Patricia Smith Churchland

How do neurons know?

My knowing anything depends on my ride a bicycle, I have to try over and neurons–the cells of my .1 More over; by contrast, learning to avoid eat- precisely, what I know depends on the ing oysters if they made me vomit the speci½c con½guration of connections last time just happens. Knowing how to among my trillion neurons, on the neu- change a tire depends on cultural arti- rochemical interactions between con- facts, but knowing how to clap does not. nected neurons, and on the response And neurons are at the bottom of it all. portfolio of different neuron types. All How did it come to pass that we know this is what makes me me. anything? The range of things I know is as di- Early in the history of living things, verse as the range of stuff at a yard sale. evolution stumbled upon the advantages Some is how, some knowl- accruing to animals whose nervous sys- edge that, some a bit of both, and some tems could make predictions based upon not exactly either. Some is fleeting, some past correlations. Unlike plants, who enduring. Some I can articulate, such as have to take what comes, animals are the instructions for changing a tire, movers, and having a brain that can some, such as how I construct a logical learn confers a competitive advantage in argument, I cannot. ½nding food, mates, and shelter and in Some learning is conscious, some not. avoiding dangers. Nervous systems earn To learn some things, such as how to their keep in the service of prediction, and, to that end, map the me-relevant Patricia Smith Churchland is uc President’s Pro- parts of the world–its spatial relations, fessor of Philosophy and chair of the philosophy social relations, dangers, and so on. And, department at the University of California, San of course, map their worlds in Diego, and adjunct professor at the Salk Institute. varying degrees of complexity, and rela- She is past president of the American Philosophi- tive to the needs, equipment, and life- 2 cal Association and the Society for Philosophy style of the organisms they inhabit. and Psychology. Her latest books are “Brain- 1 Portions of this paper are drawn from my Wise: Studies in ” (2002) and book Brain-Wise: Studies in Neurophilosophy “On the Contrary: Critical Essays, 1987–1997” (Cambridge, Mass.: mit Press, 2002). (with Paul Churchland, 1998). 2 See Patricia Smith Churchland and Paul M. Churchland, “Neural Worlds and Real Worlds,” © 2004 by the American Academy of Arts Nature Reviews 3 (11) (November & Sciences 2002): 903–907.

42 Dædalus Winter 2004 Thus humans, dogs, and frogs will repre- latches its mouth onto a nipple, and be- How do neurons sent the same pond quite differently. The gins to suck. A kitten thrown into the air know? human, for example, may be interested rights itself and lands on its feet. A hu- in the pond’s water source, the potability man neonate will imitate a facial expres- of the water, or the potential for irriga- sion, such as an outstuck tongue. But tion. The dog may be interested in a cool other knowledge, such as how to weave swim and a good drink, and the frog, in a or make ½re, is obviously learned post- good place to lay eggs, ½nd flies, bask in natally. the sun, or hide. Such contrasts have seemed to imply Boiled down to essentials, the main that everything we know is either caused problems for the neuroscience of knowl- by genes or caused by experience, where edge are these: How do structural ar- these categories are construed as exclu- rangements in neural tissue embody sive and exhaustive. But recent discover- knowledge (the problem of representa- ies in molecular biology, neuroembryol- tions)? How, as a result of the animal’s ogy, and neurobiology have demolished experience, do neurons undergo changes this sharp distinction between nature in their structural features such that and nurture. One such discovery is that these changes constitute knowing some- normal development, right from the ear- thing new (the problem of learning)? liest stages, relies on both genes and epi- How is the genome organized so that the genetic conditions. For example, a fe- nervous system it builds is able to learn male (xx) fetus developing in a uterine what it needs to learn? environment that is unusually high in The spectacular progress, during the androgens may be born with male-look- last three or four decades, in genetics, ing genitalia and may have a masculin- psychology, , neuroem- ized area in the hypothalamus, a sexually bryology, and neurobiology has given dimorphic brain region. In mice, the the problems of how brains represent gender of adjacent siblings on the pla- and learn and get built an entirely new cental fetus line in the uterus will affect look. In the process, many revered para- such things as the male/female ratio of a digms have taken a pounding. From the given mouse’s subsequent offspring, and ashes of the old verities is arising a very even the longevity of those offspring. different framework for thinking about On the other hand, paradigmatic in- ourselves and how our brains make stances of long-term learning, such as sense of the world. memorizing a route through a forest, re- ly on genes to produce changes in cells Historically, philosophers have debat- that embody that learning. If you experi- ed how much of what we know is based ence a new kind of sensorimotor event on instinct, and how much on experi- during the day–say, for example, you ence. At one extreme, the rationalists ar- learn to cast a ½shing line–and your gued that essentially all knowledge was brain rehearses that event during your innate. At the other, radical empiricists, deep cycle, then the gene zif-268 impressed by infant modi½ability and will be up-regulated. Improvement in by the impact of culture, argued that all casting the next day will depend on the knowledge was acquired. resulting gene products and their role in Knowledge displayed at birth is obvi- neuronal function. ously likely to be innate. A normal neo- Indeed, ½ve important and related dis- nate rat scrambles to the warmest place, coveries have made it increasingly clear

Dædalus Winter 2004 43 Patricia just how interrelated ‘nature’ and ‘nur- have only about thirty thousand genes, Smith Churchland ture’ are, and, consequently, how inade- and we differ from mice in only about on quate the old distinction is.3 three hundred of those;4 meanwhile, learning First, what genes do is code for pro- we share about 99.7 percent of our genes teins. Strictly speaking, there is no gene with chimpanzees. Our brains and those for a sucking reflex, let alone for female of other primates have the same organi- coyness or Scottish thriftiness or cogni- zation, the same gross structures in zance of the concept of zero. A gene is roughly the same proportions, the same simply a sequence of base pairs contain- neuron types, and, so far as we know, ing the information that allows rna to much the same developmental schedule string together a sequence of amino and patterns of connectivity. acids to constitute a protein. (This gene Fourth, given the high degree of con- is said to be ‘expressed’ when it is tran- servation, whence the diversity of multi- scribed into rna products, some of cellular organisms? Molecular biologists which, in turn, are translated into pro- have discovered that some genes regu- teins.) late the expression of other genes, and Second, natural selection cannot di- are themselves regulated by yet other rectly select particular wiring to support genes, in an intricate, interactive, and a particular domain of knowledge. Blind systematic organization. But genes (via luck aside, what determines whether the rna) make proteins, so the expression animal survives is its behavior; its equip- of one gene by another may be affected ment, neural and otherwise, underpins via sensitivity to protein products. Addi- that behavior. Representational prowess tionally, proteins, both within cells and in a nervous system can be selected for, in the extracellular space, may interact albeit indirectly, only if the representa- with each other to yield further contin- tional package informing the behavior gencies that can ½gure in an unfolding was what gave the animal the competi- regulatory cascade. Small differences in tive edge. Hence representational so- regulatory genes can have large and far- phistication and its wiring infrastructure reaching effects, owing to the intricate can be selected for only via the behavior hierarchy of regulatory linkages between they upgrade. them. The emergence of complex, inter- Third, there is a truly stunning degree active cause-effect pro½les for gene ex- of conservation in structures and devel- pression begets very fancy regulatory opmental organization across all verte- cascades that can beget very fancy or- brate animals, and a very high degree of ganisms–us, for example. conservation in basic cellular functions Fifth, various aspects of the develop- across phyla, from worms to spiders to ment of an organism from fertilized egg humans. All nervous systems use essen- to up-and-running critter depend on tially the same neurochemicals, and where and when cells are born. Neurons their neurons work in essentially the originate from the daughter cells of the same way, the variations being vastly last division of pre-neuron cells. Wheth- outweighed by the similarities. Humans er such a daughter cell becomes a glial (supporting) cell or a neuron, and which 3 In this discussion, I am greatly indebted to type of some hundred types of neurons Barbara Finlay, Richard Darlington, and Nich- olas Nicastro, “Developmental Structure in 4 See John Gerhart and Marc Kirschner, Cells, Brain Evolution,” Behavioral and Brain Sciences Embryos, and Evolution (Oxford: Blackwell, 24 (2) (April 2001): 263–278. 1997).

44 Dædalus Winter 2004 the cell becomes, depends on its epige- learning to read braille.6 So long as the How do neurons netic circumstances. Moreover, the blindfold remains ½rmly in place to pre- know? manner in which neurons from one area, vent any light from falling on the retina, such as the thalamus, connect to cells in performance of braille reading steadily the cortex depends very much on epige- improves. The blindfold is essential, for netic circumstances, e.g., on the sponta- normal visual stimuli that activate the neous activity, and later, the experience- visual cortex in the normal way impede driven activity, of the thalamic and corti- acquisition of the tactile skill. For exam- cal neurons. This is not to say that there ple, if after ½ve days the blindfold is re- are no causally signi½cant differences moved, even briefly while the subject between, for instance, the neonatal suck- watches a television program before ing reflex and knowing how to make a going to sleep, his braille performance ½re. Differences, obviously, there are. under blindfold the next day falls from The essential point is that the differ- its previous level. If the visual cortex can ences do not sort themselves into the be recruited in the processing of nonvi- archaic ‘nature’ versus ‘nurture’ bins. sual signals, what sense can we make of Genes and extragenetic factors collabo- the notion of the dedicated vision mod- rate in a complex interdependency.5 ule, and of the dedicated-modules hy- pothesis more generally? Recent discoveries in What is clear is that the nature versus point in this same direction. Hitherto, it nurture dichotomy is more of a liability was assumed that brain centers–mod- than an asset in framing the inquiry into ules dedicated to a speci½c task–were the origin of plasticity in human brains. wired up at birth. The idea was that we Its inadequacy is rather like the inade- were able to see because dedicated ‘visu- quacy of ‘good versus evil’ as a frame- al modules’ in the cortex were wired for work for understanding the complexity vision; we could feel because dedicated of political life in human societies. It is modules in the cortex were wired for not that there is nothing to it. But it is touch, and so on. like using a grub hoe to remove a splin- The turns out to be much more ter. puzzling. For example, the visual cortex of a An appealing idea is that if you learn blind subject is recruited during the something, such as how to tie a trucker’s reading of braille, a distinctly nonvisual, knot, then that information will be tactile skill–whether the subject has ac- stored in one particular location in the quired or congenital blindness. It turns brain, along with related knowledge– out, moreover, that stimulating the say, between reef knots and half-hitches. subject’s visual cortex with a magnet- That is, after all, a good method for stor- induced current will temporarily impede ing tools and paper ½les–in a particular his braille performance. Even more re- drawer at a particular location. But this markably, activity in the visual cortex is not the brain’s way, as Karl Lashley occurs even in normal seeing subjects ½rst demonstrated in the 1920s. who are blindfolded for a few days while 6 See Alvaro Pascual-Leone et al., “Study and Modulation of Human Cortical Excitability 5 See also Steven Quartz and Terrence J. Sej- with Transcranial Magnetic Stimulation,” Jour- nowski, Liars, Lovers, and Heroes (New York: nal of Clinical 15 (1998): 333– William Morrow, 2002). 343.

Dædalus Winter 2004 45 Patricia Lashley reasoned that if a rat learned are trying to under- Smith Churchland something, such as a route through a stand the structure of learning by using a on certain maze, and if that information variety of research strategies. One strat- learning was stored in a single, punctate location, egy consists of tracking down experi- then you should be able to extract it by ence-dependent changes at the level of lesioning the rat’s brain in the right the neuron to ½nd out what precisely place. Lashley trained twenty rats on his changes, when, and why. Another strate- maze. Next he removed a different area gy involves learning on a larger scale: of cortex from each animal, and allowed what happens in behavior and in partic- the rats time to recover. He then retested ular brain subsystems when there are le- each one to see which lesion removed sions, or during development, or when knowledge of the maze. Lashley discov- the subject performs a memory task ered that a rat’s knowledge could not while in a scanner, or, in the case of ex- be localized to any single region; it ap- perimental animals, when certain genes peared that all of the rats were some- are knocked out? At this level of inquiry, what impaired and yet somewhat com- psychology, neuroscience, and molecu- petent–although more extensive tissue lar biology closely interact. removal produced more serious memory Network-level research aims to strad- de½cit. dle the gap between the systems and the As improved experimental protocols neuronal levels. One challenge is to un- later showed, Lashley’s non-localization derstand how distinct local changes in conclusion was essentially correct. many different neurons yield a coherent There is no such thing as a dedicated global, system-level change and a task- memory organ in the brain; information suitable modi½cation of behavior. How is not stored on the ½ling cabinet model do diverse and far-flung changes in the at all, but distributed across neurons. brain underlie an improved golf swing A general understanding of what it or a better knowledge of quantum means for information to be distributed mechanics? over neurons in a network has emerged What kinds of experience-dependent from computer models. The basic idea modi½cations occur in the brain? From is that arti½cial neurons in a network, one day to the next, the neurons that col- by virtue of their connections to other lectively make me what I am undergo arti½cial neurons and of the variable many structural changes: new branches strengths of those connections, can pro- can sprout, existing branches can ex- duce a pattern that represents something tend, and new receptor sites for neuro- –such as a male face or a female face, or chemical signals can come into being. the face of Churchill. The connection On the other hand, pruning could de- strengths vary as the arti½cial network crease branches, and therewith decrease goes through a training phase, during the number of synaptic connections be- which it gets feedback about the adequa- tween neurons. Or the synapses on re- cy of its representations given its input. maining branches could be shut down But many details of how actual neural altogether. Or the whole cell might die, nets–as opposed to computer-simulated taking with it all the synapses it formerly ones–store and distribute information supported. Or, ½nally, in certain special have not yet been pinned down, and so regions, a whole new neuron might be computer models and neural experi- born and begin to establish synaptic ments are coevolving. connections in its region.

46 Dædalus Winter 2004 And that is not all. Repeated high rates more likely to spike. This joint input ac- How do neurons of synaptic ½ring (spiking) will deplete tivity creates a larger postsynaptic effect, know? the vesicles available triggering a cascade of events inside the for release, thus constituting a kind of neuron that strengthens the synapse. memory on the order of two to three This general arrangement allows for dis- seconds. The constituents of particular tinct but associated world events (e.g., neurons, the number of vesicles released blue flower and plenty of nectar) to be per spike, and the number of transmitter modeled by associated neuronal events. molecules contained in each vesicle, can change. And yet, somehow, my skills re- The nervous system enables animals to main much the same, and my autobio- make predictions.7 Unlike plants, ani- graphical memories remain intact, even mals can use past correlations between though my brain is never exactly the classes of events (e.g., between red cher- same from day to day, or even from min- ries and a satisfying taste) to judge the ute to minute. probability of future correlations. A cen- No ‘bandleader’ neurons exist to en- tral part of learning thus involves com- sure that diverse changes within neu- puting which speci½c properties predict rons and across neuronal populations the presence of which desirable effects. are properly orchestrated and collective- We correlate variable rewards with a fea- ly reflect the lessons of experience. Nev- ture to some degree of probability, so ertheless, several general assumptions good predictions will reflect both the guide research. For convenience, the expected value of the reward and the broad range of neuronal modi½ability probability of the reward’s occurring; can be condensed by referring simply to this is the expected utility. Humans and the modi½cation of synapses. The deci- bees alike, in the normal course of the sion to modify synapses can be made business of life, compute expected utili- either globally (broadcast widely) or ty, and some neuronal details are begin- locally (targeting speci½c synapses). If ning to emerge to explain how our made globally, then the signal for change brains do this. will be permissive, in effect saying, “You To the casual observer, bees seem to may change yourself now”–but not dic- visit flowers for nectar on a willy-nilly tating exactly where or by how much or basis. Closer observation, however, re- in what direction. If local, the decision veals that they forage methodically. Not will likely conform to a rule such as this: only do bees tend to remember which If distinct but simultaneous input signals individual flowers they have already vis- cause the receiving neuron to respond ited, but in a ½eld of mixed flowers with with a spike, then strengthen the con- varying amounts of nectar they also nection between the input neurons and learn to optimize their foraging strategy, the output neurons. On its own, a signal so that they get the most nectar for the from one presynaptic (sending) neuron least effort. is unlikely to cause the postsynaptic (re- Suppose you stock a small ½eld with ceiving) neuron to spike. But if two dis- two sets of plastic flowers–yellow and tinct presynaptic neurons–perhaps one blue–each with wells in the center into from the auditory system and one from which precise amounts of sucrose have the somatosensory system–connect to the same postsynaptic neuron at the 7 John Morgan Allman, Evolving Brains (New same time, then the receiving neuron is York: Scienti½c American Library, 1999).

Dædalus Winter 2004 47 Patricia been deposited.8 These flowers are ran- and ‘the goodies received this time.’9 Smith Churchland domly distributed around the enclosed Vum’s output is the release of a neuro- on ½eld and then baited with measured vol- modulator that targets a variety of cells, learning umes of ‘nectar’: all blue flowers have including those responsible for action two milliliters; one-third of the yellow selection. If that neuromodulator also flowers have six milliliters, two-thirds acts on the synapses connecting the sen- have none. This sucrose distribution sory neurons to vum, then the synapses ensures that the mean value of visiting will get stronger, depending on whether a population of blue flowers is the same the vum calculates ‘worse than expect- as that of visiting the yellow flowers, ed’ (less neuromodulator) or ‘better though the yellow flowers are more than expected’ (more neuromodulator). uncertain than the blues. Assuming that the Montague-Dayan After an initial random sampling of model is correct, then a surprisingly sim- the flowers, the bees quickly fall into a ple circuit, operating according to a fair- pattern of going to the blue flowers 85 ly simple weight-modi½cation algo- percent of the time. You can change rithm, underlies the bee’s adaptability their foraging pattern by raising the to foraging conditions. mean value of the yellow flowers–for Dependency relations between phe- example, by baiting one-third of them nomena can be very complex. In much with ten milliliters. The behavior of the of life, dependencies are conditional and bees displays a kind of trade-off between probabilistic: If I put a fresh worm on the reliability of the source type and the the hook, and if it is early afternoon, nectar volume of the source type, with then very probably I will catch a trout here. the bees showing a mild preference for As we learn more about the complexities reliability. What is interesting is this: of the world, we ‘upgrade’ our represen- depending on the reward pro½le taken tations of dependency relations;10 we in a sample of visits, the bees revise their learn, for example, that trout are more strategy. The bees appear to be calculat- likely to be caught when the water is ing expected utility. How do bees–mere cool, that shadowy pools are more bees–do this? promising ½sh havens than sunny pools, In the bee brain there is a neuron, and that talking to the worm, entreating though itself neither sensory nor motor, the trout, or wearing a ‘lucky’ hat makes that responds positively to reward. This no difference. Part of what we call intel- neuron, called VUMmx1 (‘vum’ for ligence in humans and other animals is short), projects very diffusely in the bee the capacity to acquire an increasingly brain, reaching both sensory and motor complex understanding of dependency regions, as it mediates reinforcement relations. This allows us to distinguish learning. Using an arti½cial neural net- work, Read Montague and Peter Dayan 9 See Read Montague and Peter Dayan, “Neu- robiological Modeling,” in William Bechtel, discovered that the activity of vum rep- George Graham, and D. A. Balota, eds., A Com- resents prediction error–that is, the dif- panion to Cognitive Science (Malden, Mass.: ference between ‘the goodies expected’ Blackwell, 1998).

10 Clark N. Glymour, The Mind’s Arrows (Cam- 8 This experiment was done by Leslie Real, bridge, Mass.: mit Press, 2001). See also Alison “Animal Choice Behavior and the Evolution of Gopnik, Andrew N. Meltzoff, and Patricia K. Cognitive Architecture,” Science (1991): 980– Kuhl, The Scientist in the Crib (New York: Wil- 986. liam Morrow & Co., 1999).

48 Dædalus Winter 2004 fortuitous correlations that are not gen- These discoveries open the door to How do neurons uinely predictive in the long run (e.g., understanding the neural organization know? breaking a tooth on Friday the thir- underlying prediction. They begin to teenth) from causal correlations that are forge the explanatory bridge between (e.g., breaking a tooth and chewing hard experience-dependent changes in single candy). This means that we can replace neurons and experience-dependent superstitious hypotheses with those that guidance of behavior. And they have be- pass empirical muster. gun to expose the neurobiology of addic- Like the bee, humans and other ani- tion. A complementary line of research, mals have a reward system that mediates meanwhile, is untangling the mecha- learning about how the world works. nisms for predicting what is nasty. Al- There are neurons in the mammalian though aversive learning depends upon a brain that, like vum, respond to re- different set of structures and networks ward.11 They shift their responsiveness than does reinforcement learning, here to a stimulus that predicts reward, or too the critical modi½cations happen indicates error if the reward is not forth- at the level of individual neurons, and coming. These neurons project from a these local modi½cations are coordinat- brainstem structure (the ventral tegmen- ed across neuronal populations and inte- tal area, or ‘vta’) to the frontal cortex, grated across time. and release dopamine onto the postsy- Within other areas of learning re- naptic neurons. The dopamine, only search, comparable explanatory threads one of the neurochemicals involved are beginning to tie together the many in the reward system, modulates the levels of nervous system organization. excitability of the target neurons to the This research has deepened our under- , thus setting up the standing of working memory (holding conditions for local learning of speci½c information at the ready during the ab- associations. sence of relevant stimuli) spatial learn- Reinforcing a behavior by increasing ing, autobiographical memory, motor pleasure and decreasing anxiety and skills, and logical inference. Granting works very ef½ciently. Nevertheless, the extraordinary research accomplish- such a system can be hijacked by plant- ments in the neuroscience of knowledge, derived molecules whose behavior mim- nevertheless it is vital to realize that ics the brain’s own reward system neu- these are still very early days for neuro- rochemicals. Changes in reward system science. Many surprises–and even a rev- pathways occur after administration of olution or two–are undoubtedly in cocaine, nicotine, or opiates, all of which store. bind to receptor sites on neurons and are similar to the brain’s own peptides. The Together, neuroscience, psychology, precise role in brain function of the large embryology, and molecular biology are number of brain peptides is one of neu- teaching us about ourselves as knowers– roscience’s continuing conundrums.12 about what it is to know, learn, remem- ber, and forget. But not all philosophers 11 See Paul W. Glimcher, Decisions, Uncertainty, mit embrace these developments as prog- and the Brain (Cambridge, Mass.: Press, ress.13 Some believe that what we call 2003). 13 I take it as a sign of the backwardness of aca- 12 I am grateful to Roger Guillemain for dis- demic philosophy that one of its most esteemed cussing this point with me. living practitioners, Jerry Fodor, is widely sup-

Dædalus Winter 2004 49 Patricia external reality is naught but an idea cre- their relative success in predicting and Smith Churchland ated in a nonphysical mind, a mind that explaining. on can be understood only through intro- But does all of this mean that there is a learning spection and reflection. To these philos- kind of fatal circularity in neuroscience ophers, developments in cognitive neu- –that the brain necessarily uses itself to roscience seem, at best, irrelevant. study itself? Not if you think about it. The element of truth in these philoso- The brain I study is seldom my own, but phers’ approach is their hunch that the that of other animals or humans, and I mind is not just a passive canvas on can reliably generalize to my own case. which reality paints. Indeed, we know Neuroepistemology involves many that brains are continually organizing, brains–correcting each other, testing structuring, extracting, and creating. As each other, and building models that can a central part of their predictive func- be rated as better or worse in character- tions, nervous systems are rigged to izing the neural world. make a coherent story of whatever input Is there anything left for the philoso- they get. ‘Coherencing,’ as I call it, pher to do? For the neurophilosopher, at sometimes entails seeing a fragment as least, questions abound: about the inte- a whole, or a contour where none exists; gration of distinct memory systems, the sometimes it involves predicting the im- nature of representation, the nature of minent of an object as yet reasoning and rationality, how informa- unperceived. As a result of learning, tion is used to make decisions, what brains come to recognize a stimulus as nervous systems interpret as informa- indicating the onset of meningitis in a tion, and so on. These are questions with child, or an eclipse of the Sun by the deep roots reaching back to the ancient Earth’s shadow. Such knowledge de- Greeks, with ramifying branches ex- pends upon stacks upon stacks of neural tending throughout the history and phi- networks. There is no apprehending the losophy of Western thought. They are nature of reality except via brains, and questions where experiment and theo- via the theories and artifacts that brains retical insight must jointly conspire, devise and interpret. where creativity in experimental design From this it does not follow, however, and creativity in theoretical speculation that reality is only a mind-created idea. It must egg each other on to unforeseen means, rather, that our brains have to discoveries.14 keep plugging along, trying to devise hypotheses that more accurately map 14 Many thanks to Ed McAmis and Paul Churchland for their ideas and revisions. the causal structure of reality. We build the next generation of theories upon the scaffolding–or the ruins–of the last. How do we know whether our hypothe- ses are increasingly adequate? Only by

ported for the following conviction: “If you want to know about the mind, study the mind –not the brain, and certainly not the genes” (Times Literary Supplement, 16 May 2003, 1–2). If philosophy is to have a future, it will have to do better than that.

50 Dædalus Winter 2004