Blog Coordinator

« Criterial causation overcomes the luck argument against MR | Main | Very Bad Wizards 59: Tumors All The Way Down (With Sam Harris) »



Feed You can follow this conversation by subscribing to the comment feed for this post.

The counterfactual analysis you provide here is not an interventionist account of causation. Instead, this is something like the simple conditional account of causation that Lewis held early in his career. Because of multiple counterexamples to this account, even it's chief supporter (Lewis) later abandoned it in favor of an interventionist account. What changes an account from the simple conditional to an interventionist is that the simple conditional account only concentrates on whether the consequent would occur. Interventionist accounts add that how, when, whether the antecedent occurs influences how, when, whether the consequent occurs(often other differences in the antecedents and consequents are also added to interventionist accounts as well).

Given your dislike for compatibilist interpretations of abilities I find it a bit surprising for you to be supportive of interventionist accounts of causation. It is precisely the idea that we ought to be thinking of abilities as causal combined with the simple conditional account (and later the interventionist accounts) that leads Lewis to his compatibilism and many early attempts to reject Lewis's compatibilism stemmed from a rejection of the core of interventionist interpretations of causation.

Also, notice, if you accept these accounts of causation you cannot claim that you are only concerned with "real possibilities" meaning those where the past and the laws are held fixed. After all, in a deterministic world there is only one set of events that are physically possible given the same initial conditions. Combining this with an interventionist account of causation and the claim that we can only consider possible worlds with the same laws and past as our own denies us the ability to vary any of the antecedents without producing trivially true conditionals. So, on an interventionist account, neither causation nor causal reasoning is possible in a deterministic world (if we are only allowed to consider your 'real possibilities). In fact, the extreme success of interventionist accounts of causation both in the sciences and in philosophy is probably the best argument for compatibilism about free will. After all, all compatibilists are claiming is that abilities ought to be analyzed in terms of interventionist accounts of causation.

Maybe I'm missing something, but from your posts it seems to me that you're a compatibilist and just don't know it yet. I think you ascribe to the compatibilist a more magical ability than they claim agents posses. All we are claiming is that abilities are a causal concept and that causation ought to be evaluated in terms of interentionist accounts of causation.

Charles, thank you for clarifying the difference between a Lewis-like counterfactual account of causation and an interventionist one.

Note that at least as far back as J.S. Mill (1843), it has been assumed that it is a necessary condition of a causal relationship that the timing of a cause be correlated with or covary with the timing of an effect. This may hold under the traditional interventionist view (though I do not know whether this, Mill's claim, has been logically proven). But this need not hold under the class of interventions that fall under the heading 'reparameterizations of the criteria that B uses to evaluate inputs from A' since such criteria can 'lie in waiting' for the right inputs from A (and/or A' and/or A'' etc.) for unspecifiable durations of time. So the Millian notion of necessary temporal covariation in time between causes and effects falls apart for this subclass of causation. Temporal covariation is not necessary for criterial causation.

There is of course an ancient assumption that a cause must precede its effects. Again, this may not hold for the class of causation I am discussing here under the name 'criterial causation.' Traditional (non-criterialist) interventionist accounts of causation might also have implicit in them that causation always operates over time, such that an intervention on A has only subsequent effects on B (and B', B'' etc.). In contrast, some situations appear to involve 'causal knots' that are locked together in one moment holistically (that is they cannot be reduced to local interactions among particles without taking global relationships into account). For example, each of the three poles of a teepee holds the other two up. Does one cause the others not to fall, or do they cause it not to fall? Well, both are true. Cause becomes inseparable from effect in such cases. There is a pattern of relationships that exists in a moment over space. So it is not the case that causal relationships necessarily have to play out over time where A causes B later which causes C later... Similarly, which part of a bubble causes it to exist as a bubble? Causation in a bubble is a causal knot where all parts simultaneously cause all others to exist in a bubble (i.e. not to pop). More relevant to the brain, when a certain pattern of inputs from neuron A satisfies criteria for neuron B's firing, is the cause of B's subsequent firing the inputs from A, the criteria that B places on A's inputs, or both, namely the relationship (the match) between them? I would argue both, so this again would count as a type of causal knot in which the simple kind of temporal sequencing or subsequence that one finds in colliding billiard balls is not the case. I would argue that models of causation need to go beyond the implicit spatial localism and temporal linearity of traditional interventionist models, and structural equation models, to incorporate global patterns and holistic causal relationships in space and time. Such patterns need not look anything like the linear and independent relationship among variables modeled using multiple regression, path analysis, factor analysis, or structural equations.

Concerning compatibilism I am indeed a compatibilist if we define FW in the Humean manner (the ability to do as we want uncoerced by outside forces). I have argued this month that everyone would be a compatibilist if that is our definition. But if we define FW as the ability to agentically shape what future becomes real by virtue of the choices we make, which might not have actually happened, had we decided otherwise, then I would be an incompatibilist, as would everyone else, by definition. I realize that you and other compatibilists here seem to bristle at this point. But I just want to emphasize that, to me at least, there seem to be essentially two debates going on in the philosophy of FW, one semantic and the other metaphysical. The former concerns the correct definition of terms 'FW,' 'can,' and 'possible.' The other concerns what must be the case to have FW and MR, and whether FW and MR are compatible with determinism or not.

Regarding your final point, are you saying that accepting an interventionist account of causation necessitates the conclusion that compatibilism is true? Does Lewis say that? It would seem to me that accepting an interventionist account would be consistent with either compatibilism or incompatibilism. If I am missing the link you are making, please unpack it more for me so I can understand if there is necessity in the link.

Charles, let me also add that I regard Judea Pearl's arguments about the nature of causation, like James Woodward’s, as idealizations of the assumptions and methods that drive the Sciences, namely, that multiple variables can each partially drive an effect (or dependent variable), which are assumed for simplicity to be independently variable (indeed that we call 'independent variables'), plus the addition of a noise term to account for all the other potentially causal factors not captured by an equation. This is formalized in the Social Sciences using multiple linear regression, which is a special case of structural equation modeling. On this account, something is causal of a multiply caused effect if changing it changes the regression outcome.

For example, Pearl writes many equations of this basic type:

B-sub-i = f-sub-i((parent-cause-A-sub-i of B-sub-i), error-terms-sub-i) where i goes from 1 to any desired number n of possible causal influences on an outcome or dependent variable B.

In other words, B is a dependent variable driven by lots of independent variables or causal factors.

This simplification definitely works for the experimental sciences, but it has lots of assumptions built into it that are generally not true in the world (e.g. of neurons/Neuroscience or social interactions/Economics), including independence of variables, and linearity. In real physical systems there are lots of non-linearities such as winner-take-all steps and sudden Thomian phase transitions that are hard to model. That is why the Social Sciences typically assume independence and linearity, at least within some small regimen where these appear to hold. One problem is that there are countless ways to behave non-linearly whereas there are only very limited ways to behave linearly, so it is easiest to assume linearity and independenceof variables. Plus the math is much easier in the linear domain. In the Social Sciences people don't typically care that these assumptions are not valid, because they are good enough to account for most outcomes within the limited experimental domain tested, where these assumptions hopefully hold. But I would think that philosophers are not just aiming for a 'good enough' account of causation like the one we settle for in the empirical end of things. I would add to that, that the sciences also need to do a better job of modeling causal non-linearities and non-independencies.

Also, let me add some of my worries about basing a philosophical account of causation on what we do in the sciences, such as, I believe, Woodward and Pearl are doing. A key point that worries me about the rising popularity of structural equation models in the sciences and now too in philosophy is that they might still be rooted in a quasi-Newtonian conception of causation as energy or momentum transfer among particles over time (see, e.g., P. Dowe’s views on causation, 1992). The idea is that we change some inputs to a system and watch for subsequent changes in its behavior. But a physical system like the brain might not change its inputs, but instead may change parameters for how inputs will be responded to. The intervention might not be on the input side, but on the decoding or response side. For example, a traditional interventionist might say, "how might the game of baseball be affected if we change the weight of a baseball to ten pounds?" A criterialist like myself might say, how might the game of baseball change if we change what will count as a ground rule double? Sure, both are interventions, but the first kind looks at physical manipulations whereas the second is about changing informational parameters. Because informational causation is fundamentally about successions of acts of decoding, informational causation will be especially sensitive to changing the code used to decode input. That is, informational causation is likely to be especially manipulable given reparameterizations of B, holding inputs from A and all other input variables constant.

I am in general fascinated by the degree to which metaphors help and hinder our thinking (a la Lakoff and Johnson’s excellent book ‘Metaphors we live by’). I would argue that there is a quasi-Newtonian metaphor underlying Judea Pearl’s thinking as described in his 2000 book ‘Causality’ and also in James Woodward’s thinking in his 2003 book. Even the language they use has a built-in metaphor that both guides their/our thinking while also potentially blinding them/us to alternative, more general conceptions of causation. For example, Pearl calls A a ‘parent cause’ whereupon it follows that B is a 'child effect.' Well, parents precede children. In addition, this metaphor leads one to think of B as a passive consequence of A because children have no say in the matter of who will count as their parents. This metaphor can blind us to the possibility that in the domain of real world causation B might in fact have some say in the matter of which inputs become its 'parent cause.'

What is missing from the traditional interventionist conception of causation is the fact that B is not passively driven by those independent variables. It can be parameterized to respond to those various causes or not, or to respond to them in different ways, depending on what the particularities of those causal inputs from the ‘parent-causes’ are. I have argued that this kind of causation is fundamentally non-classical and non-Newtonian. It de-emphasizes the importance of amounts of energy or frequencies of energy or forces, as in Newton’s picture of causation, and instead emphasizes the causal importance of phase relationships in energy such as shape and pattern and timing. This is why I have variously used and maybe even coined terms like ‘phase causation,’ ‘pattern causation’ or ‘criterial causation.’

Certainly, accounts of causation that date back as far as Mill's may have some of the problems that you address here, but none of the major contenders in the causation literature still suffer from these issues. " There is of course an ancient assumption that a cause must precede its effects. Again, this may not hold for the class of causation I am discussing here under the name 'criterial causation.' Traditional (non-criterialist) interventionist accounts of causation might also have implicit in them that causation always operates over time, such that an intervention on A has only subsequent effects on B (and B', B'' etc.). In contrast, some situations appear to involve 'causal knots' that are locked together in one moment holistically" Even in the simple conditional analysis forerunner of interventionist models of causation people in the causation literature considered it to be a necessary condition on any adequate account of causation that it did not build these temporal aspects into the concept. When Lewis defended the simple counterfactual account of causation he was quite explicit that we could not build a temporal aspect into the analysis because doing so would make backwards causation impossible and we ought not legislate from the armchair that things taken seriously in science are impossible. Some versions of transfer accounts of causation have been dismissed in the literature precisely because they build into their analysis aspects that do not take seriously the sorts of problems that you address. There are also quite a few cases discussed in the causation literature about simultaneous causation....including the cases you address here. All major accounts of causation consider it necessary to accommodate these sorts of causation.

I certainly agree that accounts like Mill's and Hume's incorporate too much of our pre-theoretic intuitions about causation and build them into the analysis....but the problems you address here are ones of earlier views of causation and not of ones that are taken seriously in the literature today. In fact, if it could be shown that an analysis of causation does not accommodate the sorts of cases you address here, it would be dismissed in the causation literature (at least since Lewis's early writings....probably a lot earlier).

I think the mistake you are making in thinking of the issue of compatibilism or incompatibilism as a matter of how we stipulate definitions of terms is one that neither compatibilists or incompatibilsits in the literature would accept. What we are arguing about isn't merely a use of terms. There are debates in philosophy that some people argue are merely terminological debates. The chief example here is whether a world that consists of only two 'atoms' in the traditional sense is a world with two objects or three. Is there a third object that is composed of merely these two indivisible particles? Many people think that there is no substance to this debate and it is purely about how we define our terms. If this is the case in the free will literature, both compatiblists and incompatibilists lose.

Instead, what participants in the debate assume is that we are arguing about the same thing. We are trying to figure out what agent's abilities are, provide an analysis of them, and then ask what they say about free will. The reason the debate is not about merely how we use the word 'ability' is because we assume that there are such things that we are trying to provide an account of. The reason that compatibilists and incompatibilists disagree is because we disagree about what abilities are.

I find it truly shocking that almost everyone isn't a compatibilist....except for the fact that many of the people who work in free will spend very little time acquainting themselves with the causation literature. Lewis's project started with giving one of the first plausible semantics for counterfactuals. From there, he attempted to define causation in terms of counterfactuals (a project that ultimately failed, but lead him to also start the interventionist accounts of causation in philosophy.) In free will, he shows how if we adopt something like a counterfactual or an interventionist account of causation, assume that causation is possible in deterministic worlds, and assume that abilities are causal concepts we are left with a compatibilist view. Notice, in order to alter the antecedents of a counterfactual in a deterministic world we must consider worlds with different laws of a different past than our own. After all, considering any world where the antecedent of a counterfactual has a different truth value than what it in fact has would require that the laws or the past be altered. So, if causation is related to counterfactuals in the manner that interventionist accounts suggest then analyzing any causal claim will require considering worlds with different laws or a different past than our own in a deterministic world. When compatibilists offer their accounts of abilities all that they are doing is offering something like an interventionist or a simple conditional account of causation for abilities. Yet, incompatibilists tend to reject the compatibilist analysis of abilities. To do so, however, requires either rejecting the interventionist accounts of causation that compatibilists use to analyze abilities, reject that abilities are a causal notion or reject that causation is possible in a deterministic world. Alternatively, an incompatibilist could accept standard compatibilist accounts of abilities and then try to show why despite appearances they do not lead to compatibilism. (Yet, to my knowledge, no incompatibilist has taken this route yet). Once you grant a Lewis style account of causation (either simple conditional or interventionist) it seems like you are committed to the compatibilist interpretation of ability. After all, all compatibilists are merely deriving what abilities are from these accounts of causation.

As far as the point about accounts like Woodward's assuming something like causation as energy or momentum transfer, this is certainly not the case. There are transfer accounts of causation (Russell, Salmon, Ehring, etc.). The earliest interventionist accounts were intended to be in opposition to these accounts. Lewis saw both his simple conditional analysis and his later interventionist approach as reductive accounts of causation. As a reductive account, Lewis thought there was nothing more to causation than what these accounts provide. So, the addition of transfer issues is irrelevant for causation. There is a pretty massive literature on debates between early interventionists (and simple conditional accounts) with their opponents who accept the transfer accounts. One of the things that made Woodward's account very beneficial in the development of interventionist accounts is that he abandons the reductive strategy of Lewis. So, Woodward's account can remain non-committal about these transfer issues. Depending on what stripe of interventionist you are, these accounts are either inconsistent with transfer accounts or they are non-commital about transfer accounts. Certainly, interventionists do not presuppose a transfer account. The earliest interventionists explicitly rejected these transfer accounts. Woodward expands the interest of interventionist accounts in the causation literature by abandoning the reductive aspect of these accounts and thereby no longer making them directly opposed to transfer accounts. Today, there certainly are some theorists attracted to interventionist accounts who are also attracted to the transfer intuitions that you mention...yet they are probably in the minority of people attracted to interventionist accounts because it is not too long ago that these accounts were seen to be in direct competition.

Also, seeing interventionist accounts of causation as being patterned off of work on causation in the sciences seems historically inaccurate. These accounts began (in philosophy) as responses to problems with the simple conditional accounts that really took off with the development of a semantics that could finally produce a logic for counterfactuals. Later interventionists saw the parallels between what was being developed in different sciences and the interventionist account that was being developed as a response to counterexamples to the simple conditional account of causation and thought that the fact that different fields seem to be converging on something similar yet doing so for drastically different reasons as evidence for interventionist accounts.

Charles, sorry for taking so many days to reply to your interesting comments. I was swamped with grantwriting.

I am impressed by your wide and deep knowledge of the history of theories of causation. Tell me please, has anyone else written about the kind of causation I think is so central to information-processing in the brain, namely causation via reparameterization of how decoders will respond to future input? If so, please tell me who I should be reading and referencing. I have read Woodward and Pearl and some of the transfer theorists and find nothing like that. Also, I see nothing like criterial causation in the early (simple conditional) or later (interventionist) accounts of causation by Lewis. All these accounts seem to primarily focus on intervening on A and looking then for effects on B, as opposed to intervening on B, changing how it will respond to future inputs from A and other nodes. I am not saying at all that these existing accounts of causation cannot be amended to accommodate this type of causation and become more complete accounts. I am just saying that these various existing interventionist/manipulationist accounts are incomplete for leaving this central (at least in biology) type of causation out of the picture. Or, am I mistaken about this? Is there an existing literature on causation via resetting of criteria for the release of future acts or responses? Thanks.

I will answer your other point about compatibilism on the other thread where I argue that the exclusion argument forces one to take an incompatibilist stance, since it is more pertinent to that thread.

Hi, Peter!

I was wondering if you could comment on something either in this thread or in one of your upcoming threads. You had stated earlier that you were going to be doing some more threads related to Libet-type stuff.

Peter, we've talked a lot about Libet in relation to Free Will, the readiness potential, different experiments, interpretations, etc.

So I'll take this question in a slightly different direction: What are your thoughts about Libet's whole idea of "Free Won't"?

I have to admit that, philosophically and scientifically speaking, "Free Won't" has never made an ounce of sense to me. (We're supposed to be able to "veto" ourselves! And we're supposed to be able to do this at the LAST POSSIBLE SECOND, and with some form of consciousness that exists independent of brain activity!)

It makes no sense. And with Libet's Readiness Potential taking increasing numbers of hits, his whole concept of "Free Won't" seems to make less sense every day.

But as I'm sure you know, some Neuroscientists have done tests of the whole "Free Won't" concept.

For example, Patrick Haggard--who, after years of trying to do experiments showing free will was an illusion---has tried with experiments to do the same thing with "Free Won't."

"There Is No Free Won’t: Antecedent Brain Activity Predicts Decisions to Inhibit"

Moreover, years ago, there were experiments by Marcel Brass and Simone Kuhn (and Haggard) that also looked at the whole Free Won't/Veto angle.

"Intentional inhibition: How the “veto-area” exerts control"

But "Free Won't," or having some kind of "Conscious Veto" that arises somewhere outside of brain activity/causation, seems like a ridiculous concept in itself.

Any thoughts?

Jeff, I will talk about this in an upcoming post summarizing why I think Libet's findings, and other findings in that vein, do not pose a threat to FW. I also hope to have another post making the same point about Wegner-style experiments. I just hope I can get to these two posts, given that my month is more than half over. More later, promise.

Thanks, Peter. I'll look forward to those threads!!

Looks like you're having another great month as the guest blogger! Very interesting stuff. : )

The comments to this entry are closed.

Books about Agency

3QD Prize 2014: Marcus Arvan