by Storrs McCall

A common criticism of free will or origination theories is that if what we do is not the result of an unbroken sequence of causes and effects, then it must to some degree be the product of chance.  But in what sense can a chance act be intentional or deliberate, in what sense can it be based on reasons, and in what sense can a person be held responsible for it?  If free and responsible action is incompatible with determinism, must it not equally well be incompatible with indeterminism? Professor McCall says no.  He argues that a new idea, that of a controlled indeterministic process, resolves a variety of classical dilemmas and opens the way to a new understanding of the relationship between actions, reasons, causes, and responsibility. Does he succeed? All of this, like a related line of argument by Professor McCann to which you can turn, is a long way from what seems to me the continuing arguableness of determinism and the unavoidableness of the proposition that both incompatibilism and compatibilism about freedom are false. But we all need to remember, with Cromwell, in our own bowels if not by those of Christ, that we may be mistaken. I guess that given the proportion of false to true views in the world, we need to remember it is arguable that we are more likely to be mistaken.


What is a controlled indeterministic process, a “CI-process” for short?  I start off by giving examples, and try to explain in what sense they are indeterministic and in what sense they are controlled.  A good first example is walking down a crowded sidewalk.  This is indeterministic because other walkers come at you unpredictably from different directions, yet it is not like Brownian motion because rarely if ever do you bump into someone.  Both you and the other walkers are highly self-controlled.  Other examples are skiing, singing, throwing darts, putting on a putting green, driving a car, and playing the piano.  Skiing is indeterministic because irregularities in the snow prevent any turn from being completely smooth.  In addition, there are marked differences between skiing in control and skiing out of control.  The same goes for driving a car.  Even for Tiger Woods, the holing of short putts is anything but a science, and throwing darts can become progressively less controlled as the evening progresses.  A 100% deterministic piano-playing machine, not pre-programmed but reading the music and depressing the keys mechanically, might be able to give a fair rendition of Bach.  But could it deal with the expressiveness demanded by Brahms?  Every full-blooded performance of the notorious third Rachmaninov concerto contains unpredictable surprises, while at the same time exhibiting superb control.

An example of a long-lived CI-process would be the composition of a novel.  When Tolstoy started War and Peace, was it already established when Pierre and Natasha would marry?  Given the initial conditions prevailing at the start of any creative composition, including the writing of this paper on CI-processes, there are always alternative themes, emphases and conclusions which remain possible but unactualized until the last word is written.  A compositional process with alternative distinct possible end-points is necessarily indeterministic.  Yet a good writer always controls her material, while operating within the framework dictated by the logic of her work.  The same goes for a Rembrandt or a Picasso.

Here are some more CI-processes: throwing a baseball, holding one’s breath (indeterministic as to the exact length of time until the next breath), tying a bowtie, doing a handstand, swimming, making an omelette, adding up a column of figures1, taking the square root of 236073, whistling, apologizing. It seems plain that those who execute a CI-process exhibit a skill or technique, whether learned or (in the case of a bird building a nest) innate.  But no manifestation of a skill is 100% perfect.  Each performance contains idiosyncracies reflecting the indeterminism and the different varieties of control found in all CI-processes, characteristics which set them apart from non-CI-processes such as the algorithmic operation of a Turing machine or of a modern computer.


The CI-processes discussed so far, such as walking, whistling or sinking a putt, can all be classified as actions.  Even writing a novel can be looked on as an extended multi-year action.  But there is another important sub-class of CI-processes which are not themselves actions, but which result in actions, namely deliberative processes which culminate in decisions.  Examination of this subclass yields the linkages between reasons, causes and responsibility that action theory seeks.

When an agent deliberates, the deliberative process she engages in can be divided into three distinct stages:

1. Listing the alternative options
2. Evaluating them
3. Choosing which option to realize.

For example, Ann this evening is faced with the choice of (A) eating out, (B) going to a movie, (C) working on a new edition of prints.  Each one of these possibilities is do-able, each excludes the others, and each of (A), (B) and (C) has reasons for, and reasons against, choosing it.  In stage 2 of the deliberative process Ann weighs her reasons for, and her reasons against, each option.  This process is not like weighing sugar.  The reasons do not come with ready-made objective weights, but must be assigned a weight, i.e. weighted by the deliberator.  The assigning of weights is controlled by Ann, and may change in the course of the deliberation as new information and new considerations come to mind.  E.g. there are conflicting reports about the movie; Hugo liked it but Tita did not.   Ann must first weight and then weigh the pros and cons of (A), (B) and (C) against one another.

Stage 2 of Ann’s deliberation ends with a list of the three options in order of preference: for example first (B), second (C), third (A).  The final stage 3, consisting of the decision to implement (B), may follow immediately, or may be delayed until the last moment for arriving at the cinema on time.  In any case Ann acts, when she acts, for a reason, and the degree of influence on the final outcome exercised by each reason is determined by the overall control she has over her decisional process.

Since the weights attached to inputs to any deliberative process are controlled by the deliberator, two CI-processes with identical initial conditions and identical inputs may result in different decisions, and consequently different actions.  This fact constitutes the indeterministic nature of the processes.  Their indeterminism does not lie in a single undetermined event, such as the chance decay of a free neutron or other unstable particle in quantum mechanics, or even in a series of such events.   Instead, deliberation is an indeterministic process, not an event.  Its indeterminism is continuous rather than discrete, and derives from the continuous nature of the control exercised by the deliberator in weighting the various options, and the importance he or she attaches to other inputs during the course of the deliberation.   In a deliberative CI-process, reasons are causes, but they are not sufficient causes.  They are causes that in Leibniz’ phrase “incline without necessitating”.  That they do not necessitate is due to the overall control exerted by the deliberator.

Because reasons incline but do not necessitate, it is possible to give an “event-causal” rather than an “agent-causal” account of deliberation and action, where the causes are reasons and the things of which they are causes are actions.2  Since reasons are not sufficient causes, the variety of causation appropriate to deliberation and decision is probabilistic causation.  One could view the causality of reasons in deliberation as a probability function, in which the probability of each option varies continuously and the probabilities p(A), p(B), p(C), … sum to one.  When the moment of decision arrives, one of these probabilities takes the value 1 and the others become 0.  Throughout deliberation the different probability values vary smoothly, in a reciprocal way.

As was stated in section 1 above, viewing deliberation as a controlled indeterministic process allows the concept of reasons-based action to occupy an intermediate position between chance and determinism.  A reasons-based action is not the product of a 100% deterministic process, nor is it the outcome of a single random event or series of random events.  Instead, it is the end-product of a process that is continuous, controlled, and indeterministic.  The indeterminism does not derive from the process’s containing isolated chance events, but from its having different physically possible terminations. Reasons-based actions consequently pass between the Scylla of determinism and the Charybdis of chance.


“Control” is a slippery word, and we need to be precise about its different meanings.3   Take the example of steering a car.  We would say that the driver controls the car’s steering if a necessary and sufficient condition of its turning to the right is that the driver turns the wheel to the right, and a necessary and sufficient condition of the car’s turning to the left is that the driver turns the wheel to the left.  Having said this, it is plain there exist lesser degrees of control.  Suppose that the steering mechanism is defective in such a way that when the wheel is turned right, the car goes right, but there may be instances in which the car turns right without the wheel being turned right.  That is, turning the wheel right is only a sufficient condition of the car’s turning right, not a necessary condition.  I shall say that in this case the driver exercises “performance control” over turning right.  Alternatively, turning the wheel right may be only a necessary condition of the car’s turning right, not a sufficient one.  I.e. sometimes the car fails to turn right when the wheel turns right, although it never turns right unless the wheel turns right.  In this case I say that the driver exercises “omission control” over turning right, so-called because omitting to turn the wheel prevents the car’s turning.  “Full control” requires both performance and omission control.

Affinities exist between performance control and what Fischer calls “guidance control”, and between full control and what he calls “regulative control”.  But much still remains to be said about these different levels of control.  In the field of human psychopathology, one would not like to be in the position of exercising only performance control, but not omission control, over the movement of one’s limbs.  However, since thoughts frequently enter one’s head unbidden, one may be precisely in that position concerning the operations of the mind.  We can think what we want to think, but sometimes we think what we don’t want to think.  Despite all this, the variety of control we possess over deliberation, including the weighting of reasons and the power of whether, when, and what to decide, is “full” control, i.e. the combination of both performance and omission control.


Under what conditions can someone who does A be said to be responsible for doing A?  This is a huge subject, discussed in dozens of papers in recent years, many of which owe their inspiration to Harry Frankfurt’s seminal 1969 paper “Alternate possibilities and moral responsibility”.  Frankfurt’s “Principle of alternate possibilities” (PAP) states that a person is morally responsible for what he has done only if he could have done otherwise.  Frankfurt argues persuasively that PAP is false, and that someone may well be responsible for what he has done even though he could not have done otherwise.  From here it is only a short step to the conclusion that moral responsibility is compatible with physical determinism.

Frankfurt’s counter-example to PAP is well-known.  Black wants Jones to perform a certain action A that Jones is contemplating performing in any case, and arranges things so that if Jones is tending towards a decision not to do A, Black will create in him an irresistible inner compulsion to do A all the same.  (E.g. Black may hypnotize Jones, or may manipulate his neural processes using electrodes implanted in his brain.)  Jones is unaware of these arrangements, and in the end decides to do A without Black’s fail-safe device coming into play.  In such a situation, Jones is responsible for doing A, even though he could not have done otherwise.  Consequently PAP is false.

Many philosophers have disagreed with Frankfurt’s reasoning, some arguing that at the time Jones decided to do A he could have decided otherwise, and that our judgement concerning his responsibility is based on this initial ability to do, or not to do, A.  Only later, when Black’s device came into play, would Jones have had no choice.4  In support of Frankfurt, others have proposed new and ingenious “Frankfurt-style examples”, in some of which Jones’ decision to do A and Black’s fail-safe device are stipulated (perhaps unrealistically) to become operative at exactly the same instant.5  At that point it becomes difficult to say which prevails; whether Jones’ decision pre-empts Black’s device, or whether the device pre-empts Jones’ decision.  My purpose here is not to pursue this point, but rather to suggest a different strategy for libertarians, a strategy that is immune to Frankfurt-style counter-examples.

Frankfurt’s indeterministic principle PAP states a necessary condition for moral responsibility.  A person is morally responsible for what he has done, only if he could not have done otherwise.  But if we add to indeterminism the notion of control, then it becomes possible to give not only a necessary condition for responsibility, but a necessary and sufficient condition.  This strengthened condition resists counter-exampling in the style of Frankfurt.  Or so I shall argue.


To replace PAP I suggest the following “Principle of Responsibility and Control” (PRC).  The principle concerns responsibility in general, not the more restricted concept of moral responsibility.

PRC.  A person is responsible for having done A at time t if and only if, throughout some time interval immediately before t, the doing of A, including whether or not to do A, was in her control.

Although it is not stated explicitly, it will be clear that PRC applies only to actions that originate within an indeterministic context.  I.e. it applies to actions that result from a CI-process, where the controller of the process is the agent.  If the question of whether or not to do A falls within the agent’s control, then ipso facto the deliberative process has at least two different possible outcomes, doing A and not doing A, and is therefore indeterministic.

Is PRC open to Frankfurt-style counterexamples?  It would seem not.  A counter-example would involve a controlled action of an agent x for which x is not responsible, and it seems implausible that there are any such.  Alternatively, it would involve a performed action, not within x’s control, for which x is nonetheless responsible, and this seems equally implausible.  However, it’s not out of the question that some Frankfurt of the future may one day uncover an ingenious counter-example to PRC, and I leave the possibility open as a challenge.  The remainder of this section is devoted to examining cases which are not counter-examples to PRC, but which support the principle.

    Imagine the following case, adapted from J.L. Austin (1956).  I go to shoot my donkey, but just as I pull the trigger you jog my arm, with the result that I shoot your donkey at time t.  Am I responsible for shooting it?  No, because during a short time interval before t the shooting of it was not under my control.  It was an accident.  This case differs from another Austin example in which I go to shoot my donkey, but because of the similarity of the animals I shoot yours instead, not by accident but by mistake.  In this case, I am responsible for shooting and killing it, but as I argue in the next few paragraphs I may or may not be morally responsible for its death.

    As was mentioned at the beginning of this section, PRC provides necessary and sufficient conditions not for moral responsibility, but for responsibility simpliciter.  If we are dealing with the concept of responsibility this is surely the way to go: first explicate responsibility, then deal with morality.  The trouble is, that in some cases the two categories of “responsibility for doing A”, and “moral responsibility for doing A”, are so intertwined that it is difficult if not impossible to separate them.  Let us try however.

    Consider Elizabeth Anscombe’s example of moving the pump handle up and down, and in so doing poisoning a house’s water supply (Anscombe (1963), p. 37).  I may be responsible for performing action A, but am I responsible for performing A under every description of A?  My (Davidsonian) answer is yes.  If the action of moving the pump handle up and down is the same action as poisoning the house’s water supply, then I am responsible for poisoning the water supply.  This holds even though I may be unaware that I am performing that reprehensible deed, i.e. unaware that the reprehensible description applies to it.  On the other hand, I may not be morally responsible for poisoning the water supply.  Examples like this demonstrate the importance of distinguishing between responsibility and moral responsibility.

    Another more difficult example, along similar lines, is the following, from Hart and Honore (1959), p.72.  If x puts poison in y’s soup and y drinks it and dies, then x is responsible for y’s death.  But suppose y knows that the poison is is the soup and drinks it anyway?  Does y’s free, intervening and knowing action break the causal chain that links x’s action to y’s death?  I confess I don’t know in this case what to say.  Maybe x is responsible for y’s death, but not morally responsible.  Or maybe x is not even responsible simpliciter, let alone morally responsible.  “To consequences no limit can be set”.  The causal consequences of our actions are endless, and causal chains can extend for hundred of years into the future.  Are we responsible for all the consequences of what we do?  Surely not.  Yet where should we draw the line?  I don’t know.  I do however believe that the concept of responsibility simpliciter should be kept separate from that of moral responsibility, and legal responsibility, and “capacity responsibility”, and from other notions such as Hart’s concept of “role responsibility”.  Let me conclude with the following passage from Hart’s Punishment and Responsibility (1968), p. 211, illustrating the diversity of responsibility-concepts.  I would hope that PRC might serve to extract one single thread of what may be called “action responsibility” from the complex tangle found in Hart’s passage.  But I am not optimistic.  There are still too many additional categories of responsibility, e.g. responsibility for omissions, negligence, etc., to be confident that our principle PRC covers them all.

    “As captain of the ship, X was responsible for the safety of his passengers and crew.  But on his last voyage he got drunk every night and was responsible for the loss of the ship with all aboard.  It was rumoured that he was insane, but the doctors considered that he was responsible for his actions.  Throughout the voyage he behaved quite irresponsibly, and various incidents in his career showed that he was not a responsible person.  He always maintained that the exceptional winter storms were responsible for the loss of the ship, but in the legal proceedings brought against him he was found criminally responsible for his negligent conduct, and in separate civil proceedings he was held legally responsible for the loss of life and property.  He is still alive and he is morally responsible for the deaths of many women and children.”


A further topic needs to be addressed: the question of rationality.  The process of deliberating and deciding can be done either rationally or irrationally, and consequently there are such things as rational controlled indeterministic processes, or “RCI-processes”.  What distinguishes a rational from an irrational deliberative process, a rational from an irrational practical decision?  At this early stage in our understanding of CI-processes, the question is best addressed by considering concrete examples. 

Chess-playing is a quintessentially rational process, which is intellectually demanding, highly controlled, and indeterministic because the opponent’s moves are unpredictable.  (A “deterministic” chess game, e.g. against a computer, in which the computer’s moves were calculable and predictable by the opponent, would not be a game at all.)  Chess-playing is an exercise in rational decision-making under uncertainty: an RCI-process.

A very different case of rational decision-making is that of Mele’s Beth (Mele (2006), pp. 106-7).  Beth is a lawyer in a large competitive urban law firm who deliberates about quitting her job and working in a small town where her moral principles would be less likely to be compromised.  She visits the small-town firm, is impressed by the enhanced quality of life of the new environment, and is just about to accept a position when it occurs to her it would be prudent to return once more and explore the town on her own.  On her second visit she encounters a degree of conservatism and bigotry she was previously unaware of, and she promptly refuses the offer and remains in the city.  Beth’s decision-making process is a rational one, based on practical rather than intellectual reason, and features a radical change in her weighting of the pros and cons of moving vs. staying after her second visit.  The change in weighting is brought by Beth, and exemplifies the rational control she exercises over her deliberation.   The latter stands as a typical case of a rational, controlled, indeterministic process.


A final example will show, I shall argue, not only that some rational controlled processes are indeterministic, but more generally that overall Laplacian determinism is false.  The RCI-processes I have in mind are those engaged in when a mathematician sets out to prove or disprove a mathematical proposition, the truth-value of which is then unknown.  Consider Fermat's Last Theorem (FLT), proved by Andrew Wiles in 1996 after 20 years' work (see for example Singh (1997)).  FLT states that the equation 

            xn + yn = zn

has no whole-number solutions for n>2.  Before 1996 there were two possibilities: either (i) Wiles would succeed in proving FLT, or (ii) he would not, and FLT would remain unproven.  (Since, as we now know, FLT is true, there existed no third possibility (iii) that Wiles would disprove FLT.)

    If Laplacian determinism were true, a "Laplacian demon" who knew with 100% accuracy the state of the physical world in, say, one million BC, and knew the laws of nature, would seemingly have been able to predict that Wiles would produce a proof of FLT in 1996.  Not only would the demon have been able to predict the shape of the ink-marks on paper when Wiles' lengthy proof appeared, he would also have predicted the headlines in the newspapers and the general acclaim that greeted Wiles' achievement.  But accepting the possibility that the demon could have predicted Wiles' proof would entail accepting that the truth of FLT, a mathematically necessary proposition, were derivable from the physical state of the world's atoms and electrons in one million BC.  That is to say, information about a mathematical truth could be derived from information about the physical world.

    But accepting this requires that we reject the watertight barrier that most philosophers believe separates the realm of mathematical necessity from the realm of physical fact and causal necessity.  The laws of mathematics and the laws of nature are mutually independent; one cannot be derived from the other.  If a Laplacian demon were asked to say whether FLT was true or false, he could justifiably reply that this lay outside his job description.  What is causally necessary is distinct from what is mathematically necessary, and information about whether, for example, every even number is the sum of two primes (Goldbach's conjecture) cannot be extracted from the state of the particles and fields constituting the universe at any moment up to now.  We must conclude, therefore, that the demon would have been unable to predict Wiles' proof of 1996.  It follows that the outcomes of some rational deliberative processes are unpredictable in principle, that universal Laplacian determinism must be false, and consequently that the idea of controlled indeterministic processes can be studied and developed within an overall probabilistic world-view such as that of quantum mechanics.


What is new about this paper is the concept of a controlled but indeterministic process, exemplified by activities like walking, running, writing a poem, playing tennis, etc., but more importantly for action theory by the process of deliberating and deciding.  When we decide and act for a reason, that reason, in competition with other reasons on which we do not act, serves as a probabilistic cause of the eventual action.  If our action is a rational one, the various competing probabilistic causes are controlled by Reason with a capital R.  If our action is less than fully rational (as thankfully many actions are), it will in most cases still be controlled by the agent, though the latter may be influenced by desire, emotion, laziness, passion, or fear.  Sometimes a Iago can manipulate an Othello, playing upon his weaknesses and sharing the responsibility for whatever crimes Othello commits.  In such cases, Othello’s control over his own behaviour is diminished, and in accordance with the principle PRC, responsibility should be reduced proportionately.   CI-processes provide a framework within which an adequate libertarian account of autonomous, reasons-based, responsible action can be formulated.


1 “I add them up from top to bottom, then add them up from bottom to top and split the difference”. (Stephen Neill)

2 See Clarke (2003) and O’Connor (2000).  Mele (2006) p. 10, remarks that “Typical event-causal libertarianism encompasses a commitment to what may be termed agent-internal indeterminism.”  But Mele himself stops short of embracing a full-blown indeterministic event-causal account of action based on reasons as probabilistic causes.

3 See Fischer (1994), pp. 132-34, and Fischer and Ravizza (1998), pp. 30-32.  The account given here differs significantly from Fischer’s.  In Aristotle’s Ethics 1102b14, self-control is “the disposition to refuse to act on desires not approved by reason; the lack of it ... is the disposition to give way to them".  (Broadie and Rowe (2002), p. 295.)  See also Mele (1995), p. 4 ff., and his (1987).

4 See Kane (1996), pp. 142-3.  Note that Frankfurt’s whole example is presented within an indeterministic framework, neither Black, nor Jones before the decision, being causally necessitated to do what he does.  Can an indeterministic example be used to argue against indeterminism, and to support determinism?   Surely not.

5 See Mele and Robb (2003); Ginet forthcoming; McKenna forthcoming; Widerker (2006).


Anscombe, G.E.M. (1963) Intention. 

Austin, J.L. (1956) “A plea for excuses”, reprinted in his Philosophical     Papers (1961).
Broadie, Sarah and Christopher Rowe (2002), Aristotle: Nicomachean     Ethics.

Clarke, Randolph (2003) Libertarian Accounts of Free Will.

Fischer, J.M. (1994) The Metaphysics of Free Will.

Fischer, J.M. and M. Ravizza (1998), Responsibility and Control.

Frankfurt, Harry (1969) "Alternate possibilities and moral responsibility",     The Journal of Philosophy 66, pp. 829-39.

Ginet, Carl “On Mele and Robb’s indeterministic Frankfurt-style     examples”, forthcoming in Philosophy and Phenomenological     Research.
Hart, H.L.A. and A.M. Honore (1959), Causation in the Law.

Hart, H.L.A. (1968) Punishment and Responsibility.

Kane, Robert (1996) The Significance of Free Will.

McKenna, Michael “Frankfurt’s argument against alternative     possibilities: Looking beyond the examples”, forthcoming in     NOUS.

Mele, Alfred R. (1987) Irrationality.

Mele, Alfred R. (1995) Autonomous Agents.

Mele, Alfred R. and D. Robb (2003), "Bbs, magnets and seesaws: The     metaphysics of Frankfurt-style cases", in Widerker and McKenna.

Mele, Alfred R. (2006) Free Will and Luck.

O’Connor, Timothy (2000) Persons and Causes. 

Singh, Simon (1997) Fermat's Last Theorem.

Widerker, David (2006) "Libertarianism and the philosophical     significance of Frankfurt scenarios", The Journal of Philosophy     103, pp. 163-87.

Widerker, David and Michael McKenna (eds) (2003), Moral     Responsibility and Alternative Possibilities.

HOME to TH website front page
HOME to Det & Free website front page