Does It Matter What Probabilistic Causation Is?

Here are two very different conceptions of what probabilistic causation is.

(1) It is the probability of causation. When e1 probabilistically causes e2, e1 stands to e2 in the same relation in which one event stands to another when the first deterministically causes the second. The difference between probabilitistic and deterministic causation is just a difference in whether there is (before the effect occurs) a chance that the causal relation will not obtain.

(2) It is the causation of probability. A probabilistic cause causes there to be a certain probability that a certain subsequent event will occur. Nothing else causal happens. It's then a matter of pure chance whether or not that subsequent event occurs. If it does, then, in a plain sense, nothing causes it.

In Libertarian Accounts of Free Will, I suggested that event-causal libertarians had better favor the first of these views, that the truth of the second would be fatal for their position. Indeed, it seems to me that if the second view is correct, and if intentional actions must be caused by psychological states, then events that are "probabilistically caused" cannot be intentional actions.

I haven't seen much discussion of this issue. I wonder if different conceptions of probabilistic causation are sometimes behind different responses to the problem of luck. What do folks think? (I mean you folks!)

Comments

[Eddy Nahmias]

Hi Randy, I agree with you that event-causal libertarians should prefer probabilistic causation (your 1) to causation plus randomness (your 2). I guess I always assumed everyone (in the free will debates) was talking about 1, because I assumed that the indeterminism suggested by the Copenhagen interpretation of quantum physics involves 1. Are there any causation theorists who talk as if 2 might be true, or whose theories require 2? Are there any free will theorists who suggest 2 might be the way to go? (Thanks for raising these issues.)

[Neil]

I'm not sure I understand what the options are. Is this right?

(1) the agent's mental states (or the event of the agent having certain mental states) has a (say) 0.7 probability of causing a volition or an intention.

(2) the agent's mental states (or the event of the agent having certain mental states) causes (with probability 1) the agent to have an intention/volition. There is then a 0.7 probability that the agent acts on that intention/volition.

I agree that picture 2 sounds worse for the libertarian. Perhaps it depends on where in the causal chain the non-causal link intervenes. If it is between intention and action, then it looks worse than if it is upstream: say with regard to whether certain reasons occur to the agent. Why? Perhaps because the agent's exercise of control needs to be closer to the locus of free will. But if you think that event-causal processes cannot involve an exercise of control, then it is hard to see what difference there could be between the pictures.

[jon kvanvig]

In response to Eddy, my sense of the literature is that the standard view is 2, not 1. Paul Humphreys is especially direct about this (he argues that the crucial idea in causation is difference-making, including making a positive difference to the probabilities under suitable control conditions), and the same idea guides Eels (1991). Salmon's mark transmission theory is a bit more difficult to classify, since it doesn't track probabilistic increase even under proper controls, as suggested first by Reichenbach (1956) (and refined by Suppes (1970)); but his view still seems to me to be a version of 2, just with a denial of the idea that probabilistic increase, under suitable controls, is necessary for probabilistic causation. In the literature on causation, I just don't see any clear sources embracing 1 over 2, and a pretty clear endorsement of 2 by many major players, and if I'm wrong about this assessment, it would be nice to hear about the sources.

[Fritz Warfield]

"It's then a matter of pure chance whether or not that subsequent event occurs. If it does, then, in a plain sense, nothing causes it."

What does "pure chance" mean so that the first claim in the quote is true? And what is the "plain sense" in the second sentence of the quote? Is it consistent with there being such plain sense in which nothing caused it that there are other clear senses in which something did? Presumably the claim that there is no (relevant) sense in which something caused it would be more important to the action issues than that there is some sense in which nothing caused it.

And reminders of where to look in the book if your best answers are recorded there are fine with me.

[jon kvanvig]

After reading Fritz's comment and looking more closely at the options, I think neither option accurately represents the standard view I was noting in the last post. The standard view I was noting denies 1, but doesn't replace it with 2. It claims that probabilistic causes are causes of events (or facts, if you prefer those), and it holds that the causing is probabilistic rather than deterministic. That's a different claim than the claim that probabilistic causes cause probabilities. Such causes, under the right controls, are commonly held to raise probabilities, and maybe there's an argument that such raising is itself causal. But the raising only happens when the controls are in place, and the causing happens even when they aren't. So what I take the standard view still seems to be neither 1 nor 2.

Not quite sure this is right, but I'm tempted to think that putting the question in terms of 1 and 2 assumes that causation is a deterministic notion, so that talk of probabilistic causation must be interpreted in terms of a scope issue on whether the probability governs the deterministic relation or whether the deterministic relation holds between a first item and a probability of a second item. I take the standard view in the causation literature to deny this presupposition: the probability is in the relation itself, and this relation holds between the same ontic units that deterministic causation holds between.

[R. Clarke]

Armstrong is someone who advances a view like (1). He says, "What a probabilistic causal law gives us is not probabilistic causality but a certain probability that causation will occur, an ordinary causation which occurs whether the law governing the causation is deterministic or merely probabilistic" (A World of States of Affairs, p. 238).

As Jon said, Humphreys holds view (2). Hausman has pretty much the same view.

Concerning a case in which two carcinogens contribute causally to a tumor, Humphreys says each carcinogen "contributes to the chance [of a tumor] on this occasion...and after they have done this, nothing else causal happens. It is...a matter of sheer chance whether the tumor occurs or not" (The Chances of Explanation, p. 37).

Hausman says "one can say that this is what probabilistic causation is: the deterministic causation of probabilistic states that then issue by chance into one outcome or another" (Causal Asymmetries, p. 201).

Neil, the disagreement concerns what a non-deterministic causal link is. I think the question you raised is where in the process leading to action such a link is said (by a libertarian) to be located.

[Benjamin Bagley]

"Indeed, it seems to me that if the second view is correct, and if intentional actions must be caused by psychological states, then events that are "probabilistically caused" cannot be intentional actions."

This strikes me as a fascinating and provocative claim! But I don't yet see why the inference is valid. Even if, as you say, there is a plain sense in which nothing causes a "probabilistically caused" intentional action, there is also a plain sense in which something does--the psychological state that made it plausible. (For, if such probabilistic causation were--as would presumably be the case in this scenario--widespread in ordinary life, this sense would just be the sense in which people ordinarily described (actually probabilistically related) events as causally related. That sense of "cause" would still, I think, be perfectly intelligible--and what sense could be plainer than that?)

If that sense would not be good enough to count putatively intentional actions as "caused by" psychological states, it's not clear to me why not. I'd think the answer would have to turn on substantive commitments of particular causal theories of actions, commitments that not all such accounts need share. (On the other hand, if a putatively intentional action were not "probabilized" by any psychological states at all, I would have no idea how it could count as intentional.)

(Again, if all this is developed in way more detail in your book, I'd be fine with just a reminder of where to look!)

[R. Clarke]

I don't see the issue as one concerning senses of 'cause', but as one concerning what causation is. Perhaps my original post is misleading on this point. Also, I don't myself endorse the Humphreys/Hausman view.

Perhaps a causal theorist can reasonably say that even if probabilistic causation is what these guys say it is, that's good enough for probabilistically caused intentional action. (As Benjamin suggests, it seems to be good enough for probabilistically caused things of other sorts.)

However, should a libertarian accept that on some occasion your prior psychological states contribute to the chances of each of two or more alternative decisions, and after they've done this, it's a matter of sheer chance which decision is made, and still, the one that is made is made freely?

[Franz JANSEN]

In response to Randolph, to my mind “ a probabilistic cause causes …” (your 2) is not a unique causation but a multi-factorial causation in the macrocosm, when the dominating causal factor at a certain time point is ignored. A dice shows “probabilistic causation” for an observer, but for a physicist all physical laws of the movements of a dice are known and could theoretically be predicted, so that only their high complexity does not allow any determinist prediction. Therefore “probabilistic causation” in general (your 1 also) becomes a relative definition, well adapted for the description of highly complex situations. Only from the viewpoint of the observer there is “probabilistic causation”, but from the viewpoint of a physicist there is deterministic causation, although too complex to be analyzed.

In the free will problem a person in a certain situation may be totally indecisive for finding the best alternative. Nevertheless, from the viewpoint of an outside observer her general tendency or “probabilistic causation” is well known. Finally the person may become decided by an internal psychological event. Although from the inside a person may know the precise determinist cause for her decision, the observer from the outside can only see her general tendency confirming or not the viewpoint of “probabilistic causation”, since the real, determinist cause is ignored from the outside. In the macrocosm, not considering the quantum mechanical problems, “probabilistic causation” can often be ignorance of underlying complex determinist causation.

[Fritz Warfield]

"However, should a libertarian accept that on some occasion your prior psychological states contribute to the chances of each of two or more alternative decisions, and after they've done this, it's a matter of sheer chance which decision is made, and still, the one that is made is made freely?"

Probably the libertarian should not accept this. But I don't know what "sheer chance" is yet, so I don't know if libertarians have to accept this and I don't know whether it's bad if we do.

[Bryan White]

I'll dip a toe into the shallow end of the grown-ups' pool, since there's a standing invitation for opinions from outsiders.

Fritz touches on something I see often in criticisms of free will, though certainly I'm more accustomed to less sophisticated arguments than those occurring here.

It seems to me that the libertarian ought to welcome the contribution of psychological states and "sheer chance" outcomes so long as the chance is located merely in the outcome. That is, there is no apparent ontological entity "Chance" that causes said outcomes. The outcomes simply occur as a result of allowing for the metaphysical feature of indeterminism and should be embraced as a feature of the libertarian view. So long as the other requirements of libertarian free will are met, "chance" in the aforementioned sense need cause no particular concern.

In sympathy with Fritz, I think "chance" is often used in the debate such that it seems somewhat ambiguous. We should certainly have an intuitive objection at the suggestion that our decisions are *caused* by chance. Otherwise, not so much.

[Christina Schneider]

The first to remember is that a "probability p " for e1 is void unless there is (at least) the complementary "probability 1-p" for non-e1. Only within probability-distributions are "probabilities" meaningful.
Concerning the question: The situation 1) can easily be interpreted that there is a probability distribution of the psychological states to be effective with respect to their intention (e2). Psychological states are effective only with probability.
The situation 2) introduces (somehow by psychological states) a probability distribution for the possible effects. And this is quite different. The "effects" would indeed be random. (Unlesss it was the intention to perform a random experiment with respect to the possible effects.)

I think the first situation is better for your purpose.

Best regards Christina

[Bryan White]

Christina,
Please pardon my barbarism.

Other than "this is quite different" I fail to see any clear difference. Situation 1 is summarized with "Psychological states are effective only with probability."

Situation 2 is described as "a probability distribution [somehow by psychological states] for the possible effects"

It seems to me the second might serve as a fair paraphrase of the first (no doubt because I'm missing some obvious subtlety?).

Suppose we opt for situation 2. Somehow, psychological states introduce the probability that A will result 89 percent of the time and ~A will result 11 percent of the time (thought experiment using identical scenarios). In what specific sense is this "random"? And why is this element of chance a threat to libertarian free will?

Can we not model either outcome as under the control of the chooser? And further, each reflecting a rational course of action given the deliberations in each case?

[R. Clarke]

When we model a choice as under the control of an agent, do we provide a theory of what it is for an agent to exercise such control? This is something that I look for in a theory of free will.

[Fritz Warfield]

What's the relation between a choice being free [for an agent] and a choice being "under the control of an agent"? Am I right that for people who favor "control" language those are the same thing or at least two-way entailment between them?
Or is that wrong (perhaps the relation is one way entailment...which way?)?

If they are the same thing, then we all had better look for it in a theory of free will. But if they are different, I'm pretty sure I'd rather look for a theory of freedom than a theory of control.

[Bryan White]

R. Clarke wrote:
"When we model a choice as under the control of an agent, do we provide a theory of what it is for an agent to exercise such control?"

I let the compatibilist define control, then I offer to show how a libertarian free will (indeterministic) model may add the ability to do otherwise (freedom).

Again agreeing with Fritz, I look for a theory of freedom. Then I use the notion of control to increase the appeal of that theory to a compatibilist.

[Michael Robinson]

Fritz,

Doesn't it seem that the relation between a choice's being free and a choice's being under the control of an agent depends, at least partly, on the sort of control in question? Take guidance control and regulative control, for example. Certain sorts of incompatibilists seem to hold that, whereas there is a two-way entailment relation between a choice's being free and a choice's being under an agent's regulative control (at least, when understood in a way that is incompatible with determinism), the entailment relation between a choice's being free and a choice's being under an agent's guidance control is only one-way: A choice's being free entails that it is under an agent's guidance control, but not vice versa.

[Christina Schneider]

Brian, may I give you an example.
A scientist intends to start a quantum experiment whose outcomes have a preassigned probability distribution.Scenario a) The scientist is mentally and physically in a perfect state (control) and starts the experiment. The outcome e2, say, will show up with probability p2.
Scenarion b) The scientist wants to perform the experiment, but unfortunately he is drunk and has therefore only a probability p1 less than 1 to succeed and 1-p1 not to saucceed. (reduced control). He runs the experiment and as a result the outcome e2 will obtain with probability p1 p2 (since he must manage to start the experiment correctly and if he succeds in that, e2 will show up with probability p2)

[Fritz]

Michael,
You a 3rd general possibility to the 2 I provided: perhaps there is no one relation (iff equivalence, one way entailment, or any other) between freedom and control because "control" is not a single notion to be related to freedom in any of those ways.

That may well be correct. If it is, isn't this simply more reason to look for a theory of freedom rather than a theory of "control"?

[Bryan White]

Christina,
Your examples help somewhat to clarify R. Clarke's intent (though it makes him appear to say that the drunken man's controlled action was not a responsible action). But it does not address my point about the supposed implications of chance. We can rule out responsibility with great ease for the drunken man where he does not control the result. Chance apparently has nothing to do with the loss of responsibility, since were he similarly not in control of his actions in a deterministic scenario we would not attribute to him any greater degree of responsibility (with the element of chance removed). How would the drunken man fail to be responsible where he has control of the outcome only probabilistically? I don't see chance as the problem, but rather the potential failure to offer an alternate possibility likewise under the control of the agent.

[Seth Shabo]

I'm still trying to get clear about Randy's original worry, which I take to be this:

Suppose we want to say that DB (an agent's belief-desire states at the time of action) probabilistically caused him to rob the poor box. And suppose that we understand the nature of probabilistic causation in accordance with (2) above. In that case, we might say that, given DB, there was (say) a .6 objective probability of his exhibiting poor-box-robbing behavior then. And--I take the worry to be--there is a problem about how this behavior can count as (constitute, whatever) the intentional act of robbing the poor box.

Is that the worry? If so, where's the rub exactly? Why couldn't this be a case of intentional action? Could we say that DB caused the behavior in question by causing the agent to form an intention to rob the poor box? Or is the worry that such a mental state wouldn't qualify as an intention?

[Michael Robinson]

Fritz,

When you say "look for a theory of freedom rather than a theory of control," I trust this means look for a theory of freedom rather than *only* a theory of control, or that the former should be the primary focus, and not that we should look for a theory of freedom to the exclusion of looking for a theory of control, correct? Since most accounts of freedom, presumably, will require that a free action be something over which an agent exercises some kind of control, it seems that providing an account of freedom is going to have to involve giving a theory of control of some sort.

[Fritz Warfield]

"Since most accounts of freedom, presumably, will require that a free action be something over which an agent exercises some kind of control, it seems that providing an account of freedom is going to have to involve giving a theory of control of some sort."

I think there are two groups of people talking about "control". Some use it in a completely unhelpful way -- as distracting alternative language for "free with respect to" (= "having control over..."
Others give a particular unpacking of a stipulative notion of "control" and then need to spend a whole lot of time clarifying what they take to be the relation between "control" in the specified sense and freedom.
Maybe there is some merit to some version of the latter approach. For example, if one manages to identify a notion of control that is very plausibly an impportant necessary condition on freedom, then it can figure in an account of freedom. But my sense of the literature is that this hasn't happened and that what has happened instead is people worry about control (in various senses) without any clean case for it standing in some interesting relation to freedom.

So yes, I think freedom should be the primary focus. And that "control" talk is typically nothing but a distraction.

A final way of putting the point: the sense in which I would require a free action to be one which an agent has control over is the trivial one. The kind of "control" required for free action is being free with respect to the action. I doubt, that is, that account of freedom will factor into a "control" condition + "the rest of what's needed for freedom".

[R. Clarke]

I don't think I used 'control' at all in my original post. Somewhere along the line, one of the comments included that word, and my response to the comment used it.

Here are a couple of ideas that seem correct (perhaps they need qualification). An intentional action is an exercise of agential control. Something over which an individual exercises no agential control isn't an intentional action performed by that agent.

On one conception of freedom, if a choice is free (freely made by an agent), then it is up to that agent whether that choice is made. (Fritz, I gather that you favor a conception of freedom something like this.) If an agent has no control over whether a certain event occurs, then it isn't up to that agent whether that event occurs; then it isn't a free choice.

I don't think that either of these ideas opens up the possibility of solving big questions about free will by shifting from talk about freedom to talk about control. I hope I haven't suggested that.

Seth, suppose that the thief actively forms the intention to rob the poor box--i.e., that his forming this intention is his making a choice to rob the poor box. Suppose that this choice is probabilistically caused by his having the desire and belief DB. Now suppose that this probabilistic causation comes just to this: the thief's having DB contributes to there being a .6 chance that a choice to rob will occur, and after that, nothing more causal happens. It's a matter of sheer chance that the choice to rob occurs. (Fritz asks: What exactly does that mean? I don't know; but it's what proponents of this view of probabilistic causation say.)

I retract my earlier suggestion that in this case, the "choice" can't really be an intentional action. Let's accept that it is. Still, can it be a free choice, one freely made; can it be up to the thief whether this event occurs?

That was my question.

[Seth Shabo]

OK, that helps. So the question is how the intentional action the thief performs qualifies as up to him, not how it qualifies as an intentional action.

As to what might be meant by saying that the outcome is a matter of "sheer chance," given the probability distribution established by DB, consider this.

Suppose that someone freely chooses to activate a random-outcome generator (where the randomness is objective and metaphysical), and freely assigns the probability distribution for each of the two possible outcomes at .6 and .4. Once the device is activated, the agent makes no contribution to which outcome ensues. That outcome is no more up to her than is a similar outcome that results from a naturally randomizing process.

If the thief's decision to rob the poor box given DB is relevantly similar to the outcome of a randomizing process, there is indeed a question about how it can be up to the thief whether he robs the poor box. At a minimum, if we understand the claim that the outcome comes down to "sheer chance" in the case of the randomizing devices, we should understand the analogous claim about the thief's action. It is simply a way of saying that thief makes no more contribution to which outcome ensues than he does in the case of a random-outcome generator.

(If there's something I'm missing about "sheer chance" and why this expression is problematic, I'd be interested to hear it!)

[Seth Shabo]

(By 'freely assigns the probability distribution,' I meant freely selects the probabilities, say, by turning a dial on a machine--in case that was unclear.)

[Franz JANSEN]

Seth, would you define libertarian indeterminism only as probabilistic, similar to the quantum mechanical approach, since you argued that an agent’s believe and desire “probabilistically caused him to robe the poor box“?

In a simple free will example of a mother offering her daughter a necklace and asking her to choose between 6 different alternatives, indeterminism does not seem to depend on probability, although it could become totally random.

Case 1: The daughter is from the beginning attracted by only one necklace and therefore decides with determinism.

Case 2: The daughter is in the beginning completely indeterminate, but during a deliberation process she eliminates with her motivational base one by one less interesting alternatives, in order to select the most attractive one. This is not probability, but motivation dependant selection. The daughter is initially indeterminate but finally determinate and thereby shows a kind of ephemeral indeterminism.

Case 3: The daughter remains indeterminate during the whole deliberation process and finally decides by pure luck for one of the alternatives. This is a kind of permanent indeterminism taking decisions with total randomness, but not with probability, and resembles the random choice of a lottery number.

If Case 1 (determinism) is a frequent feature, Case 2 (ephemeral indeterminism) is much more frequent. However, Case 3 (permanent indeterminism) is rare, but can be found in certain indecisive characters or when the slightest motivation is lacking, such as for the choice of a lottery number.

Ephemeral indeterminism is indeterminism limited to the initiation phase, which leads to determinism in the final phase with the help of a motivational base. In contrast permanent indeterminism, such as choosing a lottery number, looses any sense for free will expression, since the final choice entirely depends on luck and not on probability, so that personal motivations or psychological states can no longer interfere in the final decision.

[Bryan White]

Franz JANSEN wrote: "(P)ermanent indeterminism, such as choosing a lottery number, looses any sense for free will expression, since the final choice entirely depends on luck and not on probability, so that personal motivations or psychological states can no longer interfere in the final decision."

This may be a candidate for the type of ambiguous invocation of chance I wrote about earlier.

On what basis is it determined that the choice depends purely on luck if that is not a metaphysical stipulation (that is, it is assumed to be true)?

I suspect that *if* we were able to have a subject choose a lottery number repeatedly under identical conditions we might well see some numbers chosen with a higher probability than others. Perhaps some numbers are preferred analogous to the preference for the necklace.

That's the thing with randomness. Sure, we can make it a metaphysical assumption if we like. But that begs the question as to whether indeterminism is necessarily random, even for choosing lottery numbers.

It may be the case, as Franz suggests, that some decisions are deterministic within a certain time frame even under an overarching framework of indeterminism. And it may likewise be the case that some decisions are truly random within that same framework. But I am skeptical that the human will--the thing that moves us--is entirely separable from "motivations," including psychological states.

I am reminded of Bob Doyle's "Cogito Model." Doyle would explain indeterministic features of the will via a truly random "prior" cause--quantum phenomena (I place "prior" within quotation marks because Doyle locates such causes within the deliberation process rather than prior to deliberation). And Doyle appears to fend off the charge that the will is random as a result by stipulating that the will can selectively ignore some of those random "prior" causes.

I suppose I should try to have a point. :)

Franz's case 2 scenario is, I think, capable of enough flexibility to closely resemble his case 1 and case 3 scenarios in terms of function. It ends up not being a question of whether libertarian freedom is manifested, but whether the cases involve the need to invoke moral responsibility. The choice to play the lottery might have a notable moral dimension. Much less so the choice of numbers.

[Fritz Warfield]

Hi Randy,
Sorry about the control diversion... --- several conversations going on at once.
About your original question; in the absence of an argument for a special problem for thinking there is free agency in this case, I don't immediately see a problem. I don't see any all things considered problem thinking the description is consistent with intentional action (though I do see why some initially wonder about this). And given this, I don't see anything that would apply that though the action may well be intetional it's not free.