Flickers of Freedom

Neuroscience vs. Free Will? (Redux) Go....

Here is the abstract from a new paper by Adam Bear and Paul Bloom entitled, "A Simple Task Uncovers a Postdictive Illusion of Choice":

Do people know when, or whether, they have made a conscious choice? Here, we explore the possibility that choices can seem to occur before they are actually made. In two studies, participants were asked to quickly choose from a set of options before a randomly selected option was made salient. Even when they believed that they had made their decision prior to this event, participants were significantly more likely than chance to report choosing the salient option when this option was made salient soon after the perceived time of choice. Thus, without participants’ awareness, a seemingly later event influenced choices that were experienced as occurring at an earlier time. These findings suggest that, like certain low-level perceptual experiences, the experience of choice is susceptible to “postdictive” influence and that people may systematically overestimate the role that consciousness plays in their chosen behavior.

Unsurprisingly, the paper is generating some buzz in the popular press.  See here, here, here, here and here. Adam Bear has even written a popular press piece about this research in Scientific American entitled, "What Neuroscience Says about Free Will: We're convinced that it exists, but new research suggests it might be nothing more than a trick the brain plays on itself." So, it appears we go once more into this breach. What say you Al, Eddy, Neil and the usual crew of folks working on neuroscience, free will, etc.?

Related articles

Nahmias on Free Will at Scientific American

Consciousness and Moral Responsibility

Is Consciousness Important for Free Will? Why (Not)?

Superorganisms and Skepticism about Free Will

Over at Daily Nous and Practical Ethics, there are posts about this interesting piece by Peter Kramer and Paola Bressen entitled, "Humans as Superorganisms: How Microbes, Viruses, Imprinted Genes, and Other Selfish Entities Shape Our Behavior." Here is the abstract:

Psychologists and psychiatrists tend to be little aware that (a) microbes in our brains and guts are capable of altering our behavior; (b) viral DNA that was incorporated into our DNA millions of years ago is implicated in mental disorders; (c) many of us carry the cells of another human in our brains; and (d) under the regulation of viruslike elements, the paternally inherited and maternally inherited copies of some genes compete for domination in the offspring, on whom they have opposite physical and behavioral effects. This article provides a broad overview, aimed at a wide readership, of the consequences of our coexistence with these selfish entities. The overarching message is that we are not unitary individuals but superorganisms, built out of both human and nonhuman elements; it is their interaction that determines who we are.

Here is a table from the paper (also posted over at DN):

I have a longstanding interest in toxoplasma gondii (here is an old post of mine over at the Experimental Philosophy blog from all the way back in 2005!)--especially the controversial claim that the "manipulation hypothesis" purportedly applies to humans and not just other animals (see here and here). Needless to say, as a skeptic about free will and moral desert, I find the suggestion that we are not unitary agents but rather superorganisms intriguing when it comes to our agency and responsibility. From the manipulation question to the issue of whether we have a unified self, the data summarized by Kramer and Bressen seem relevant to the issues we explore here on this blog.  

It's no surprise that this issue has peeked the interest of scientists who are interested in free will.  For instance, here is a piece by Cristoff Koch in Scientific American entitled, "Protozoa Could Be Controlling Your Brain"--which begins with the following lead in/hook: 

THE ANCIENT DEBATE surrounding the existence of free will appears unresolvable, a metaphysical question that generates much heat yet little light. Common sense and volumes of psychological and neuroscientific research reveal, however, that we are less free than we think we are. Our genes, our upbringing and our environment influence our behaviors in ways that often escape conscious control. Understanding this influence, the advertisement industry spent approximately half a trillion dollars worldwide in 2010 to shape the buying decisions of consumers. And extreme dictatorships, such as that in North Korea, remain in power through the effective use of insidious and all-pervasive forms of propaganda. Yet nothing approaches the perfidy of the one-celled organism Toxoplasma gondii, one of the most widespread of all parasitic protozoa. It takes over the brain of its host and makes it do things, even actions that will cause it to die, in the service of this nasty hitchhiker. It sounds like a cheesy Hollywood horror flick, except that it is for real.

Another eminent scientist (and free will skeptic) has taken a similar line.  Here is an excerpt from an interview with Robert Sapolsky as Edge.org:

In the endless sort of struggle that neurobiologists have — in terms of free will, determinism — my feeling has always been that there's not a whole lot of free will out there, and if there is, it's in the least interesting places and getting more sparse all the time. But there's a whole new realm of neuroscience which I've been thinking about, which I'm starting to do research on, that throws in another element of things going on below the surface affecting our behavior. And it's got to do with this utterly bizarre world of parasites manipulating our behavior. It turns out that this is not all that surprising. There are all sorts of parasites out there that get into some organism, and what they need to do is parasitize the organism and increase the likelihood that they, the parasite, will be fruitful and multiply, and in some cases they can manipulate the behavior of the host.

As we learn more about "toxo" and other organisms that may influence our macro-behavior without our awareness, it is unsurprising that some might suggest we reevaluate the traditional way of viewing human thought, agency, and behavior. It is surprising, then, that there is no mention of toxoplasma gandii in all of the Philosopher's Index. While the aforementioned manipulation hypothesis was initially met with skepticism when applied to humans, the science has steadily pushed further towards that conclusion. Now this paper by Kramer and Bressen has compiled the findings on various other micro-organisms that (may) influence our behavior in ways that satisfy their interests while running counter to our own.

So, what say you, philosophers of action? What sense are we to make of the gathering data on microorganisms and macrobehavior? Can these findings be summarily dismissed? If so, why? If not, how should it affect what we say about agency and responsibility? If you hate headlines like "My brain made me do it," just think of how you'll respond to "My toxo made me do it"! For skeptics, this is just more grist for the mill. But for the rest of you? You may have some explaining (away) to do...

 

Neuroscience and Free Will Revisited

Free will makes the news once again.  This time the issue is raised over at the Genetic Literacy Project in a piece entitled, "Neuroscience of Free Will: Does Reaching for a Beer with a Robotic Arm Mean Free Will Doesn't Exist?" It's a surprisingly balanced piece, with quotes from our own Al Mele and some mysterious team from George [sic] State University. I can only assume it's Eddy Nahmias' doppelganger in a nearby world.

I am not sure why the fact that the drink is a beer is relevant, but I am suddenly finding myself strangely thirsty. Perhaps we're less free than we think after all (if such simple priming has me posting this and leaving shortly hereafter for the pub)! Or, perhaps weakness of will is real and some of us (for instance, me) do worse at avoiding it than others. Al's got a theory about that too.  But it ain't "lite"!

 

Unconscious Mens Rea, part 2

In my last post I discussed the case of Bert, the father who honestly forgot it was his weekend to take care of his kids, and was found guilty of child abandonment. I indicated that those who don’t attribute moral responsibility to agents for harm caused by lapses could support punishing Bert based on purely forward-looking aims of punishment, such as deterrence. Indeed, I suggested that some intuitions finding Bert responsible may stem from beliefs about deterrence and maybe not due to the presence of some condition that justifies holding Bert morally responsible.

However, I’m now going to come clean and admit that I think Bert IS morally responsible for the harm he caused to his kids. (Many of you who commented seemed to agree.) I’m going to argue that, if the facts of Bert’s case justify punishing him on grounds of deterrence, then those same facts also mean that Bert meets the conditions for minimal moral culpability.

Punishing Bert based on pure deterrence grounds indicates that we think Bert may be specifically deterred from child neglect and abandonment in the future due to his experience of being punished. This seems right: I can imagine that Bert, despite getting a sentence of probation, may be shamed enough by his encounter with the family court system to be more careful about caring for his kids in the future. Similarly, I can imagine a parent who heard about Bert’s case in the news being more careful about her custody schedule after hearing about Bert’s arrest. (This may be due to the worry about being punished, or just a heightened awareness of a possible lapse with regard to their kids’ care.)

If we think that Bert is capable of conducting himself such that he doesn’t lapse and violate the law due to the threat of punishment, this means we think the standard of care that constitutes a violation of the law is known, or should have been known, by Bert (if somewhat nebulously). Further, that Bert’s punishment can have this effect on his behavior suggests that the behavior triggering and targeted by the punishment is within his control.

I think this may mean that anyone who supported Bert’s forward-looking punishment actually implicitly supported assigning him direct moral responsibility. The efficacy of his forward-looking punishment rests on him having had the capacity to manage his behavior such that he met a particular standard of care. But the control and knowledge conditions discussed above sound suspiciously like grounds to attribute moral responsibility, even though Bert possessed no conscious awareness that the decision related to the lapse would cause criminal harm. For example, Fischer and Ravizza (1998) claim one has guidance control over an action if it is caused by a moderately reasons-responsive when an act is issued from a mechanism that is regularly receptive to reasons. If we think application of punishment to Bert means he, and others, may be deterred in future, we presumably also think the same threat of punishment may have deterred Bert, and the others, in the past. Thus we think Bert could have responded to the law as a reason not to lapse – that is, we think his act issued from a reasons-responsive mechanism. 

To put the argument in slightly different, more legalistic, terms: Those who would deny moral responsibility to Bert might support punishing him based upon forward-looking aims of punishment. But application of such punishment entails the idea that persons have the capacity to avoid lapses due to the threat of punishment. In Bert’s case, this means he both knew of and had the capacity to meet the legally required standard of care. When one has the capacity to know of and meet legally required standards of care, they are morally obligated to do so. Thus Bert is morally responsible for the violation of the standard of care due to his lapse.

This analysis of Bert’s case indicates his lapse may be best understood using a negligence standard. In most cases, a defendant found guilty of a crime is found to possess conscious mental states with regard the criminal harm she caused: she acted for the “purpose” of achieving the harm; she “knew” the harm was likely to occur; or she “recklessly disregarded” the likelihood of harm. However, in some cases criminal guilt is assigned even where there the defendant seems to have possessed no conscious mental states with regard to the criminal harm utilizing a negligence (or even a strict liability) standard. The law does this in cases where the criminal harm can be properly attributed to the defendant even though they lacked such agency, because he has culpably violated an important standard of care.

Violation of a standard of care is the hallmark of negligence. HLA Hart claimed instances of negligence were culpable as cases of unexercised capacity of choice to avoid risk (Hart, 1968). That is, Hart argued that assuming defendants have the capacity to choose to avert risk, they must have that capacity in at least some cases when they don’t avert risk, even where that risk was not consciously foreseen. Moore and Hurd, however, indicate negligence cases are a “dog’s breakfast,” meaning many of them appeal to different capacities, and have little in common besides a lack of specific mental states (Moore and Hurd, 2011). They provide a list of categories of capacities that defendants may fail to exercise such that they are negligently culpable, including: failure to address a dispositional lack of awareness of the risk; failure to address a risk related to character or psychological flaws the defendant knew about; and violation of a known mini-maximum (such as never leave a baby unaccompanied in the tub) (Moore and Hurd, 2011).

Note that in each of these types of negligence cases there is a capacity to abide by a standard of care (one was aware of or should have been aware of), and then a failure to exercise that capacity – exactly what is required for a punishment to have a deterrent effect. This is what amounts to “minimal culpability” in the law. Even in cases of strict liability – where the law claims to require the defendant possess no mental states with regard to the criminal harm – that harm has to be tied to a culpable (not “purely innocent”) act; it cannot be the result of an earlier, fully legal, decision (Brown, 2012). Brown argues that strict liability is applied in categories of cases where actors are oftentimes negligent, and in some cases reckless. The classic case of strict liability is statutory rape, where a defendant is held responsible for having sex with someone underage, even if they did not know what the person’s age was. Legal scholar D. Brown argues courts are especially willing to imply this required “minimal culpability” in cases involving a victim’s age or where the offense involves a weapon (Brown, 2012). “Strict liability does not reject culpability’s relevance but merely serves it imperfectly; strict liability is a clear rule with the inevitable weaknesses of clear rules—it is over inclusive for some cases” (Brown, 2012).

In my next post I want to explore minimal moral culpability in more detail. Can we hang such minimal culpability on indirect moral responsibility via tracing conditions? Matt King says no, and I’m inclined to agree.

---

 

The Origins of Imagination

I have argued this month that the ultimate source of LFW is the human capacity to imagine virtually anything and then enact it. But how is this awesome engine of novelty and creativity, rooted in spontaneous and unprecedented recombinations of existing representations and operations, realized in the brain?

Capacities that set humans apart in kind, not just in degree, from other animals include capacities for art, music, analogical reasoning, abstract thought, the spontaneous generation and use of symbols, and the ability to reason abstractly, as well as the ability to manipulate symbols recursively and syntactically. I have developed a theory (http://www.dartmouth.edu/~petertse/TseMoralityChapter.pdf ) that all these modes of cognition share a common root cause. My central claim is that neuronal circuits that were functionally distinct and modularly encapsulated in our chimp-like ancestors came to interact through a new type of attentional binding. Cross-modular binding, where operators previously limited to operands within a given module could now operate over the operands of other modules, is, I argue, the root cause of uniquely human modes of cognition.

Although animals can learn arbitrary associations, they are not using symbols as we do. Beyond lacking syntax, they are not typically able to effortlessly assign or reassign an arbitrary meaning to a given symbol once an association has been learned. It is this arbitrary and flexible relationship between a symbol and its referent(s) that is the hallmark of true symbolic thought, setting it apart from mere association.

A symbol has various aspects. One is the perceived sign, which can stand for one or more referents. A sign’s referent(s) can belong to many different types of mental representations, which themselves need not be signs. If the mental representation is one of an object (or type of object) in the outside world, then the sign symbolizes that thing in the world. When a sign has a referent (i.e., a meaning) for a particular perceiver, it is a symbol for that perceiver. A symbol is assumed to possess two key defining properties: (1) a symbol is a mental representation that can be stored in long-term memory or held in working memory that can stand for one or more arbitrary stored or online representations; (2) a symbol can be flexibly remapped to an existing or new referent without a need for many trials of learning to build up an association. In other words a symbol is arbitrary. Its meaning is not based on the likelihood or degree of co-occurrence between a sign and its referent. It is simply assigned.

Only humans, as far as we know, possess the capacity to process symbols in the sense of (2). A mind capable of (1) but incapable of (2) can learn complex associations between an object or event and some referent, but the association is not truly symbolic. Classical conditioning is an example of an ancient form of associational learning. In contrast, a three-year-old child can pretend that a block is a truck and then, a moment later, pretend that it is a monster. This capacity to instantly remap the referent of a representation maintained and manipulated in working memory (a WMR) is unique to humans and is at the heart of why our cognition is truly symbolic.  Symbols and symbolic thought are inherently attentional in nature, because they involve (now or involved in the past) the binding of a sign with arbitrary referents, whether temporarily in working memory or more durably in long-term memory. With unintentional repetition or intentional practice, arbitrary representations can become bound together in memory. This is the relatively slow process of associative learning that humans share with other animals. With attention, however, arbitrary representations can become bound together in working memory after one instance. Attentionally bound symbols and referents can be “chunked” and stored as a unit  in long-term memory, recalled as a unit, and later processed without the need for further attention. However, at the stage of encoding, one-shot binding of symbol and referent occurs because symbols and referents occupy a common WMR.

Many nonhuman animals appear to have the capacity to monitor, select, ignore, track, and otherwise attend to objects and therefore must have analogs of human WMRs in their cognitive architecture. A dog’s WMR of a tree, for instance, presumably contains color and shape information represented in different neural populations or maps, as well as other information that the dog has learned to associate with this particular tree or trees in general. All types of information, however, are about the tree. A dog’s WMR is encapsulated in the sense that it cannot contain information about irrelevant objects or events. WMR encapsulation presumably helps animals survive, because irrelevant information is not represented, permitting the animal to remain undistracted and unconfused by matters irrelevant to survival and the matter at hand.

Human WMRs, unlike animal WMRs, can contain any information which can be attended or downloaded from memory into the WMR. As such, human WMRs can enact truly cross-modular binding. Because any representation can be downloaded into the WMR of a tree, a tree can be taken to stand for “my friend Bob” if Bob and the tree occupy the same WMR. This requires tagging the tree component of the WMR as real (i.e., pointing to the world) and the Bob component as not real or referential (i.e., pointing to a representation that need not be in the world). In the absence of such a tag, a person might take a tree to really be Bob. Cognitive modularity and encapsulation protect animal minds from cognitive “noise.” That humans are prone to misrepresentation, psychosis, delusion, and hallucination is the price we pay for the cognitive freedom afforded by demodularization.

Whereas animals seem to be capable of internally modeling events that might happen or might have happened, humans go beyond this “literal” capacity to one of imagination that can model events and objects that could never happen and could never exist in the real world. Imagination became possible when arbitrary contents and operators could be downloaded to a common WMR. For example, the representation of wings could be downloaded into the WMR of a tree. The operator that places one object onto another could then be accessed to create a new representation of a tree with wings. A human WMR of a tree can contain everything that a dog’s would, plus information that has nothing to do with this particular tree or any tree. For example, a man looking at a tree can simply volitionally decide that it stands for his wife. This is accomplished by downloading representations of his wife into the WMR holding the tree information.

Analogy emerges because this downloading to the present WMR need not take place in a conscious manner or a manner dictated by the plans or goals of a central executive. Because of automatic cross-modular binding, he may see a tree standing next to a bush, and this may remind him of his wife and son because of their similar size relationships. This happened in a stimulus-driven manner because similarity in one domain automatically triggered activation on other maps encoded in terms of the same relationships, and these new activations, in this case corresponding to representations of his wife and son, were automatically downloaded to the WMR containing the representations of the tree and bush, which then could function as symbols of his wife and son. He might say that the tree and bush remind him of his wife and son, but this fact became available to his conscious report after the link between disparate representations had been made. Thus, symbols and symbolic relationships, although inherently attentional in nature, because initially mediated by the attentional construct of a WMR, are not necessarily volitionally invoked.

A dog cannot be reminded of anything by a tree and a bush other than things that have a direct link in its experience or in the world with a tree and a bush. Animal cognition, lacking any basis for symbolism, reminding, or metaphor, is inherently literal.

A key consequence of the emergence of cross-modular binding was that operators could be bound with new types of operands, such that operators from one module could operate on the operands of previously encapsulated modules. Nestable motoric operators could operate upon symbols rather than just physical actions, giving rise to the possibility of syntax. Another example would be that operators designed to decode the emotional content of vocalizations could now operate over nonvocal sounds, giving rise to the possibility of music. And basic aspects of art and human aesthetics may have emerged as a consequence of operators that became “disencapsulated” or “universalized.” For example, operators that evolved to discern genetic health in a potential mate could now operate over visual scenes, which could then attain the status of eroticism or beauty, although obviously a scene cannot be a mate.

The essence of abstraction is the ability to detect patterns that transcend  immediate sensory input. Pattern extraction is common to the perceptual systems of many animals, but in nonhuman animals, pattern extraction may be modularly isolated from the other contents of cognition and, thus, domain specific. Once modularity broke down because of cross-modular binding, the computations underlying pattern extraction and recognition could be applied to data that were not perceptual in nature. An example of perceptual pattern recognition that became abstract pattern recognition in humans involves the inference of causality. Other animals may be able to detect patterns of cause and effect in the flow of sensory input, because they carry out computations over sensory input that evolved specifically to solve this task. Because animals seem to be limited to causal pattern extraction from the current flow of sensory input, they may be limited to detecting physical causal relations that arise from spatiotemporal contiguity and simultaneity. In the human lineage, however, the computations dedicated to the extraction of patterns of causal relation in the sensory domain came to operate over the contents of other modules, permitting the extraction of patterns of causal relation that transcended spatiotemporal contiguity. For example, once causal relations could be detected in the contents of memory, patterns of cause and effect could be recognized that transcended the here and now. One could become aware of a time before one’s life or after one’s death. One could entertain the fact of one’s impending death. Simply put, whereas other animals can “connect the dots” perceptually, they cannot do so abstractly. Humans, in contrast, came to connect the dots to such an extent that hidden causes such as invisible beings and forces were invoked to explain events. Once patterns could be detected in information that was not perceptual, the human mind gained the capacity to become “unglued” from the here and now, free to connect the dots in novel ways beyond the here and now.

In sum, unconstrained  imagination is the ultimate source of human freedom. It evolved as a consequence of the demodularization of  neural circuitry associated with volitional attentional operations over operands downloadable  into a mental workspace where, virtually, anything is possible.

Why Regress Arguments Fail

There is an old regress argument against the possibility of self-forming acts that implies that we are not ultimately responsible for (the consequences of) how we choose to act. This is the regress whereby one would have to choose what kind of chooser to become on some grounds, which would require having to be able to choose those grounds on some metagrounds, and so on, ad infinitum.

In one variant, Strawson (1998) summarizes this ‘basic argument’ as follows: (1) One acts (or reacts to input) in a particular manner because of one’s physical/mental organization (or character or nature). (2) If one is to be ultimately MR for what one does, one must be ultimately MR for one’s physical/mental organization. (3) But one cannot be ultimately responsible for one’s physical/mental organization at the time of making a choice, because, simply put, one is what one is when one makes a choice. To choose the physical/mental organization that leads to that same choice, permitting a different choice to be made, would be a case of causa sui (self-causation), which is logically impossible. (4) So one cannot help but choose as one does at each instant because one cannot choose what one is (the character or nature that one has) at each instant, and one therefore is not ultimately MR for anything that one does.

The impossibility of self-causation is a valid argument against the possibility of an ability to choose the present physical/mental organizational grounds for making a choice at the present instant. But it is not circularly causal for a present choice to alter the physical/mental organizational  grounds of future choices. Criterial causation gets around the impossibility of self-causation by having neurons alter the physical grounds, not of the present mental events they realize, but of potential future mental events. They accomplish this by triggering changes in the physically realized informational/physical criteria for firing that must be met by future neuronal inputs before future neuronal firing occurs that realizes future mental events. In the case of a self-forming choice, one does not choose or alter the physical organizational basis for making a present decision or choice about how to be, one chooses or alters the physical organizational basis for making future decisions.

Choices that follow from such ‘future self’ self-forming acts we will have willed to take place by virtue of having set up those particular criteria in advance. When those criteria are satisfied in the future, leading to some decision, action or choice, those criteria cannot be changed, but because criteria can be changed in advance, we are to an extent free to decide on the parameters defining how we will decide or behave in the future. Criterial causation therefore offers a path toward a degree of self-formation and a degree of ultimate MR, where the brain can determine how it will decide/behave given particular types of future input. Self-formation involves reparameterizing how our future selves will choose to act.

Implicit in such regress arguments is that an infinite stack of decisions would have to be made in an instant. But decisions do not happen in an instant. They are cybernetic processes that come to conclusions over durations. If I want to decide on guests to invite for dinner, I might deliberate, playing out different scenarios in my mental workspace. The process is cybernetic because it involves feedback and comparison with some criteria I am trying to fulfill. If I imagine two guests who I then realize have split up, I will experience an error signal, and replace one or both of them with someone else, and then play things out again.

Self-forming choices are also durational and cybernetic, involving feedback, comparison with optimal fulfillment, and error signals. For example, say I have been fighting with my wife. After deliberating, I conclude that I am in part to blame because of how I react to her comments about, say, me working too much. I decide that, henceforth I will not respond in an angry way. I have set up grounds not for that decision, but for future decisions. Next time she makes a comment I might feel myself getting angry or defensive. But because I have preset new grounds by reparameterizing myself, there can be an error signal that indicates to me that I am heading towards a violation of these reformed grounds for my deciding. I correct myself, suppress my kneejerk anger, and respond in a way consistent with the reformed or reparameterized self I chose. I might say ‘you are right, I will come home earlier.’ I can then reparameterize myself to decide to come home earlier in the future. Days later, stressing out at work at 6PM about some grant deadline, I can feel an error signal or conflict between wanting to stay and my promise to get home. Whether I decide to honor my promise or not, I am beginning to develop a new character because I now feel torn, whereas previously I worked late without even thinking about it or its consequences.

If self-forming decisions about future choosing are not instantaneous, but rather cybernetic processes, then we can, as demanded by type 2 LFW, bootstrap ourselves over time into new characters. We might envision overcoming alcoholism, bind ourselves in various ways (e.g. reparameterize ourselves by swearing not to drink, reparameterize our world by throwing out all alcoholic drinks, joining AA etc.) such that when we feel the urge in the future, we are less likely to succumb. Months later we might have a new non-alcoholic character. The fact that there are millions of recovered alcoholics is testament to the human capacity to reform character.

Such bootstrapping might give rise to ‘Sorites worries.’ These are the ancient Greek heap arguments: A grain of sand is not a heap; Adding another grain does not make it a heap; On that logic, a million grains is also not a heap. But at that stage it patently is a heap. Similarly, a small act of future-self-reparameterization is not an entirely new character, and so on. And yet, after beating alcoholism or workaholism via the cumulative effects of thousands of small self-forming decisions, one patently does have a reformed character. Why do such heap arguments fail? Because the definitions of ‘heap’ or ‘character’ are criterial and therefore fuzzy. They set conditions on amounts or degrees of certain traits, not number.  

Of course we are not utterly free in making our choices because we cannot choose some of the initial bases of our choices. For example, what smells good or disgusting is just dictated to us by the inherited organization of our nervous system. We are not free to determine our initial conditions, since we have to start with some innate criteria at birth.  However, although we cannot choose what smells good, we can choose what we eat, because that decision can be made in light of criteria that we set on the basis of reason, before we smell anything delicious. For example, say a woman love’s the smell of cooked meat, and has a first-order desire for it. But based on her second-order desire not to hurt animals, and her reasoned conclusion that it is immoral to do so, she can set up criteria for deciding what to eat in the future, such that when she smells roasted meat she will decide not to eat it, but something vegetarian instead.  

A person can be responsible both for setting up the kind of criteria that would lead to certain kinds of future actions/choices and also for failing to inhibit criterial satisfaction (i.e., block the passing of the critical threshold) on the basis of higher-order criteria about what counts as right and wrong, through, for example, inhibition from executive centers in the frontal lobe. Consider the blameworthiness of a policeman who shoots and kills an unarmed citizen. On this account, he would not be responsible for pulling the trigger at the moment criteria for firing are passed, leading to a motor command to pull the trigger; instead he is responsible for (a) having certain decision criteria in place that he in part voluntarily placed there or did not voluntarily challenge when he could have, and (b) for failing to inhibit criterial satisfaction on the basis of knowledge of what is right and wrong, in the moments before the criteria were satisfied that led to him pulling the trigger. In a moral sense, the crime was not committed at the instant he pulled the trigger, but moments (or even longer spans of time) before that, by his having made decisions to put criteria in place or having failed to make decisions to train his nervous system to inhibit criterial satisfaction, which then led him to have the kind of nervous system that would lead him to pull the trigger under circumstances like those where he did pull the trigger, killing yet another unarmed citizen in America.

Onward and Upward with Downward Causation

Accounts of free will that require supernatural or contra-causal interventions have given libertarianism a bad name, and violate basic assumptions of physicalism and science. For any naturalistic variant of LFW to exist, several necessary conditions must be met. First, indeterminism must be ontologically real, rather than just a matter of epistemic uncertainty. Under indeterminism, I argued, Kim’s exclusion argument fails to rule out MC, leaving us with an opening to develop a believable account of MC. To get there, the following facts must in turn be true of neural processing: (1) quantum domain randomness must be amplified up to the level of randomness in macroscopic neural information processing, which (2) would have to be able to harness this randomness to fulfill information processing aims, and (3) there would have to be a role for the subclass of information processing that we call ‘mental,’ particularly the subclass of the mental that we regard as consciously volitional, in the specification of the ends to which such harnessing will apply, if conscious willing is to be agentic or causal of the realization of such aims. In my book I laid out a detailed case that these conditions are met, permitting the physical realization of a type 1 LFW. In order to have a type 2 LFW an additional condition would have to be met, namely: (4) the nervous system would have to be able to make decisions about how it would like to change itself, and then have the means to change itself, over time, into the intended type of decider.

LFW requires non-illusory downward MC. ‘Downward’ here means that events at a supervening level can influence outcomes at the rootmost level. In this context it would mean that information can bias which possible particle paths are realized. There is no wiggling out of this. If we want MC, and a FW or MR rooted in mental events that have/cause real consequences, we must defend the position that an informational entity, such as an intention or plan, can bias what possible particle paths open at the rootmost level can and do become real. But an entity at a supervening level cannot, logically, change its own physical basis because there can be no causa sui. This is where criterial causation via reparameterization comes in, because this allows what supervenes now to constrain what can supervene in the future.

How might such constraining work in the brain? The key pattern to which neurons respond is temporal coincidence. A neuron will only fire if it receives a certain number of coincident inputs from other neurons. Criterial causation occurs where physical criteria imposed by synaptic weights on coincident inputs in turn realize informational criteria for firing. This permits information to be downwardly causal regarding which indeterministic events at the rootmost level will be realized; Only those possible rootmost physical causal chains that meet physically realized informational criteria can drive a postsynaptic neuron to fire, and thus become causal at the level of information processing. Typically the only thing that the set of all possible rootmost physical causal chains that meet those criteria have in common is that they meet the informational criteria set. To try to cut information out of the causal picture here is a mistake; The only way to understand why it is that just this subset of possible physical causal chains—namely those that are also informational causal chains—can occur, is to understand what *informational* criteria delimit that class of possible outcomes.

As Eddy Nachmias put it my Oct. 2013 thread here: “the fact that informational state S1 could be realized by a range of physical states P1-PN and that info state S2 counterfactually depends on S1 but *not* any one of the specific physical states, including the one that actually realizes S1 on this occasion (e.g., P3) suggests that S1 is what makes a difference to S2 in a way that P3 does not. If we want to causally manipulate S2, manipulating P3 may not do it (e.g., if we alter it to P1 or P4, or one of the other S1 realizers); rather, we need to manipulate S1 (yes, by altering its realizers in the right way, but the right way will involve considerations of the S-level, not the P-level). S2 *rather than* S7 occurs *because* S1 *rather than* S1' occurred, and not because P1 rather than P4 occurred.”

Information only comes into existence by virtue of a decoder receiving input that matches its conditions (typically placed on the phase relationships or patterns in incoming energy) for the release of some effect, say, an action potential sent to other such decoders. But a decoder also serves as a ‘filter’ on the set of all potentially causal inputs, since it will only change the system of decoders in which it is embedded, namely, by firing, if its physically realized informational criteria are met.

Information cannot be anything like an energy that imposes forces, because it is not material even when it is realized in the material substrate. Information’s causal power consists in ‘filtering’ informational causal chains out of the set of all possible physical causal chains by constraining which sets of possible physical causal chains can occur. Although every informational causal chain is also a physical causal chain, most physical causal chains are not informational causal chains. Information is downwardly causal not as a material force, but as constraints that only allow the realization of sets of possible physical causal chains at the rootmost level that also comprise informational causal chains. Physical laws are not violated by this. Every possible physical causal chain conserves energy and momentum and so on. But only those possibilities allowed by physical laws which also meet informational criteria pass the physically realized informational filter, and become informationally causal, either by reparameterizing the criteria by which other neurons will assess future input, by changing their synaptic weights, or by triggering other neural firing.

Information is multiply realizable because which particular set of spike inputs ̶ and thus what particular information ̶ will make the neuron fire is unforeseeable, so long as the physical/informational criteria for firing are met. If neural causal chains are also informational causal chains, and informationally equivalent informational causal chains are realizable in multiple different neural or particle causal chains, then the parsimonious model is one of information causing information. Yes, there must always be some physical realization of information, but under physical/informational criterial causation, which one it happens to be is irrelevant so long as informational criteria are met. Chains of successive informational criterial satisfactions and criterial resettings afford the physical realization of downward MC.

Derk Pereboom in a previous thread said “if on some proposal, a dualist or a nonreductivist one, M and P are distinct causes of E, the threat posed by exclusionary reasoning will be neutralized by any response on which the number of causes is reduced to just one. There are two ways to achieve this: a first is by eliminating all but one of the causes, and the second is by identifying the causes.” If mental events are a type of information, and information is identical to acts of decoding immaterial (i.e. not made of mass) relationships or patterns among physical inputs, then mental events are identical to some class of acts of decoding. But note, this identity does not make mental events have physical properties like mass or momentum, because the identity is not with physical events at some instant, but with a process that is realized in physical events. Moreover, acts of decoding patterns cannot be reduced to a level where the patterns are not explicit, say the rootmost level, because the decoder only responds to a pattern at a level where it is explicit. Thus the identification is not with events at the microscopic level or even the neuronal level, but at the level of decoding the non-physical patterns to which the decoder is sensitive. Under determinism supervening informational criteria cannot filter out possible but non-informational causal chains at the rootmost level, because there is only one possible causal chain. But if indeterminism is the case, supervening informational criteria can make a difference regarding which possibilities at the rootmost level happen. That is, under indeterminism but not determinism, there is non-redundant causal work for informational criteria to do.

But how does this give the brain the capacity to freely will? It is not enough for neurons to filter out non-informational possible physical causal chains. It must be the case that some neural activity that we associate with volition, can control the parameters that neurons will apply in the future to enact such acts of filtering. Control comes from executive circuits that can plan, imagine, deliberate and make decisions in light of highest level demands and needs, and that can ‘rewire' circuits and reparameterize neurons, by changing synaptic weights, to embody new criteria for firing that will fulfill current executive ends. The downward causation afforded by the informational filtering of possible rootmost causal chains becomes agentic downward causation when executive circuits can rewire lower level circuits to fulfill whatever criteria they demand.

How to Escape Kim’s Logical Headlock

If the logic of the EA (given in the previous post) is valid, then only if either (i) or (ii) is incorrect, is there potentially room to develop a theory of MC. So any theory of MC that attempts to meet “Kim’s challenge” must explicitly state which premise, (i) and/or (ii), is incorrect.

To begin with, note that without the sufficiency of c, Kim cannot apply the “exclusion of over-determination” principle, so cannot rule out mental causation on the basis of the EA. The sufficiency of c is crucial if the EA is to succeed at excluding MC.

 In my book and on this blog last in Oct. 2013 I argued that the EA does not hold under indeterminism. I will repeat the argument again here because I think it is a necessary part of the foundation for everything I later want to say about how MC (including volitional MEs associated with FW) and downward causation work in the brain. If I am wrong about there being an escape from Kim’s logical headlock, the foundation of the house of MC I want to build collapses. That is, if the EA also holds under indeterminism, then MC, FW and MR are ruled out, period, and I would have to become a hard incompatibilist, whether determinism or indeterminism is the case. Is there an escape from such a bleak worldview?

Here goes. Premise (ii) does not hold if indeterminism is the case, because any particular present microphysical state is not necessitated by its antecedent microphysical state or states. In other words the traditional definition of causal closure that “every physical event has an immediately antecedent sufficient physical cause” is not satisfied, because when a cause c can be indeterministically followed by any number of possible effects e_i, then c is not a sufficient cause of any of the possible e_i, because they might not happen if they have not yet happened, and they might not have happened even after they have happened. 

Classical deterministic laws are laws that apply among sufficiently causal actualia, where both c at t1 and e at t2 are actual events. Quantum mechanical laws are instead deterministic at the level of possibilia, but indeterministic at the level of actualia, because which possible outcome will occur upon measurement is only probabilistically specifiable. Nonetheless, under quantum mechanics c is sufficiently causal of its entire set of possible outcomes e_i with their associated probabilities of occurring. It is just that c is not a sufficient cause of any particular one of its many possible effects that happens to happen when measured. Classical deterministic and modern quantum mechanical laws both operate deterministically, and causation is sufficient, but over different types of physical entities. Actualia and possibilia, while both physical, have mutually exclusive properties. Actualia are real and exist now or in some past moment; they have a probability of 1 of happening or having happened. Possibilia are not yet real and may never become real, and exist in the future relative to some c, and have a probability of happening between zero and one. A given event cannot be both actual and possible at the same time.

Closure, therefore, applies to different types of physical events under ontological determinism and indeterminism. “Closure” entails that the set of physical events is closed; Any particular effect will be a member of the same set to which a sufficient cause itself belongs. Determinism is closed at the level of actualia; Any particular cause or effect will be a member of the set of all actual events in the universe across all time. Indeterminism, in contrast, is not closed at the level of actualia because a non-sufficient actual cause and one of its possible outcomes that may never happen are not both members of the set of actualia. Rather, quantum theory is closed (and deterministic) at the level of possibilia: Any particular outcome or event will be a member of the set of all possible outcomes or events in the universe across all time, and any possible cause is sufficient to account for the set of all of its possible effects. Under indeterminism physical explanations are of a different type than under determinism, though both actualia and possibilia are physical, and theories of either are physical explanations.

An indeterministic causal closure thesis could be restated as follows: “(ii*) the set of all possible microphysical states is completely diachronically necessitated by antecedent possible microphysical states.” The realization thesis for the indeterministic case might be: “(i*) all mental states are synchronically determined by underlying sets of possible microphysical states.” But claim (i*) is contrary to the definition of supervenience. Mental events do not supervene on sets of possible physical states, they supervene on specific, actually occurring physical states. Since it is absurd to maintain that mental events synchronically supervene on sets of possibilia, we can rule (i*) out. It remains to be shown whether (i), i.e. supervenience on actualia, can be combined with (ii*), i.e. causal sufficiency and closure among possibilia, to yield (iii).

An actual microphysical state and the set of all possible microphysical states are different kinds with mutually exclusive properties (e.g., real/~real; present/~present). The essentially syllogistic structure of the exclusion argument requires staying within a logical kind. It is logically valid to draw from the major premise (ii) ‘All physical events are caused by preceding sufficient physical causes’ and the minor premise (i) ‘mental events are realized in physical events’ the conclusion (iii) that ‘the physical events that realize mental events have preceding sufficient physical causes’. But now we are splitting ‘physical’ into two types with mutually exclusive properties, possibilia and actualia. The conclusion (iii) of the syllogism holds only if both the major and minor premises hold and are both are about actualia as in (ii) and (i), or both are about possibilia as in (ii*) and (i*). If one premise is about possibilia and the other about actualia, the conclusion does not follow, because the premises are about exclusive entities. For example, (ii) and (i*) would read ‘All actual physical events are caused by preceding sufficient actual physical causes’ and ‘mental events are realized in sets of possible physical events,’ which violates syllogistic logic as much as ‘all men are mortal’ and ‘Socrates is a robot’. Conversely, (ii*) and (i) would read ‘The set of possible physical events are caused by preceding sufficient possible physical causes’ and ‘mental events are realized in actual physical events,’ which similarly violates syllogistic logical form.

In conclusion, if determinism is the case, then the EA holds, and MC, FW and MR are ruled out, as is compatibilism. However, if indeterminism is the case, the EA does not hold and MC and FW and MR are not logically ruled out, at least by the EA.

This still leaves open the hard work of explaining how MC might work in a brain that harnesses indeterministic events to make information downwardly causal, which I will take up in my next post on downward causation.

The Exclusion Argument Excludes Compatibilism Too

First, let me apologize for not answering people’s comments promptly. Now that my grant applications are done, I should be quicker to respond, and promise to respond to the many comments I have not responded to yet.

There is a deeper, logically prior problem than the FW and MR problems, namely, the mind-body problem. Granted, there are really several mind-body problems (among them the neural basis of consciousness, the neural code, the problem of how the mental is physically realized, and physically yet downwardly causal). But regarding FW and MR, the most fundamental issue is whether mental events (MEs) can be causal. If not, then the conscious deliberations we so treasure as the basis of both FW and MR are at best epiphenomenal. In reading the FW debates from recent decades, it seems that the majority of argumentation among philosophers who believe in FW involves compatibilists responding to the consequence argument (CA), and incompatibilists responding to Frankfurt cases. I think this is optimistic in the same way that arguing over how the furniture should be arranged while the house is on fire is optimistic. First we have to put the fire out, because unless we do that, there is simply no point in arguing any further.

What is the logical fire threatening to burn down the house of (free) mental causation (MC)? It is Jaegwon Kim’s exclusion argument (EA). If the EA is true, then there can be no MC. But if MEs cannot be causal, volitional MEs such as willings cannot be causal. If a fundamental component of any definition of FW is that a willing causes events to happen, but willings cannot be causal because MEs cannot be causal, then FW is not possible. Ruling out MC would rule out FW. Similarly, if a fundamental component of any definition of MR is that an act of willing causes consequences to happen for which we are in part blameworthy or praiseworthy, but willings cannot be causal because MEs cannot be causal, then MR is not possible either. No MC, no FW and no MR.

One way to try to skirt this would be to let go of mental causation, and simply try to ground FW and MR solely in the neural basis of mental events. Someone making this move would redefine FW away from mental FW and the conscious deliberations that nearly everyone really hopes to find causal. Mele, for example, hedges his bets when he frames causation disjunctively by saying that a ME such as an intention OR its physical basis causes a decision. If the EA holds, however, a ME cannot be causal, leaving only open the possibility that its physical basis is causal as a physical-on-physical event. That might be a way to save FW from the EA, but would come at the cost of accepting that mental willings cause nothing. But are physical neural events really willings? Neurons have mass. Would willings then have mass? Regardless, this move fails because the EA makes neural causation epiphenomenal as well, assuming determinism, since it also reduces to particle-on-particle causation. If we lose both MC and neural causation to the EA under determinism, we lose FW and MR utterly.

Another way to look at this is that the EA is a more devastating argument against compatibilism than the CA. The CA is problematic, I think, for at least two basic reasons. First it makes no explicit mention of causation, and second it is a modal argument. Being modal it leaves open ambiguities in the definitions of (and referents of) basic terms used in the CA (e.g., are we talking about laws-in-themselves, laws-as-modeled by science, or just conceivable laws?; are the laws in the past laws of causation or something else?; Is causation necessitarian in that it completely specifies what will happen in the future given what is?; Does causation only run forward in time?; are we talking about objectively being able to realize certain possibilities, or only conceivably being able to realize certain possibilities, whether in the past or the future?). This kind of ambiguity in what we are talking about when we talk about the past, the laws or a modal operator like van Inwagen’s ‘N’ leaves a lot of wiggle room for both confusion and talking past one another in the seemingly endless or at least stalemated back and forth over the CA. The EA, in contrast, is explicitly (not implicitly like the CA) about causation, and the EA does not provide any modal or conditional escape routes for compatibilists to take, because the EA is not a modal argument at all.

The EA rests on a premise of the causal closure of the physical. “Causal closure” means that causality at the level of particles is sufficient to account for all outcomes and interactions at the level of particles. Kim (2005, p. 17), applying Occam’s razor, advocates the “exclusion of over-determination” when modeling physical causation. In his words: “If event e has a sufficient cause c at t, no event at t distinct from c can be a cause of e.”  If particle-level causality is sufficient to account for particle behavior, and neurons are made of particles, MEs, assuming that they supervene on neuronal and particle events, can play no causal role in neuronal or particle behavior. In other words, MEs cannot cause fundamental particles to behave differently than they otherwise would have if they had only interacted according to the laws obeyed by particles. And the same goes for neural events.

Put succinctly (Kim, 1993, pp. 206–210): If (i) the “realization thesis” is the case, then each ME is synchronically determined by underlying microphysical states, and if (ii) “the causal or dynamical closure of the physical thesis” is the case, then all microphysical states are completely diachronically necessitated by antecedent microphysical states, then it follows that (iii) there is no causal work left for MEs as such to do.

I believe the EA holds if determinism is the case, ruling out MC and thereby mental FW and mental MR, though I would be delighted if someone could convince me that that is not true. Condition (i) must be true, assuming monism/physicalism, because mental events must be realized in their neural, and ultimately particle-level basis. But if determinism is true, then (ii) is also true because then events are sufficient causes of their subsequent outcomes, at least under a causally necessitarian understanding of determinism, such as held by, say, Newton or Einstein. If determinism is true (iii) follows. Thus if determinism is true, the EA rules out MC, which in turn rules out the kind of mental FW and MR most people want. Therefore FW and MR are not compatible with determinism. Note that this basis for adopting incompatibilism (of FW and MR with determinism) does not depend, like the CA, on any modal arguments with multiply interpretable operators, definitions of basic terms, or referents. The EA is so deadly for compatibilism in its simplicity because it follows from widely held assumptions of reductionistic physicalism about mental realization and physical causation.

My next post (above) will repeat some things I have written about concerning how to escape the EA, before moving on to how MC (and FW as the class of volitional MEs) might actually work in the brain.