Wednesday, November 04, 2015

Do Neurons Literally Have Preferences?

Carrie Figdor has been arguing that they do.

Consider these sentences, drawn from influential works of neuroscience (quoted in Figdor forthcoming, p. 2):

  • A resonator neuron prefers inputs having frequencies that resonate with the frequency of its subthreshold oscillations (Izhikevich 2007).
  • In preferring a slit specific in width and orientation this cell [with a complex receptive field] resembled certain cells with simple fields (Hubel and Wiesel 1962, p. 115).
  • It is the response properties of the last class of units [of cells recorded via electrodes implanted in a rat’s dorsal hippocampus] which has led us to postulate that the rat’s hippocampus functions as a spatial map. ... These 8 units then appear to have preferred spatial orientations (O’Keefe and Dostrovsky 1971, p. 172).
  • These are completely standard, unremarkable claims of the type that neuroscientists have been making for decades. Figdor suggests that it's best to interpret these claims as literal truths. The verbs in these sentences work like many other verbs do -- "twist", "crawl", "touch" -- with literal usage across a wide range of domains, including organic and inorganic, part and whole.

    Figdor's view sounds bizarre, perhaps. People literally have preferences. And rats. Maybe frogs. Not trees (despite 22,000 Google hits for "trees prefer", such as "Ash trees prefer moist, well-draining soil for optimum growth"). Definitely not neurons, most people would say.

    One natural way to object to Figdor's view is to suggest that the language of neurons "preferring" is metaphorical rather than literal. I can see how that might be an attractive first thought. Another possibility worth considering is that maybe there are two senses of "prefer" at work -- a high-grade one for human beings, a thin one for neurons.

    Figdor responds to these objections, in part, with technical linguistic arguments that I am insufficiently schooled in linguistics to evaluate. Does conjoining human and neuronal cases of "prefers" pass the zeugma test?

    However, from seeing others' reactions to Figdor -- she gave a talk here at UCR a couple weeks ago -- I'd say it's not a fine sense of technical linguistics that drives most people's rejection of Figdor's claim. (In conversation, she says agrees with me about this; and in newer work in progress she is de-emphasizing the technical linguistic aspects to focus on the bigger picture, including how terms evolve over time in deference to scientific usage.) What gives folks the heebie-jeebies is the thought that "preferring" is a psychological notion, and so if Figdor is saying that neurons literally have preferences, she appears to be saying that neurons literally have minds or psychological states. And we certainly don't want to say that! (Do we?)

    Figdor is not some far-out panpsychist who believes that neurons tingle with experiences of delight when they receive the stimuli they prefer. But she is far out in another way -- a more sensible and appealing way, perhaps. Once we see the actual source of her radicalism, we can start to appreciate the importance and appeal of her work.

    It's natural -- common sense -- for us to approach the world by dividing it into things with minds (you, me, other people, dogs, birds...) and things without minds (stones, trees, pencils, fingernails). Reflecting on intermediate cases, such as various types of worms, one might sense trouble for a sharp distinction here, but vagueness along a single spectrum of mindedness isn't too threatening to common sense. The essential difference between the minded and the un-minded remains, despite a gray zone.

    Figdor's picture challenges all that. If what she says about "prefer" also goes for some other important psychological terms (as she thinks it does), then mentality spreads wide into the world. Some psychological terms -- "prefer", "decide", and "habituate" are her examples, to which I might add "seek", "learn", "reject", and many others -- appear to spread wide; while other terms, such as "meditate", "confess", and "appreciate", might apply only to humans (or maybe a few other species). Each psychological term has a range of application, and the terms that are more liberally applicable will attach to all sorts of systems that we might not otherwise tend to regard as privileged with any sort of mentality.

    Figdor has taken, I think, a crucial step toward jettisoning the remnants of the traditional dualist view of us as imbued with special immaterial souls -- toward instead seeing ourselves as only complex material patterns whose kin are other complex patterns, whether those patterns appear in other mammals, or in coral, or inside our organs, or in social groups or ecosystems or swirling eddies. Some complexities we share and others we do not. That is the radical lesson of materialism, which we do not fully grasp if we insist on saying "here are the minds and here are the non-minds", demanding a separate set of verbs for each, with truly "mental" processes only occurring in certain privileged spaces.

    With that thought in mind, let's go back to "prefer". Do neurons literally prefer? I don't know whether the linguistic evidence will ultimately support Figdor on this particular case, but I think we can approach it evenhandedly, letting fall wherever they may the technical tests of metaphor and polysemy and other considerations from linguistics and philosophy of language -- figuring that of course some of our mental state verbs literally refer to patterns of behavior that spread widely, and at different spatiotemporal grain, across the complex, multi-layered, dynamically evolving structures of our world.

    [image source]

    --------------------------------------

    Carrie writes:

    Thanks to Eric for posting on my work-in-progress and the opportunity to clarify a few things. First, the technical linguistic stuff is actually my attempt to understand why it could possibly strike anyone as "natural" or "reasonable" to think these uses are metaphorical. Who "naturally" thought Hubel and Wiesel intended their descriptions of their data to be metaphorical? To the contrary, the cry of "Metaphor!" reflects not an astute semantic analysis of their uses but an automatic response to my claim that they should be interpreted literally: "They just can’t possibly be literal." The idea that they are metaphorical is actually one of the weakest semantic alternatives to a literal view.

    That said, Eric is correct that I am not a radical panpsychist. Rather, I’m interested in a plausible, non-ad hoc explanation of the ever-expanding uses of psychological language throughout biology at all levels of complexity. Basically, I think psychological concepts are transitioning to scientifically determined standards for proper use, leaving behind the ideal-rational-human, anthropocentric standards we now have. There’s a lot more to that story, and I hope to make it public very soon.

    17 comments:

    Brandon N. Towl said...

    Interesting thoughts; I'm now motivated to read Figdor's paper!

    Interestingly, my reaction was similar in spirit but not in content. I feel there is an issue with the claim that neurons literally have preferences. But I don't think it stemmed from any kind of dualism or need to keep psychological terms applied to only a narrow range of things. No, I think my issue stems from the fact that "preferring" is a rather complex state, that plays a role in a complex mental economy. Knowing that a person prefers X, we can make inferences about what they will think and do-- but only if we also know that they believe Y, desire Z, etc. etc. I don't see the orientation preferences for V1 neurons being of the same kind. The neurons are too simple. Even for my dog, a sentence like "Toby prefers hamburger, but he will choose to eat dog food and not steal meat from the counter, because he believes he will receive a correction if he is caught..." makes complete sense and seams to be well within the psychology of this creature. I can't see a sentence like that working for a neuron.

    Or maybe the issue is that is could work, but why bother? It's taking an intentional stance towards something for which the design stance is sufficient, to borrow Dennet's terms.

    Unknown said...

    Does '...transitioning...' require transformationing as well..just look at neurological evolution; because of time we humans wonder about preference...Is time-evolution's-time literal or metaphorical...

    Joshua Rust said...

    Interesting! I'm reminded of a book, Embodied Minds in Action, by Robert Hanna and Michelle Maiese, where they single out a class of phenomena--what they call "natural causal singularities" (NCSs)--which are neither completely deterministic nor completely indeterministic:

    "[E]xamples of natural causal singularities include the Big Bang and black holes, the roiling movements of boiling water (as opposed to water boiling at 100 degrees centigrade/212 degrees Fahrenheit at standard pressure, which is deterministic), weather systems, traffic systems, ecosystems, planets like the Earth, solar systems, stars, star systems, and the biological processes and endogenously produced overt movements of living organisms, including the intentional body movements of conscious, intentional animals and real persons. Indeed, General Relativity predicts the existence of natural causal singularities such as the Big Bang and black holes.10 But even more importantly from our point of view, the biological processes and overt body movements of individual living organisms constitute a class of naturally creative little bangs with the same essential properties as the dramatically larger natural causal singularities.

    "What are the essential properties of natural causal singularities, whether large or little? Natural causal singularities are nomologically unique, actual world dependent, unprecedented, unrepeatable, situated, forward flowing, non-random processes with thermodynamic self-organization, existing in a natural world that also has some rough-grained general deterministic laws and some rough-grained general probabilistic or statistical laws. This sets natural singularities sharply apart from both completely deterministic events and completely indeterministic events." (261)

    Neurons are NCSs. Perhaps, we might include "haveing preferences" among any NCS's essential properties--a property that might even follow from their being self-organized. This isn't panpsychism, in that not every physical phenomenon/event is self-organizing, and that while being a NCS is perhaps necessary for consciousness, there is no requirement that being a NCS is sufficient for consciousness (though Hanna and Maiese happen to think that NCSs imply the "internal standpoint of a naturally purposive dynamic system" (120)).

    bryce huebner said...

    Very cool! Can't wait to read the new stuff; Carrie and I talked about this project when it was just getting started, and I'd forgotten all about it.

    I'm in the process of writing a tiny bit of a paper right now about bacterial behavior. It's in the context of a broader discussion of group behavior, and I realized that it was little more than bad ideology that leads philosophers to focus on complex prokaryotes in thinking about adaptive forms of group behavior. One of the things that keeps striking me as I read through the relevant literature is that the behavior of these tiny machines is really complex and highly adaptive. Their preferences might be simple, and they're probably grounded in little more than an impulse to pursue nutrients and avoid toxins, but bacteria seem to pass the William James test for mentality (fixed ends, with varying means). This is leading me to think that the whole idea of mindedness is poorly conceived, or that bacteria are minded...or maybe that's just my Spinozist proclivities showing. Regardless, given what I'm learning about bacteria, I'd be shocked if neurons didn't have preferences! So I'm really looking forward to reading Carrie's new stuff!!

    Eric Schwitzgebel said...

    Thanks for the comments, folks!

    Brandon: One possibility would be to analogize with "crawling". You might think crawling requires going slowly on legs, but arguably worms arguably literally crawl; if so, there's some commonality that a thinner thing that "crawling" gets at in both cases. You *might* be able to say the same about "preferring" -- that there's some thinner thing that both your dog on the neuron have in common. Or maybe it doesn't work for "preferring" in particular. I'm not committed specifically to that (though Carrie is). But the general idea of thinking that psychological verbs can have wide application rather than only applying literally to some privileged cases -- I find that interesting and appealing.

    Unknown 10:48: A bit cryptic for me to have much of a reaction to. Please feel free to expand with a clarification.

    Josh: Interesting analogy. It does have a similar flavor of seeing causal commonality across a wide range of cases that aren't typically classed together in folk psychology and folk physics. I struggle a bit with the idea of being "nomologically unique". I kind of want to go Cartwright, Dupre, or Horst here and suggest that there are a variety of possible nomological regularities that could encompass complex systems, without one one privileged nomological system, so that at a fine enough grain pretty much everything is unique and at a coarse enough grain nothing is.

    Bryce: That sounds pretty interesting. My hunch is that Carrie would suggest that the complexity is multi-level so that even saying "probably grounded in little more than an impulse to pursue nutrients and avoid toxins" is too oversimplifying and concessive to the picture that favors systems that look like mammals.

    Wayne said...

    Eric, great that you're discussing Carrie's work here.

    I havent talked to Carrie recently about this, so I've likely forgotten the nuances in her very interesting view, but on "preferences" in neurons...why not just understand that in terms of tuning, and that idea in terms of generation of firing rates for stimuli with specific parameters of interest (e.g. direction of motion as in the tuning of neurons in MT). Thus, one might look up direction tuning in MT neurons.

    It's not the psychological meaning of the word that is at issue but a specific technical meaning tied to the ideas just noted. I don't think the neuroscientists around me would have taken themselves to be implying anything psychological at all in their use of "preference" and they would drop the term and replace it with non-psychological terms if this were to become an issue in a given context, say here. They'd just show you a graph that relates one physical parameter to another.

    Here's another point: Preference and tuning go hand in hand with information in the statistical sense which is, following Dretske, not sufficient for representation. One might think that "preference" in the psychological sense goes hand in hand with representation.

    I don't think this invalidates Carrie's point. Rather, it's an emphasis on treating certain terms as technical and dissociable from immediate and essential connections to mentality.

    Unknown said...

    Eric, Josh, Carrie...
    Does 'psychological language' prefer behavior over nomology and panpsychism...
    Does philosophical thought require nomenclature that is philosophical in nature...
    Is the pull to metaphysics all ways calling philosophers to use the mind to find will...

    Lee J Rickard said...

    Superb stuff: a deeply interesting paper and an excellent analysis. This is why I subscribe to your feed!

    Callan S. said...

    I'm thinking much the same as Wayne - in terms of evolutionary science, no one treats such a scientist using the word 'design' to refer to some sort of god entity. It's just a very useful shorthand to say this creature is 'designed' for flying, that creature is 'designed' for swimming.

    Here 'preference' is being used as a shorthand. And I think Wayne is right - if neuroscientists realised they were being taken for implying some particular quality that they don't mean by 'preference', they'd drop the term and use one that avoids conveying the quality they have no interesting in conveying.

    In the end one could say a weighted die has a preference for rolling a six. But it's a shorthand for various physical effects.

    Carrie Figdor said...

    Hi to everyone (some of whom I know and have discussed some of this with before): I won't hijack Eric's blog but I did want to say I've taken note of the comments (in a couple of cases I discuss and address them in chapters of the MS) and it's good to be reminded of how persistent they can be. The main question to think about is what determines the standards of proper use (and from that, why we should assume that our untutored intuitions have got them right). There's been some great recent work on bacteria (Auletta, J.A. Shapiro etc.) and Fred Keijzer's recent paper on the Sphex wasp that show how we create an intuitive chasm (we: complex; they: simple; we: unpredictable; they: hardwired; etc.) where there may not be one. That's compatible with differences within a category (e,g, not all real decisions are weighty) even for us.

    David Duffy said...

    Bunge suggests that it is reasonable to talk about value of X to any form of life (in that access to X improves fitness) even if an organism doesn't necessarily have behaviour that maximizes X. So It seems to me that "a plant prefers X", as opposed to "a bacterium prefers (and seeks out) X" are different usages. It doesn't help me with the neuronal preference, except perhaps as per a neuronal Darwinism sense: a neuron that doesn't exhibit a useful preference for features of the world might lose connectivity. I don't think anyone actually thinks this way ;)

    Unknown said...

    Duffy, Carrie...
    What do plants, animals, humans have in common, well we are (I am) here...
    Could untutored intuitions and connectivity remind us to also Observe ourselves...
    Without understanding-seeing ourselves live life-it then can become simply a waste of time...

    chinaphil said...

    My immediate thought was along the same line as Bryce's: that in order to intelligibly talk about preferences, we would have to have at least two levels of state/activity. One level would be what the unit is doing/is happening to it, and on the other level would be some kind of assessment of this event. There would probably also have to be multi-input analysis so that the higher level can make something analogous to a "choice".

    But I think any analysis of the meaning of words like prefer should acknowledge the role of ignorance in them. We don't say balls prefer to fall to earth because we (believe we) have absolute knowledge of what they will do. If you watch a bacterium dance around responding to unseen stimuli, you would be likely to ascribe "preferences" or behaviour to it. If the stimuli are illuminated so that every movement the bacterium makes becomes obviously "caused", then people would be less likely to use psychological language to describe the bacterium's actions. In that sense, talking about neuron preferences makes sense if we are talking about a few known factors among a host of unknowns. If we basically know all the things that make a neuron tick, then I think we'll find it odd to use psychological terms.

    (Incidentally, this is another reason why talking to AI is going to be so hard - they're not going to want/need to use psychological terms to describe our behaviour, because much of it will be so transparent to them.)

    David Duffy said...

    (Incidentally, this is another reason why talking to AI is going to be so hard - they're not going to want/need to use psychological terms to describe our behaviour, because much of it will be so transparent to them.)

    This is Lem's joke in Imaginary Magnitude - that AIs can actually see each other's consciousness, but they are skeptical about ours'.

    Eric Schwitzgebel said...

    Yeah, chinaphil, that might be some of the folk psychology of it.

    David -- I started Imaginary Magnitude but got distracted by other things. I'm inspired to pick it up again!

    Scott Bakker said...

    Everyone I turn it seems I see shades of blind brain theory, anymore! I applaud the turn to heuristics that so many seem to be making, but with heuristics comes *neglect,* and Figdor's argument looks splendid only so long as one neglects neglect.

    I'm with Wimsatt in that I think beggars can't be choosers, that science is as science does, and that there will always be a wide array of tools used. But the flip side of affirming this pluralism has to be the recognition of the specificity of the tools used, as well as the fact that we have no way, at the theoretical level at least, of knowing when our applications become *misapplications,* impediments rather than implements.

    The fact is, the toolbox is itself an empirical domain of inquiry. Until things get sorted, it seems to me that we should be exceedingly wary of tools regularly found in apparently insoluble problem ecologies. And it seems pretty safe to say that 'psychological tools' have a very problematic track record, and should be avoided wherever they can be--even as facon-de-parlers.

    The fact is, there is a very stark difference between knowing what makes something tick and knowing what something wants. Until we know what this difference consists in (and I think I do!) Figdor simply has no way delimiting potential 'crash spaces.'

    Saulius Simcikas said...

    It's strange that example of neurons was chosen, not something simpler and more fundamental like "The octet rule refers to the tendency of atoms to prefer to have eight electrons in the valence shell."