-- Ambrose Bierce, The Devil's Dictionary
"To be or not to be", brooded the Prince of Denmark. Such agonizing choices are thankfully rare, but decisions, trivial and momentous, fill our days. Was something there? What was that? What to do? Whether to do it? When to do it? Was it the right choice?
In the last few years, a number of investigators have begun to claim that they are studying decision processes (e.g. Parker & Newsome, 1998; Schall & Thompson, 1999; Shadlen, 1999). In these writings, the word decision is used in an innocent, colloquial sense. According to this view, decision making, the most deliberate of mental operations, seems to have been probed by the neurophysiologists electrode. We will learn, though, that most of the neurophysiological data has more to do with choice than with decision. This book will summarize and review the recent discoveries made by monitoring the activity of neurons in the brains of monkeys performing a range of complicated, demanding tasks. The exposition will also draw on theories and findings from other research domains as needed.
What kinds of choices have been studied? Many studies have investigated how the brain resolves ambiguity perceived in the environment - that is one kind of decision. Other studies have investigated how the brain controls action, what to do and when to do it -- this is another kind of decision.
Neural Correlates of Choosing
Considered in the most general terms the process of deciding takes time. During this time the brain (or some subset of circuits) undergoes a transition from multipotent to determined, from plurality to unity, from possible to actual. If the proper variables were measured in the brain, there seems no reason why one could not observe this transition. Imagine a state variable that describes the probability of a choice among alternatives. If there were 2 choices, say, and everything else were equal, this variable would start around 50% and then over time with the accrual of information the variable would approach 100% probability of one of the choices (see Figure). In the chapters to follow, neural correlates of this probability variables like this and their transition will be described. Often the neural representation of a choice process consists of a comparison of activation between two or more populations of neurons that encode the alternatives. The choice variable is derived from some comparison of the balance of the current activation in the different populations of neurons (see Figure 10). A challenge we will face is trying to relate the measurement of the evolving different states of activity across populations of neurons to the actual probability of choosing one of the alternatives. We will see that the neural correlates of a choice can be achieved without necessarily resulting in a movement that expresses the choice. An outcome of this analysis is that we should be careful to distinguish the process by which a choice is arrived at from the state of having chosen but not yet executed.
The Need for Definitions
Science travels on its vocabulary. If words are not used consistently and effectively, then progress is impeded. The recent literature in behavioral neurophysiology is full of words like 'decision', 'intention' and the like. Authors of papers in neuroscience journals routinely write of monkeys "expecting," "attending," "intending," and so on. They make attributions that would make a cognitive psychologist cringe. What do these words mean? When a neurophysiologist says them, does he mean the same thing as a philosopher, a lawyer, or the man in the street means? If they are all referring to the same thing, then the neurophysiologist's work may be relevant for the philosopher, the lawyer and the man in the street. But if the neurophysiologist means something different, then his findings may not be meaningful to philosophical arguments, legal cases or daily life. We wish for this not to be true. Therefore, we will invest some effort in considering and defining the terms like "choice," "decision," "action," "intention," and so on. The goal is to replace habitual, unprincipled use of terms with more disciplined usage. Beyond the issue of relevance outside the laboratory, this exercise is important. Can these words be defined in a way that provides service for conceiving more discriminating hypotheses and designing more informative experiments? The goal of the rest of this chapter is to review the philosophical background of these basic words. Readers interested in understanding the neural correlates of decisions may regard this philosophical treatment as dry as bones. But if the skeleton is weak or misformed, on what can the flesh of neurophysiology hang and move?
This enterprise in cognitive neuroscience will fail without serviceable definitions of these critical words. Neuroscientists understand the valuable but grueling effort of obtaining a detailed knowledge of neuroanatomy and neurophysiology. In the same light, we should not be loath to undertake the grueling work of philosophy. If we care to have a precise and accurate understanding of things like action potentials and areas of the cerebral cortex, and if we wish to consider how things like action potentials and areas of the cerebral cortex relate to things like choices and decisions, then we ought to care to have an equally precise and accurate understanding of things like choices and decisions.
Another reason for the importance of defining these terms so carefully is the fact that although we wish to understand human thought and behavior, most of our information about the signals generated by the brain is obtained in nonhuman primates. Do terms like choice, decision and intention apply to nonhuman primates? If not, then can the monkey be a model of man? If so, then are there any particular limitations of the application of these terms to monkeys?
Deciding, Choosing and Acting
The development of this material is derived from Oldenquist (1967), Dennett (1984), Ryle, and others. An adequate understanding of the terms 'choosing', 'deciding' and 'doing' is important for gaining insight into the relation of mind and brain. The concepts of choose, decide and intend are historically and typically applied to private mental processes; that is, they are subjective. But if mental events actually arise from physical-chemical biological processes, according to the modern neuroscientific view, then are 'choice' and 'decision' fundamentally mental events? We must arrive at a view that can show how the subjective 'choice' of an agent can be explicated in terms of objective (that is publicly observable) events such as those occurring in the brain.
We cannot proceed without dealing with the concepts of 'intention', 'action' and 'consequences'. How should we distinguish between an action and that which precedes it, and that which follows it? In other words, where does an action begin, and where does it end? Are intentions, decisions, and choices just muscle contractions caused by nerve impulses? Or are choices not really actions, but instead they are preliminaries to actions which nevertheless stand in certain important relations to actions? If intentions, decisions, and choices are actions, then is whatever follows them not an action but instead a consequence? How does one specify where and when an action ends and its consequences begin?
The neural basis of choosing and deciding is central to the relation of freedom and determinism. Can decisions and choices be caused? Can decisions and choices cause actions? If choices are caused, i.e., determined, then in what sense would they be regarded as free? Are actions identical with sets of events that in principle can be predicted on the basis of antecedent conditions and relevant causal laws? On the other hand, if an action is more than a set of physical events, does this preclude causal determination of actions?
Space and competence prevent reconsidering all of the arguments surrounding these profound questions, but a thesis of this book is that new information about the brain processes related to choosing and acting can inform the arguments, and we should not shy away from these implications that invigorate the inquiry.
The Difference Between Choosing and Deciding
We might paraphrase William James and say that everyone knows what choosing and deciding is. Certainly we have a lot of experience choosing and deciding. However, as we begin to investigate the mechanisms of choosing and deciding, we shall find that intuition born of introspection is not always reliable. Also, we will discover that distinguishing softly between choosing and deciding is useful. We can contrast several alternative views about the nature of choice. First, choices may be mental acts that precede and perhaps cause actions. At this stage, let us consider a mental act in its most innocent sense to mean some conscious thought process. Second, choices may not be mental acts but instead are overt actions done in a context of alternatives. Or third choices, when distinct from decisions, are overt actions done in a context of alternatives, and when they are not overt actions, they are identical with decisions. Or fourth choices may or may not involve mental acts and fundamentally are overt actions for which certain sorts of explanations in terms of purposes can be given. We will find that the last three statements are refinements to deal with problems engendered by the first alternative.
We can establish certain properties that define and distinguish choices and decisions. One characteristic of choices is that they are not true or false. They can be good or bad, wise or foolish, defined in relation to the attainment of some goal. In other words, it would make no sense to say that my choice of the blue plate lunch special is true or false, but it would make sense to say that my choice was satisfying. Let us be clear; the assertion that choices are not true or false is not the same as the assertion that one could alarm truly or falsely that a certain choice was made. Likewise, we should distinguish two distinct senses of the word 'decide'. We can decide that something is the case. This kind of decision can be true or false.
This kind of decision will not be considered further because it cannot be investigated in nonhuman primates very effectively. We will see that it is tricky enough to attribute a perceptual or intentional state to a monkey. I would not care to try to attribute something like a misunderstood concept or a false proposition to a monkey. Deciding that enters the domain of decision theory (Eells E, 1982; Rational Decision and Causality, Cambridge: Cambridge University Press). This is certainly relevant, but we must understand how the brain decides to before we can understand how the brain decides that.
Alternatively, we can decide to do something. These kind of decisions are not true or false but can be good or bad like choices. This book is about how the brain "decides to". What is learned may be relevant for our understanding how the brain "decides that".
How strict is the distinction we should draw between choosing and deciding? In some cases choosing may be synonymous with deciding specifically when a choice among alternatives is made but the expression of that choice is delayed. However, situations in most choosing is more closely related to action than is deciding, and deciding is more closely related to intending than is choosing. For example, it is perfectly intelligible to claim that the decision that this book will win a Pulitzer prize is false. The definition of intention will be considered later in this chapter. An example will help make this point. The members of my laboratory occasionally eat lunch at a particular restaurant. Each individual orders a particular favorite meal; we are creatures of habit. The choice of meals is made automatically, with no deliberation, no look at the menu to examine the alternatives. Looking for any excuse to eat out, we take new lab members or visitors to this restaurant. In contrast to the habitual, predictable choice that we make, the new person must examine the menu, evaluate the options before making the choice. Whereas, the old members of the lab chose automatically, the visitor engaged in a conscious decision process before choosing. The choice of the old lab members could be predicted with reasonably high accuracy. But the decision and choice of the visitor could not be predicted. Everyone had to choose (otherwise no food would appear), but only the visitor had to decide. This is the essence of the distinction between choice and decision we need to make.
Even though choices per se are not true or false, statements about choices are true or false. "Jones chose to attend class" can be verified as true or false. This kind of verification is possible also for first-person past-tense statements like "I chose to write this book." However, some first-person future-tense statements of choice and all first-person future-tense statements of decision appear to be unintelligible. "I will choose (or will decide) to buy this book next week." The statement that one has not yet chosen or decided means that one has not made up one's mind. A defining feature of decision is that they cannot be predicted --even by the agent. One does not know what one will choose in the future. Consider this statement by Michael Jordan :
"I feel that way right now. Ask me in two or three months and I may change. I don't think I will. I'm pretty sure that's my decision."
-- Michael Jordan on his retirement from professional basketball. Associated Press, 17 July 1998.
Preferences or restrictions may narrow one's options. But in the face of equally attractive alternatives, one cannot say which one will choose in the future. If one can say what one will choose in the future when given the alternatives, then one has already decided. This is the case for the old members of my laboratory when we go for lunch. Some future-tense statements of choice can be intelligible if the choice involves limiting to some range of alternatives that will later be narrowed down. "I will buy one of the books from MIT Press" is an example.
Thus, a fundamental property of decisions is that they cannot be predicted, they can only be committed. Dennett (1984, p 77-78) writes, "This invisibility of causal paths is not just a matter of the invisibility (to us) of other minds. From our own first person 'introspective' vantage point, the causal paths are equally intractable. Consider decisions themselves, presumably the focal events in the mental life of a genuinely free agent. Are decisions voluntary? Or are they things that happen to us? From some fleeting vantage points, they seem to be the preeminently voluntary moves in our lives, the instants at which we exercise our agency to the fullest. But those same decisions can also be seen to be strangely out of our control. We have to wait to see how we are going to decide something, and when we do decide, our decision bubbles up to consciousness from we know not where. We do not witness it being made; we witness its arrival." (His emphasis) "Once we recognize that our conscious access to our own decisions is problematic, we may go on to note how many of the important turning points in our lives were unaccompanied, so far as retrospective memory of conscious experience goes, by conscious decisions." We should note as well that the unpredictability of decisions is as true for me predicting my decisions as it is for me predicting your decisions. We can see, then, two senses of 'choice', one identical with decision and incompatible with foreknowledge and another distinct from decision and compatible with foreknowledge -- "When I get to the bookstore, I am going to choose to buy this particular book." To qualify as "decision", according to this sense of the term, there must be a conscious, explicit concomitant of willful consideration.
What is a Choice?
The goal of this section is to arrive at a clear understanding of what it means to choose. We have distinguished decision as the conscious deliberation between alternatives with unclear outcomes. We have asserted that choices are closer to actions. Let us now examine the merits of different senses of 'choice'. First we consider the claim that a choice is a kind of action but that a choice is distinct from the overt action which is performed. For example, according to this definition, the overt act of signaling to the waiter in the restaurant which meal we desire is preceded by a covert act of choosing which meal. This view leads to two problems. If all actions are produced by preceding choices, then an infinite regress arises. If a choice is an action, then it must be preceded by a prior choice, which is itself an action preceded by a choice. The unacceptable conclusion, then, is that no one can act until they perform an infinite number of choices. A second problem is implicit in the sense of choice as a covert action. If choices are actions and if we maintain that we are able to choose any action which we believe we are able to perform, then it follows that we can choose to choose. For example, Jones believes she can skip class. On this view of choice, it follows that she can choose to skip class. If she reasons logically, it also follows that Jones believes she can choose to skip class. Now, we just said that we can choose any action that we believe we are able to perform. If Jones believes she can choose to skip class, then it follows that she can choose to choose to skip class. But the concept of "choosing to choose" makes no sense. If these arguments are sound, then it seems that we should reject the view that choices are actions.
An alternative view is that choice means "doing this instead of that", in other words performing one action out of a range of alternatives (Nowell-Smith, 1984, 1958), or obtaining one object from an array of alternatives. For example, ordering one of the possible meals at a restaurant. More particularly, we can claim that choices are overt actions done in a particular way. This claim rejects the view that there must be a mental act of choice prior to every action. Some actions in the face of alternatives can arise out of habit. This is a fitting description of certain members of my laboratory when we go to our favorite restaurant. Other circumstances present only Hobson's choice.
Three objections to this view have been suggested. Sometimes we choose to do something well before we do it, e.g.,"Jones chose to attend Vanderbilt two years before she graduated from high school". Therefore, choosing cannot always be the same as doing. An argument against this objection is that choosing has another meaning when we speak of choosing in advance. As we said above, this sense of choose becomes synonymous with decide. We can see the truth in this when we consider that in some contexts doing is a necessary condition for choosing (otherwise you'll never receive your meal at a restaurant), whereas doing is never a necessary condition for deciding (you can decide to have the same meal on the next visit). In the circumstance where doing is not a necessary condition for choosing, we can substitute 'decide' for 'choose' without confusion. Thus, according to this argument, the most fundamental meaning of choose is "doing this rather than that" because, unlike the other sense of choose in advance it is not synonymous with decide.
To sharpen these concepts, let us consider another example. Jones has just dropped her coins into a soft-drink vending machine. She presses the button to receive a Pepsi. Is it necessary that any deliberative, mental acts of choice pass through her mind before we correctly can say that she chose? She may say to herself "I choose Pepsi", but it does not seem necessary that she do so. What, then, is necessary? Jones must perceive the vending machine, believe she has alternatives and recognize Pepsi among the alternatives. But these processes do not constitute choosing. The view we are considering claims that she chooses just when she reaches to press the Pepsi button. We can emphasize the point that Jones's choosing Pepsi does not require preceding deliberation by recognizing that we can imagine easily Jones engaged in conversation at the same time as she makes the selection so that no thoughts pass through her mind which could plausibly be said to consist of choosing. The point of view we are considering holds that in the case of habitual choice nothing happens which should be called the choice beyond the act of pressing the button. Therefore, it is not necessary that a mental act of choice must precede every action performed in the face of alternatives.
The Difference Between Choosing and Deciding
Using the same example, we consider again the difference between choice and decision. Decision is more closely related to deliberation and intention than is choice. In the example of Jones at the vending machine, we should not say that she decided to select Pepsi because there was no deliberation, no delay between inserting the coin and pressing the button because of her habitual preference. Contrast the habitual choice of her favorite Pepsi with the deliberation that would transpire if the vending machine were out of Pepsi and she had to select from among less desirable alternatives.
As we said earlier, decisions cannot be known in advance. However, choices in the sense of doing this instead of that, selecting Pepsi instead of Coke, can be known in advance. In other words, we would be more confident speaking of a 'habitual choice' than we would 'habitual decision' (without being misunderstood). We can say that Jones decided to choose Pepsi before she arrived at the vending machine. This indicates that decision can precede choice. It also indicates that deliberation and resolution can be completed before a choice is made -- if by choice we mean the action of hitting the Pepsi rather than any other button on the vending machine. The converse is not sensible. We would not want to say that Jones "chose to decide". In this case 'chose' means the same as 'decided', i.e., to make up her mind about a future course of action. According to this point of view, to choose in advance means the same as to decide. Thus, whenever choosing is not "doing this instead of that", choosing is the same as deciding. The sense of choosing in advance is what we mean when we speak of making difficult choices.
However, nuances of a distinction remain between 'choosing in advance' and 'deciding'. The difference depends on an affinity that remains between choosing in advance and choosing in the fundamental sense of "doing this instead of that". The critical issue involves knowing how close the agent comes to making the selection. We wish to distinguish between a statement like "Jones decided what drink to have" and a statement like "Jones chose what drink to have". Imagine that she gave money and instructions to a friend who was going to the vending machine; then we should say "Jones chose Pepsi". If she merely thought to herself, "I will have a Pepsi", it is still possible that she could change her selection upon arriving at the vending machine and discovering a more attractive alternative. In this case we should say "Jones decided". This highlights another useful distinction between decision and choice. Decisions can change because they are not commitments; Jones could change her mind. Choices in the sense of doing this instead of that cannot be changed, because a commitment was made as the action was performed. If a commitment is made, we would say Jones chose in advance. In general, we would not care to say that Jones chose Pepsi before she arrived at the vending machine. But, we would be comfortable saying that Jones decided to have Pepsi before she arrived at the vending machine.
Is action a necessary condition of choice? Is a choice not finished until some basic act is produced, an utterance or a signature for example? That is, is doing this a necessary condition for choosing this? When we choose to do something among alternatives at once, the definition of choice we are considering posits that it is logically impossible to have chosen if one does not in fact perform the action. However, this seems curious because it does not actually appear to be logically impossible. For instance, what if Jones' body is suddenly paralyzed just before she reaches for the Pepsi button? Would we not wish to say that the choice was made even though no action was performed? Perhaps not. On the one hand, if there had been prior deliberation, then the sense in which we would say that the choice had been made is better understood as decide. On the other hand, if there had been no antecedent deliberation, if it was Jones' habitual choice that was prevented, then we would say there had been no decision, because there had been no deliberation, and no choice because there had been no action.
Linking Choosing and Deciding to Brain Processes
The scientific goal of this enterprise is to gain the ability to predict choices based on monitoring the brain. As a matter of fact, I have had this experience. When Nikos Logothetis and I recorded the activity of neurons in an area of the visual cortex of monkeys viewing a binocular rivalry motion display (Logothetis and Schall, 1989), we discovered that while recording from certain neurons we could predict what report the monkey would make 1.5 seconds later. I know of colleagues who have had the same experience (Mike Shadlen, personal communication). Likewise, if we understand how choices are made, then we should be able to influence choices by manipulating the brain. This has been accomplished in a number of ways. One of the most dramatic examples showed that mild electrical stimulation of an area of the visual cortex influenced the choice of monkeys in a motion discrimination task (Salzman and Newsome, 19__).
Establishing a clear distinction between choose and decide will be critical if we are to come to an understanding of the neural basis of decision making broadly considered. Clearly, to investigate choosing and deciding, it is necessary to create experiments in which subjects are presented with alternative responses. This has been done in a variety of settings (e.g. Newsome, Ramo, Schall, Wise). Macaque monkeys can learn to distinguish among different stimuli and even apply arbitrary rules (e.g. D. Pelligrino & Wise). The activity of individual neurons can be monitored in various parts of the brain while monkeys respond to the array of stimuli. Several investigators have described brain processes that are closely related to the choices monkeys make (e.g. Newsome, Schall...). The extent to which the neural processes correlate with the choices, they can be said to underlie in some way the choice. But should the neural activity be related to deciding or choosing?
Many behavioral neurophysiological studies have inserted a delay between the presentation of the stimuli on which the choice is based and the time when an action can be produced. This manipulation is supposed to provide time to monitor the decision process. We can imagine the neural correlate of choice as the transition between the balanced to the determined state that leads to this versus that action. If a delay is interposed between the neural state transition and the initiation of the action, then can we say that the longer the delay, the more the choice becomes like a decision. But if the neural state transition is maintained throughout the delay period, then the choice is equivalent to decision. What happens if the neural transition leading to the choice fades away? If it is committed to memory, then this can be regarded as a decision that will guide future behavior. The decision can be recalled from memory when the action is to be performed. When it is time to execute the action, the agent can say to herself, "Now, what did I decide I would do?" If the agent remembers, then the choice is executed according to the earlier brain state transition. However, if memory fails, then one cannot remember what one decided! And deliberation must begin anew. In this sense, monitoring brain states can inform us about whether an agent has chosen and whether the choice is still in place.
The ability to monitor brain processes associated reliably with choosing means that something that was subjective -- making a choice -- may become objective, i.e. publically observable. This will be one of the most interesting aspects of the work we will explore. One of the criteria for claiming that we understand the brain process responsible for decision process entails that we be able to predict and even influence the choice. Prediction represents an important level of understanding. But when we start considering the idea of predicting choices, we must confront core issues in the debate over free will and determinism, for can a choice that can be predicted be said to be free? What else would we need to know in addition to the state of a specific brain process to be able to predict a choice? The neural activity in and of itself is not sufficient. We will see that we need to know the alternatives, the context and even at least something about the history of the agent. To understand this, we will need to develop the distinction between different explanations of choices -- reasons and causes.
Choices and reasons
The position we have been developing does not explain the meaning of 'rather' in 'choosing is doing this rather than that'. The theory must explain what 'rather' means and how it matters to allow the definition of choosing to escape the preceding objections to the view that choices are actions. When we say that choosing is doing or taking one among alternatives, we exclude cases in which it makes no difference what the alternatives are. As Shakespeare observed, there's small choice in rotten apples. If Jones is happy to have any soft drink and her pressing of the Pepsi button reflects no particular preference, then we should say that she did not make a choice. According to this view, then, to say that one does this rather than that implies that there is a connection between which action one performs and some preference, intention, principle, resolution or habit. To explain this connection is to give a reason why this was chosen as opposed to that. If no reason can be produced, the theory would hold that there was no choice.
The relation of choices to reasons is also implicit in the claim that choice can be judged as good or bad. To say that "choosing this was good" means there is some basis for explaining how and why this is better than that. Reasons reflect preferences -- Pepsi over Coke for example. To explain how and why this is preferable to that, one lists the distinguishing features and how they relate to one's preferences or goals. Thus, a good choice is one that leads to the satisfaction of a preference or accomplishment of a goal. Bad choices lead to dissatisfaction and frustration. Therefore, we can say that every choice aims, with or without success, to suit a goal, principle or habit.
Reasons for actions are explanations of them in terms of goals, preferences or habits. Reasons are different from causes. The cause of some action must precede it. However, the time at which a reason is produced, or at which it is entertained by some mind, is irrelevant to whether or not it is a reason. The reason for an action can be formulated and stated before or after the action. In fact, a reason for an action need not even be formulated or stated explicitly to qualify as a reason. Similarly, the time at which a causal explanation is given can be after an action is performed even though the causal event must precede the effect. Reasons, as distinct from causes or statements of reasons or statements of causes, are not datable. When we say that Jones chose Pepsi because she preferred it over other soft drinks, we are not describing some event that preceded and caused the choice of Pepsi. Instead, we are saying that Jones considered Pepsi to be more suitable than any other soft drink to satisfy her thirst.
Reasons explain why but not how a choice was made. In this sense, an account of how the choice was made in terms of brain events would not be complete because it would not capture why. But if we wish to have a complete account of choosing and deciding in terms of brain processes, then we must also explain the neural correlates of reasons for choices. Can such a thing be done? A premise of this book is that we can if we operationalize reasons for choices through an understanding of the factors that influence the state changes in the sensorimotor parts of the brain. We can experimentally factor the influence of explicit memory (goals), implicit memory (habits) (e.g. Bichot & Schall, 1999) and motivation (preferences) (e.g. Platt & Glimcher, 1999; Hikosaka; Schultz). This is just what we need to do to explicate choice in terms of publically observable events.
The distinction between reasons and causes in the explanation of choices is a necessary aspect of the theory we are considering. Remember, if choices are actions that cause subsequent actions, then the act of choosing must precede the chosen action. But this is just the point of view we are trying to work away from. [cites Ryle, 1949, The Concept of the Mind, Hampshire 1959; Melden, 1961]. Thus, antecedent events play no logical role in the explication of the concept of choosing. Reasons are necessary conditions for choices, but the time at which evidence for the existence of a reason shows itself is not part of this necessary condition. Therefore, even if every choice were preceded by a mental act of selection, this would be only a contingent fact about choices, not part of the meaning of 'choice'.
What is an action? What does it mean to do something? One approach to this general question is to ask what is the order of events involved in the production of a voluntary action? One reasonable list begins with deliberation leading to a decision followed by the intention to act including brain processes, nerve conduction, muscle contraction, followed by a body movement. The movement can result in a series of further events extending into the future that appear to be causally related. Which of these are preliminary to action? Which are action? Which are consequences of action?
Let us consider the preliminaries of action. Often one does y by doing x, e.g., get a Pepsi by putting money into a vending machine. We may suppose that x (putting money into a vending machine) is really the action, and y (getting a Pepsi) is a consequence of the action x. But what if to do x, one must do w (grasp and release coins)? Then x and y would be consequences of w and so on. An infinite regress is avoided, though, because we are able to do some things directly, things that are not the consequence of some other action. This leads us to define a basic action as an action we perform without any preliminaries, such as raising a finger. We cannot say how we raise our finger; we will it and it happens. This view argues that all that we really do are basic actions and that all other occurrences are more correctly viewed as consequences of these basic actions.
The most plausible candidates for basic actions are choices (hitting the Pepsi button), acts of will ("Look left"), muscle contractions and the like - whatever can logically be the action that is causally first in the sequence of occurrences leading to the desired goal. We must distinguish many kinds of accompaniments of actions from actions themselves. Choices, we have already determined, are not actions themselves that are separate from overt actions. Neither are decisions actions. Effort is not an action; it characterizes how some action is performed. Efforts cannot be done alone and are not even intelligible apart from the actions which they modify. One can raise one's arm directly with no intervening steps, but an intention is always an intention to do something, a choice is always a choice of some action, and we cannot exert will unless it is the will with which we act or try to act.
Muscles contractions in most contexts are not basic actions but are necessary conditions for actions just as brain activation and nerve conduction are commonly inaccessible but always necessary preliminaries to action. Jones does not think about nor even have access to the events that make her fingers press the Pepsi button on the vending machine. These preliminary events can, though, under unusual circumstances become basic actions. Imagine that Jones is an amputee who learns to contract a particular muscle to operate a prosthetic limb. In this case, contraction of this specific muscle per se is the action. Or consider developing prosthetic neural implants that patients learn to operate (Chapin & Nicolelis, 1999). Producing the correct pattern of neural activation becomes a basic act least until the device operates according to design. Once the neural prosthetic operates consistently, the attribution of the basic act may migrate from the production of a pattern of neural impulses to producing the desired movement of the prosthetic. Thus, under typical conditions muscle contractions and neural impulses are necessarily preliminaries to actions but they are best regarded as events and not actions.
The distinction between actions for which an agent can be held accountable and events which happen may be clarified by considering data from electrical stimulation of the brain. In human patients, electrical stimulation of primary motor cortex artificially produces a movement . The patients report that they did not feel as if they willed the movement; instead, they report that the movement was as if someone else moved their body -- even though the neural pathway that would normally produce the movement was activated (Gray Walter; Dennett). In contrast, electrical stimulation of SMA produces movements or the subjective report of the urge to move. Also, alien hand syndrome.
The definition or attribution of an action depends on context. What we correctly call an action, as distinct from mere events, depends in important ways on what we can easily imagine to interest a person or to form part of some intelligible plan or aim. The muscle contraction practiced to operate a prosthetic limb meets this criterion, but the muscle contraction necessary for pushing buttons on vending machines does not. This analysis is based on the possibly unwarranted assumption that there can be only one correct description that applies to what one does. In fact, we will now learn that certain alternative descriptions of actions are equally valid (A. Goldman).
Actions and Consequences
We do actions to achieve goals. In other words, we do certain actions to produce desirable consequences. What criteria distinguish actions from consequences that follow them? For example, suppose that Jones speaks and in doing so answers a question, the consequence of which is that a test was passed. One could say that Jones answered a question and that a consequence of her action was that a test was passed. However, it is equally appropriate to say that Jones said a test-passing answer which in turn had various consequences such as finishing college. Hence, it will be useful to refer to "a test being passed" in a way that is neutral in regard to the categories action and consequences. Thus, we should consider everything that Jones produces when she acts to comprise a temporal series of occurrences extending indefinitely into the future. The example we are considering includes the occurrences "words being uttered", "an answer being given", "a test being passed", and "college being finished" and so on. We shall learn that we must resist the temptation to call all elements of the action-consequence sequence "events" rather than "occurrences" and to assume that they are causally related to one another. [See Viscinzy, Laws of Chaos]
We are free to regard "answering a question" as an action and "passing a test" as a consequence. We can also regard "passing a test" as an action with "finishing college" as a consequence. Because different descriptions of actions can apply, there is no rigid distinction between action and consequence. We are free to construe many occurrences either as consequences or as constitutive of an expanded version of the action. However, even though more than one description of an action can be valid, an infinite number of descriptions is not possible because we have criteria for classifying occurrences as constitutive of actions or as consequences. Intention, broadly defined, is one such criterion. We often say that what a person does is what she intends and that what she does not intend is a result or consequence. But this cannot be a sufficient criterion. If it were strictly so, then there could not be unintentional actions nor intended results.
When a person is intentionally performing an extended action, e.g., writing a book, it seems more appropriate because of the indirectness with which such things are accomplished to call them 'goals' and when they are achieved to call them 'results' and reserve the term 'action' for the various means to the goal, e.g. typing or writing.
Likewise, it cannot be disputed that unintended results occur, e.g., the coins are dropped into the vending machine but the vending machine is empty. The temptation for both extended goals and unintended results may be to restrict the term 'action' to the occurrence that is more causally primary. But this leads to the restricted use of 'action' to describe only basic actions, simple body movements. If this is so, then we are obliged to maintain that no one ever does anything except make simple movements of the body. Alternatively, we can grant that some 'later' consequences can be classified either as actions or as consequences of body movements. If we choose this alternative, then we cannot classify occurrences as results rather than as actions solely on the grounds that what precede and produce them can also be classified as actions.
We must work out how such criteria can be applied to nonhuman primates. If there is to be a neurophysiology of decision, choice and intention, then we must imagine how nonhuman primates can make decisions and choices and have intentions. Thus, many authors speak of a monkey making a voluntary eye movement in a task to get a juice reward. The basic act of an eye movement has a reason defined by a goal and motivation. One might object that the inability to verify plan and goals, verbally prohibits this attribution to nonhuman primates. However, it has been clear that natural, social behavior in nonhuman primates including macaques' so full of guile and cunning, can only be understood in terms of purposes, goals and plans (Science 1999 284:2070).
Consider a monkey performing a task to earn a juice reward. We should regard the body movements he makes as actions and receipt or withholding of the reward as a result. In fact, macaque monkeys -- which are the subject of effectively all neurophysiology experiments of interest for us -- have the mental capacity to plan only somewhat extended, arbitrary actions to achieve distant goals (See Tanji, but see recent Hikosaka paper). Therefore, the fact that we are concerned with neural correlates of simple choices, basic actions and direct results prevents us from having to be two distracted by the complexity of action descriptions.
Actions and causes
We should be careful in supposing that actions are to consequences as causes are to effects. First, not everything caused by an action is correctly regarded as a consequence of an action. [See the Douglas Adams passage on the chicken bone choking the fox... See also Rules of Chaos. See also Poincare chapter]. The effects of one's actions extend indefinitely into the future, but we limit the description of the consequences of what one does by considering what is important in the situation, what is of interest to us, what is not too distant in time and what can be explained by referring to intentions. These rather loose criteria are applied to categorize the products of human actions as consequences rather than merely effects of a cause.
Second, y can be a consequence of x when y is not at all caused by x. Consider the example of Jones raising her hand to ask a question in class versus Jones raising her hand to stretch versus Jones raising her hand to point. The signaling of a question is not caused by Jones raising her hand because there is nothing more to signaling a question than raising one's arm. In other words, there is no event that is part of signaling a question in class that is not also part of raising one's hand. If x causes y, then we know that two physical (or perhaps mental) events occur, but when Jones raises her hand to ask a question, only one physical event occurs. The example of Jones signaling a question by raising her arm contrasts with other sorts of actions such as Jones interrupting class by shouting a question. In the latter case there is an obvious causal relation between events constitutive of the action (shouting a question) and events constitutive of its consequence (interrupting the class).
The distinction between the two examples is captured by distinguishing between "doing y by doing x" and "doing y in doing x". The second formulation describes cases where doing x is part of what we mean by doing y. Thus, "Jones raised her hand" entails "Jones signaled a question". Also, "Jones signaled" can be construed as either an action or a consequence of an action. But how are we to distinguish the cases when Jones raised her arm to point or to stretch rather than signal? The view we are exploring would hold that which of these three actions Jones actually performs cannot depend on the events she produces. Thus, we must seek other criteria that justify the application of one action description rather than another. Also we must see whether it makes sense to say that these criteria cause Jones' action to be of one sort rather than another.
Practices and intentions
In the realm of human and other primate interaction at least two criteria may be applied to identify the correct action description. The first is the applicability of social practices or conventions. The second is the intention with which the action was done. We will see that actions like signaling or pointing, or other socially appropriate actions make, are logically dependent on social practices and intentions. Now, when we introduce a concept like social conversation to explain choices and actions, we must pause to consider whether this only applies to humans but not to nonhuman primates. In fact, the current view of many who study the behavior of chimpanzees, vervet monkeys, and even macaque monkeys is that certain behaviors can only be understood in a social context (e.g., Science 1999;284:2070).
Now, given a particular state of affairs, the existence of an intention or applicable practice logically entails but does not cause a given action. The existence of an intention or social practice logically entails but does not cause a given action. However, the relationship between intention or social practice and the production of an action can be complicated. That is, given an appropriate movement of the body (e.g., raising a hand), it is sometimes sufficient and other times necessary that one intend it for it to be correct to say that an action take place. Likewise, for an action to take place it is sometimes sufficient and sometimes necessary that a social practice or convention apply to it. Which of the alternatives - sufficiency or necessity - holds depends on the particular circumstance in which the action is done. For example, if Jones raises her arm, then whether she stretches, points or signals often depends on which of these she intends to do. Intending to point, because it is a communicative act, is usually a necessary condition for pointing, and if it is performed in the appropriate social context, then it is sufficient for pointing as well. However, if the party with whom Jones wishes to communicate is not looking at Jones when she points, then her raising of her hand was not sufficient to achieve her goal of pointing.
Thus, the intention to point is a critical part of how the action is interpreted. If this is so, then it seems that there are some actions which, at least in certain circumstances, cannot be done unintentionally. Similarly, other kinds of actions can, in many contexts, be done unintentionally. The professor may misinterpret Jones raising her hand to stretch or point as the action of signaling a question.
The concept of intention has been invoked to define decision and choice and to characterize action. The term intention has also been employed to characterize certain kinds of neural activity (e.g. Snyder). We must now develop a rigorous understanding of intention. Intentional acts are actions that can be explained in terms of the agent's purposes, goals or plans. According to this definition, the existence of prior mental events is not a necessary condition for the concept of intention.
The concept of intention seems necessary to answer the question -- why did she do that? If a person is asked, they report their reasons, what they intended to do or what consequences were intended. Clearly, the concept of intention has a rigorous usage in human interactions such as the legal arena (___?).
Can the concept be applied to nonhuman primates? It seems equally intelligible to ask -- why did the monkey do that? One answer can be in terms of neural processes. But this is not totally satisfactory or complete. The explanation needs to address the reasons based on preferences, goals and beliefs.
We can see that the concept of intention can be used in four ways (Aune, 19___). First are expressions of intention such as "I will buy Jones a Pepsi if the vending machine is not empty." Second, are ascriptions of intention such as "Jones intends to buy a Pepsi soda if the vending machine is not empty." Third, are descriptions of the intention with which some action is done such as "Jones intention in buying a Pepsi was to satisfy her thirst." Fourth, are classifications of actions as intentional such as "Jones shot him intentionally." The logical relation between these four uses of the concept of intention is clear. But this means that the word intention is ambiguous. The word may refer either to a state or episode or to the intentional object of such a state or episode. In other words, the word can refer to the state of intending or to that which is intended.
The two basic problems are first what sort of state or episode an intention is and second how are such states related to intentional actions. More opinions have been expressed on these questions than we can deal with here (see for example, Anscombe, 1957; Aristotle, 19__; Ryle, 1949; Dennett, 1984). Nevertheless, few would argue with the claim that at least one of the things conveyed by expressions of intention is, other things being equal, a readiness to initiate actions that one believes will obtain the intended object or circumstances. The validity of this claim is substantiated by the all too tendency for people to profess intentions that they really do not have and the strategy for refuting the profession. The straightforward refutation of a professed intention is producing evidence that the claimant made no attempts or exhibited no inclination to do what he said he intended to do. Thus, if Jones is in the presence of a vending machine but does not insert coins and press the Pepsi button, then we should conclude one of at least three things. First, Jones changed her mind. Second, Jones did not realize she was in the presence of the vending machine. Third, Jones did not really intend to obtain a Pepsi.
So the disposition to perform some act is a central defining feature of intentions. However, this disposition can be complex. First, a pure (?) Intention to perform an act is not intelligible. An additional necessary clement of intention is the circumstances under which the act will be performed. Contrast Jones saying "I will get a Pepsi" with Jones saying "I will get a Pepsi when we arrive at the vending machine if there are any left." The second statement has a credibility that the first is missing. Accordingly, the intention to do this under these circumstances can only entail the disposition to do this when one believes that these circumstances are present. We would have some insight into Jones' intention if the action is simple, such as a movement of her body -- "Jones intends to hit the Pepsi button on the vending machine." However, if the action is more complex -- "Jones intends to go to the vending machine," then we still do not know exactly what she will do. She may, for example, take the elevator or the stairs or go to a different vending machine than she usually does, or she may send a friend on the errand. Thus, when faced with intentions about complex, extended actions, to understand what someone will do, we must also know her beliefs about the preferred means by which the complex intention can be realized. In other words, what specific actions a person having the intention to do this under these circumstances will be disposed to perform can only be identified by referring to her preferences and her beliefs concerning what it means to be in these circumstances and what actions constitute the doing of this or will bring about that this is done.
When we see that understanding intention requires knowledge of these additional factors, then we see how complex it will be to gain insight into the neural basis of intention. Most of the claims, if not all, in the neurophysiological literature (e.g. Snyder et al) use the word intention in place of basic response preparation. We must distinguish rhetoric from accurate descriptions.
Now, because an intention involves the agent's beliefs about the conditions appropriate for its realization and beliefs about the preferred means of its realization, it follows that in addition to any purely behavioral disposition an intention will also involve some conception of what is intended.
Two properties of intentions. First, intentions may or may not be realized. Also an agent's intentions cannot always be determine by the way she behaves toward the object of her intentions; the object might not even exist. Second, a given action may be intentional under one description but not under another. For example, "Jones purchased her friend what she believe was her favorite drink, a Dr. Pepper. However, her friend prefers Root Beer." Thus, Jones did not intend to purchase her friend's least favorite drink. Having an intention amounts to conceiving in a particular way an action or state of affairs while in a state of readiness to do things that will, one believes, directly or indirectly realize the desired goal.
We are interested in intentions because we want to understand one another's behavior. Would we care to wonder about the intentions of a macaque monkey? We know that macaques can produce behaviors that call for descriptions in terms of intentions. For example (citation), a juvenile macaque whose mother is highest ranking in the troop has been known to vocalize as if he were being attacked by a lower ranking female who had a piece of fruit (?). When the mother ran to rescue her son, chasing hte allegedly offending female away, the juvenile retrieved the fruit that was left behind. Describing the connection between faking an attack and retrieving fruit seems to require or warrant the invocation of the concept of intention as much as any human action. Thus, even though work with monkeys prevents verbal exchanges, examples like this justify the conclusion that macaque monkeys have at least a measure of intention appropriate to relate to human agents. The same cannot be said for mice or rats, so this conclusion limits the range of species appropriate for investigation. The point is that reference to intentions is necessary to explain the behavior of humans and at least some nonhuman primates.
We now consider another defining property of intentions. Explanations of actions in terms of intention correspond or at least allude to the steps of reasoning that the agent might have taken in deciding what to do to realize his intention. Intending to get a Pepsi, Jones might say to herself, "I'm thirsty. I would like a Pepsi. I need to get enough money before going to the vending machine. I will take the elevator instead of the stairs." Thus, besides having the disposition described above, an agent who has a certain intention is likely to think thoughts that have the logical form of an intention, i.e. "I shall do this under these circumstances." Of course, the agent does not have to rehearse the thoughts about the intention until the first action is performed. The definition we are establishing only requires that the agent be disposed to think such thoughts, for example, when she is deliberating about what to do. So in this sense having an intention is a matter of being disposed first, when deliberating to formulate certain thoughts corresponding to the verbal statements used to express the intention and second, to perform actions that, one believes, will directly or indirectly obtain the intended goal.
Beyond the dispositional sense of intention discussed so far, we can also conceive of a nondispositional sense, that is to engage in acts of intending. Because resolving is an act that can be regarded as a special case of intending that is, an immediate consequence of deliberation leading to choice. Intentions in the form of dispositions may arise effortlessly and unconsciously like other seemingly spontaneous thoughts. However, intentions following deliberation and choice develop explicitly, consciously. This sense of intention has the character of resolve.
Intention is related to voluntary movement. Voluntary movements are those that are intended, one might say.
Ansombe GEM (1957). Intention, Oxford University Press.
Aristotle Nichomachean Ethics, Book III.
Ryle G (1949) The Concept of Mind.
Dennett D (1984). Elbow Room. Cambridge: MIT Press
Chisolm R (1964) Human freedom and the self. The Lindley Lecture, 1964, University of Kansas in Watson G (ed): (1982) Free Will, Oxford, Oxford University Press.