!

Why? That's a believe in occam's razor.