Embodied knowledge vs. flesh and blood

As DiMaggio (1997) originally noted, most sociological theories of action make assumptions about the nature of cognition even as they dismiss any explicit discussion of cognition in favor of “social” explanation. Thinking about how culture comes to be taken up by the mechanisms of cognition and how it influences action through those mechanisms would, theoretically, address deficits in sociological theories of action and, at the same time, correct the bias towards extreme individualism that pervaded the cognitive sciences from the 1950s to the 1990s (which, as Dryfus (1992) has been screaming for his entire career, made them useful for writing chess-playing programs and little else). Persons, according to this view, are not mere symbol-processing machines, but culturally-informed symbol-processing machines, whose chaotic interaction with the myriad cultural forms of everyday life naturally produces both behavioral and cultural variation (DiMaggio, 1997: 272).

As new theory tends to do, these symbolic-schematic accounts of how action comes to be solved some problems and created a few more. In cognitive science, the symbol-processing model simply failed to manifest its promises in the fields of artificial intelligence and robotics. From the 1980s through the early 2000s, most programmers and engineers tried to mimic intelligent behavior by writing programs composed of internally consistent symbol systems. While this produced some laudable feats (one thinks of Deep Blue’s famous triumph over the then world chess champion Gary Kasparov), they were limited to extremely bounded tasks that lent themselves to abstraction. In contrast, physical tasks that nine-month-old babies did with ease were arduously recreated by robotics engineers only to fail as soon as the environment in which they were performed was slightly altered. This begged the question: if human intelligence is basically a complex symbol-processing mechanism, then why are artificial symbol-processing systems so unbelievably inept at tasks so simply any human could perform with without any amount of thought or attention?

In sociological theory, the symbol-processing model of culture and cognition painted a picture of an agent who, rather than simply responding to culture, could explore and engage with it. But the nature of the mechanism(s) that allowed for this remained opaque. In other words, if culture is internalized as cognitive architecture, what is the process of internalization? How are the cultural “logics,” “schemas,” and “heuristics” that, in interaction with the social world (or “stimuli” for the cognitive scientists) acquired and applied?

Embodiment in Social Theory

Enter the embodiment perspective. The turn towards embodiment, both within culture and cognition (Ignatow, 2007; Strand & Lizardo, 2015; Winchester, 2016) and, increasingly, within cognitive science itself (Edelman, 2004; Rowlands, 2011), has been an attempt to address these issues. In social theory, the embodiment perspective accounts for culture’s internalization by theorizing that the systems of thought that ground our ability to engage with the world – perception, the formation of habits, and the execution of habitual behavior – are essentially informed by the iterative interactions of the body with the world. For some thinkers, a capacity for “deliberation” is a feature of embodiment (Joas, 1996; Winchester, 2016), this capacity itself depends on the repertoire of habits that result from the body’s immersion in the world. Our capacity for action and the cognitive schemas and logics on which it depends finds its root in the body’s grounding in a stable world from which, through infinite experimental explorations from the first day of life until the day we die, it amasses “embodied knowledge.”

This theory of cognition has been extremely fruitful for cognitive scientists and robotics engineers. Robots fitted with exploratory learning algorithms have fared far better at problem-solving in various arenas compared to their symbol-processing predecessors (Edelman, 2004). In sociology, too, the conceptualization of knowledge as fundamentally embodied is enjoying somewhat of a heyday in sociological theory (e.g. Martin, 2011). And no wonder, since theories of embodied knowledge have several advantages over symbol-processing theories of cognition. For example, they provide an explanation of how cultural knowledge is acquired, maintained, and changed over time. In addition, they lend themselves to habit-oriented theories of action. And finally, they continually situate subjects within the world they inhabit, making a retreat into the theatre of the mind in order to “deliberate,” “calculate,” or “problem-solve” in a wholly abstract fashion analytically unnecessary. This feature of the embodiment perspective has been particularly attractive for action theorists interested in dismantling the legacy of the Cartesian model of the human subject (Crossley, 2013; Scheper-Hughes & Lock, 1987; Turner, 1984; Whitford, 2002), and for sociological theory more generally because it provides a detailed explanatory account of the inseparability of individual and society (Joas, 1996; Martin, 2011).

Beyond Representationalism

Nevertheless, despite the radical situatedness advanced by contemporary theories of embodiment in culture and cognition, a specter of their theoretical predecessors remains. Specifically, the theorization of embodied knowledge tends to conceptualize that knowledge not as a feature of the flesh and blood of the physical body in the world, but as a series of representations of bodily capacities developed and stored in the brain. Ignatow (2007: 122), for example, refers to a “repertoire of embodiments…stored in memory with cognition and language rather than in a separate location.” This makes sense intuitively. The brain, after all, is the ultimate site of the choreography of habitual behavior. We might speak of “muscle memory,” but the effortless sequencing of movements to which that phrase refers relies on patterned neuronal connections in the motor cortex. By themselves, the muscles that articulate activity know nothing of these connections. It is therefore often easy to ignore the physical body in favor of the cognitive representations that map the repertoire of habits it has access to.

But to do so is to mistake the choreography for the dancer. When we neglect the role that the flesh and blood of the physical body plays in the development and maintenance of habitual behavior, we describe embodiment only in its foundational capacity, its ability to give rise to the world immersion that characterizes experience in moments of habitual flow. Even in these moments, however, embodiment is continually vulnerable to breakdown. When we are ill or injured, for example, the cognitive infrastructure that encodes embodied knowledge can no longer make itself manifest. This aspect of embodiment – its vulnerability to disorientation and ungroundedness – is as much a feature of its nature as its ability to act as the bedrock of being-in-the-world.

This is an observation that Maurice Merleau-Ponty made more than half a century ago. Like contemporary theorists of culture and cognition, Merleau-Ponty (1962, p. 102) conceived of habit formation as “a rearrangement and renewal of the corporeal schema”; but he was also always careful to emphasize that the corporeal schema, or “habit-body”, was only intelligible when married to a corresponding “body at this moment.” The specific habit-creating character of human subjectivity, “always already” immersed in its world, relies fundamentally on the fact that the flesh and blood of the physical body (unlike its cognitive representation in the nervous system) extends into that world.

As such, the body is simultaneously an objective part of the world, on the one hand, and the foundation for subjective experience, on the other. This insight allows Merleau-Ponty to account both for the effortless enactment of habitual behavior that structures daily life and the ever-present possibility of a breakdown in the flow of experience it gives rise to: “The fusion of soul and body in the act, the sublimation of biological into personal existence, and of the natural into the cultural world is made both possible and precarious by the temporal structure of our existence” (Merleau-Ponty, 1962, p. 97, italics added).

Recognizing the possibility of breakdown as an essential element of embodiment is important for its conceptualization for two reasons. First, it is simply an accurate description of the reality of embodied experience: our habits are accessible and deployable only to the extent that we possess a body capable of enacting them. “Embodied knowledge” is not enough. Second, a recognition of the tenuousness of embodied knowledge opens up a novel space for theorizing how ruptures in the flow of existence produce behavioral variation. Like disjunctures between ideology and the material conditions of life (Swidler, 1986), or ruptures in the relationship between habitus and history (Bourdieu, 2004), breakdowns in the relationship between the physical body and the cognitive structures that map its history of activity give rise to opportunities for creative behaviour, as subjects are forced to contend with the experience of being “thrown” into an action that they are newly incapable of performing.

References

Bourdieu, P. (2004). The peasent and his body. Ethnography, 5(4), 579–599.

Crossley, N. (2013). Habit and habitus. Theory & Society, 19(2–3), 136–161.

Drefus, H. L. (1992). What computers still can’t do: A critique of artificial reason. Cambridge, MA: MIT Press.

Edelman, G. (2004). Wider than the sky. New York: Yale University Press.

Ignatow, G. (2007). Theories of embodied knowledge: New directions for cultural and cognitive sociology? Journal for the Theory of Social Behaviour, 37(2), 115–135.

Joas, H. (1996). The creativity of action. Chicago: University of Chicago Press.

Martin, J. L. (2011). The explanation of social action. New York: Oxford University Press.

Merleau-Ponty, M. (1962). Phenomenology of perception. New York: Routledge.

Rowlands, M. (2011). The new science of the mind: From extended mind to embodied phenomenology. Cambridge, MA: MIT Press.

Scheper-Hughes, N., & Lock, M. (1987). The mindful body: A prolegomenon to future work in medical anthropology. Medical Anthropology Quarterly, 1(1), 6–41.

Strand, M., & Lizardo, O. (2015). Beyond World Images: Belief as embodied action in the world. Sociological Theory, 33(1), 44–70.

Swidler, A. (1986). Culture in action: Symbols and strategies. American Sociological Review, 51, 273–286.

Turner, B. S. (1984). The body and society: Explorations in social theory. London: SAGE.

Whitford, J. (2002). Pragmatism and the untenable dualism of means and ends: Why rational choice theory does not deserve pragmatic privilege. Theory & Society, 31, 325–363.

Winchester, D. (2016). A hunger for god: Embodied metaphor as cultural cognition in action. Social Forces, 95(2), 585–606.

“Learning By Nodes”: Dendritic Learning and What It Means (Or Not) for Cultural Sociology

In a paper published earlier this year in Scientific Reports and further discussed in a later ACS Chemical Neuroscience article, a group of researchers argues that learning might not function like we previously thought. The researchers (Sardi et al. 2018a, 2018b) explain that the dominant conceptualization in cognitive neuroscience of how learning works—synaptic learning, or “Hebbian learning” (Hebb 1949)—is wrong. Instead, using a series of computational models and experiments with synaptic blockers and neuronal cultures  (see Sardi et al. 2018a:4-7), the authors find evidence for a different type of learning—what they refer to as “dendritic learning.” Just as “Copernicus was the first to articulate loudly that the earth revolves around the sun and not vice versa, even though all the accumulated astronomical evidence at that time fit the old postulation,” the researchers proclaim, as are they the first to “[swim] against conventional wisdom” of Hebbian learning theory (2018b:1231).

Of what consequence is this newfound process of dendritic learning for cultural sociology? Should we care at all? I’ll try to briefly describe some of the potential consequences of dendritic learning for cultural sociology; but, spoiler alert, I am not sure one way or the other if these consequences amount to being consequential for how we do sociology. But perhaps taking a peek at what dendritic learning is and how it is different from conventional understandings of how learning works is a nice place to start.

copernican-universe
Figure 1. Are We Witnessing a “Revolution of the Cognitive Spheres”?
Note: Image from Copernicus’ On the Revolutions of the Heavenly Spheres (Palca 2011).

LINKS VS. NODES

Going on 70 years, the prevailing explanation for how learning works has been synaptic learning. Building from Hebb’s (1949) The Organization of Behavior, the idea behind synaptic learning is that if there is an activity that stimulates a neuron which in turn stimulates another neuron, and if that activity is repeated over time, then the first neuron becomes a more efficient stimulator of the second neuron and the two become more strongly connected in the brain.

Neuron-neuron stimulation occurs through synapses, the chemical (usually) or electrical (less frequently) structural gaps between neurons transmitting information across them. Synaptic learning, then, is a type of “activity-dependent synaptic plasticity” (Choe 2015:1305). Repeated practices or exposures to a certain stimulus modifies the synaptic strength between the two neurons: when the practice/exposure is repeated, the two neurons become more tightly associated in the brain, and when the practice/exposure is not repeated, the association weakens. This process occurs relatively slowly.

Synaptic learning is the inspiration behind the old adage that “neurons that fire together wire together.” Until very recently, this was the way we assumed new neural coalitions formed in biological neural networks. Consider an example from Luke Muehlhauser over on the Less Wrong blog (Muehlhauser 2011). Think back to Pavlov’s experiment on classical conditioning (Pavlov 1910):  a dog is given food when the researcher rings a bell, and the timing between the bell ringing and the presentation of food is manipulated. At first, there is no association between the neurons stimulated by bell ringing and the neurons that trigger salivation; they are, ostensibly, mutually exclusive actions. However, if the researcher rings the bell and the food is presented to the dog at the same time (or in close enough time intervals), the neurons that fire when food is present and the neurons that fire with bell ringing are activated together. Over repeated trials, the synapses between “bell ringing” and “salivation” neurons become stronger and, eventually, simply ringing the bell induces salivation without the presentation of food (see Figure 2).

Screen Shot 2018-10-16 at 6.56.46 PM
Figure 2. Synaptic Learning with Pavlov’s Experiment
Note: Reprinted from Less Wrong blog (Muehlhauser 2011).

Sardi and colleagues refer to synaptic learning as “learning by links” (Sardi et al. 2018a:1), since learning occurs through the synapses that link the neurons together. Their research, however, suggests a different type of learning—dendritic learning, also known as “learning by nodes” (Sardi et al. 2018a:2). In short, with this mode of learning, the workhorse of the neuron for learning purposes is not the synapses, but instead the dendrites. In a neuron cell, dendrites are the long, treelike extensions that connect the cell body (the soma, which contains the cell nucleus) to the synapses that themselves “connect” the neuron to other neurons.

Take a look at Figure 3, a neuron cell’s anatomy. The dendrites are responsible for taking in information from other neurons and passing it along into the soma, while the axon is responsible for passing the information on to other neurons via the axon terminals—which are themselves connected to the next neuron’s dendrites through synapses, thus propagating information transmission across the neural network. Without dendrites, information cannot be transmitted into the body of the neuron: e.g., damaged or abnormal dendrites are linked to brain under-connectivity issues associated with autism (Martínez-Cerdeño, Maezawa, and Jin 2016). Trying to construct new neural networks without dendrites is like trying to have group deliberation with all talk and no listening.

Screen Shot 2018-10-16 at 8.44.31 PM
Figure 3. A Neuron’s Anatomy
Note: Reprinted from OpenStax (2018), redirected from Khan Academy (2018).

So, how does dendritic learning differ functionally from synaptic learning? While synaptic learning is based on the idea of synaptic plasticity, dendritic learning revolves around the notion of (you guessed it) a sort of dendritic plasticity: given increasing or decreasing levels of exposure to a neuron-activating stimulus, the extent of the neuron’s “dendritic excitability” can grow or diminish while the strength of the synapses remain relatively constant (Neuroskeptic 2018).

Consider Figure 4. Across both panels, the teardrop object at the bottom represents the neuron cell body, which is where the firing happens if the input signals from the dendrites are strong enough for an outgoing signal to be pushed from the cell body down through the axon and into the dendrites of the next neuron. The long treelike branches are the dendrites, and the tips are the synapses that connect the neuron’s dendrites to the axon terminals of other (not shown) neurons. The left panel illustrates conventional synaptic learning, where the synapses themselves are weighted (indicated by the red valves at the tips of the branches) upward or downward depending on the extent of stimulus exposure. The right panel shows dendritic learning: it is the extent to which a neuron’s dendrites are in a high state of stimulation, and not the strength of the synapses linking the neuron to other neurons, that determines the strength of the input signal and therefore whether or not the neuron fires. In dendritic learning, then, there are far fewer “learning parameters,” since the dendrites are responsible for the learning and not the synapses (see the right panel of Figure 4) (ScienceDaily 2018).

Screen Shot 2018-10-16 at 9.34.34 PM
Figure 4. Synaptic Learning (left) vs. Dendritic Learning (right)
Note: Reprinted from ScienceDaily (2018).

IMPLICATIONS (?) FOR CULTURAL SOCIOLOGY

The “Neuroskeptic” over at Discover Magazine reviewed the evidence from the Sardi et al. papers and suggests that “[a]t best they have shown that dendritic learning also happens [in addition to synaptic learning],” and that “[they] don’t think Copernicus has returned to earth just yet” (Neuroskeptic 2018). I agree with Neuroskeptic in terms of what this means for neuroscience, largely because they are the neuroscientist and I am not. That said, there does seem to be the potential for some implications for how we do cultural sociology. But the potential may be greater for some subfields than for others.

I’m Not Sure What this Adds for How Sociologists Study Learning

The existence of dendritic learning has at least two major implications for cognitive neuroscience. First, learning may happen at much faster timescales than previously thought. Second, weak synapses matter a lot. In terms of timescale, it seems that the brain isn’t that bad at quick adaptation—at least relative to traditional Hebbian learning. As Sardi and colleagues note, “[t]his dynamic brain activity leads to the capability that when we think about an issue several times we may find different solutions” (Shrourou 2018). For the importance of weak synapses, the researchers point out that dendritic strengths are “self-oscillating” (2018b:1231), where weak synapses effectively “temper” the dendritic weights and prevent them from taking on extreme values. In other words, “dendritic learning enables stabilization around intermediate [dendritic strength] values” (Sardi et al. 2018a:4). These implications are pretty important for neuroscientists and medical researchers studying various diseases of the brain (Sardi et al. 2018b:1231-32).

What does all this mean for cultural sociologists? It might be too early to tell. Dendritic learning might be faster than synaptic learning, but the time scales in the experiments are in much smaller intervals (minutes) than the learning processes of interest to sociologists. The researchers note that future studies should “investigate . . . [dendritic learning] efficiency and available learning time scales in more realistic scenarios” (2018b:1231), so it’s an empirical question whether or not the learning speed differentials between synaptic and dendritic learning are a wash with longer timescales. So, in terms of theoretical leverage, dendritic learning may or may not offer much over and above how we already talk about learning in culture and cognition studies (see Lizardo et al. 2016:293-95). At the end of the day, for cultural sociologists it may all look like GOFILT—Good Old Fashioned Implicit Learning Theory—in which case the difference between synaptic and dendritic learning can be taken as ontologically true but analytically inconsequential. Only time (pun) will tell.

The Payoff May Come Sooner for Computational Social Science

In addition to understanding the learning processes behind biological neural networks and brain disorders, Sardi and colleagues also note that this “paradigm shift” matters for developing machine learning algorithms built to mimic human learning (2018b:1231). In natural language processing, for instance, if synaptic learning isn’t the baseline model of human learning (itself an empirical question), then perhaps analytical strategies that build associations between terms or documents based on term frequencies and co-occurrences aren’t based on the best cognitive model for machine learning.

But at face value I’m skeptical of this last proposition—I like word count methods for analyzing meaning, others do too (Nelson 2014; Underwood 2013), and I’ve read enough papers that make defensible claims using them to sell me on their continued use. That said, we have not seen dendritic learning rules implemented into machine learning algorithms yet (but see Sardi et al. 2018a:2-3 for an example of dendritic learning rules in a series of perceptron models), and it might prove particularly consequential in deep learning tasks and artificial neural network models. These sort of machine learning algorithms have not gained much traction in sociology, though, so, for now, it seems that the utility of distinguishing between synaptic and dendritic learning for culture and cognition studies is truly a waiting game.

I can continue all of my work without making these distinctions, and I suspect that most of the people reading this post are in the same position.

REFERENCES

Choe, Yoonsuck. 2015. “Hebbian Learning.” Pp. 1305-09 in Encyclopedia of Computational Neuroscience, edited by D. Jaeger and R. Jung. New York: Springer.

Hebb, Donald O. 1949. The Organization of Behavior: An Neuropsychological Theory. New York: Wiley.

Khan Academy. 2018. “Overview of Neuron Structure and Function.” Khan Academy. Retrieved October 16, 2018 (https://www.khanacademy.org/science/biology/human-biology/neuron-nervous-system/a/overview-of-neuron-structure-and-function).

Lizardo, Omar, Robert Mowry, Brandon Sepulvado, Dustin S. Stoltz, Marshall A. Taylor, Justin Van Ness, and Michael Wood. 2016. “What Are Dual Process Models? Implications for Cultural Analysis in Sociology.” Sociological Theory 34(4):287-310.

Martínez-Cerdeño, Verónica, Izumi Maezawa, and Lee-Way Jin. 2016. “Dendrites in Autism Spectrum Disorders.” Pp. 525-43 in Dendrites: Development and Disease, edited by K. Emoto, R. Wong, E. Huang, and C. Hoogenraad. Tokyo: Springer.

Muehlhauser, Luke. 2011. “A Crash Course in the Neuroscience of Human Motivation.” Less Wrong. Retrieved October 16, 2018 (https://www.lesswrong.com/posts/hN2aRnu798yas5b2k/a-crash-course-in-the-neuroscience-of-human-motivation).

Nelson, Laura K. 2014. “Computer-Assisted Content Analysis and Sociology: What You Should Know.” Bad Hessian. Retrieved October 17, 2018 (http://badhessian.org/2014/01/computer-assisted-content-analysis-and-sociology-what-you-should-know/).

Neuroskeptic. 2018. “Is ‘Dendritic Learning’ How the Brain Works?” Discover Magazine. Retrieved October 16, 2018 (http://blogs.discovermagazine.com/neuroskeptic/2018/05/11/dendritic-learning/#.W8aX4P5KjdT).

OpenStax. 2018. “Neurons and Glial Cells.” OpenStax CNX. Retrieved October 16, 2018 (https://cnx.org/contents/GFy_h8cu@9.87:c9j4p0aj@3/Neurons-and-Glial-Cells).

Palca, Joe. 2011. “For Copernicus, A ‘Perfect Heaven’ Put Sun At Center.” NPR: Morning Edition. Retrieved October 16, 2018 (https://www.npr.org/2011/11/08/141931239/for-copernicus-a-perfect-heaven-put-sun-at-center).

Pavlov, Ivan. 1910. The Work of the Digestive Glands. London: C. Griffin & Company.

Sardi, Shira, Roni Vardi, Amir Goldental, Anton Sheinin, Herut Uzan, and Ido Kanter. 2018a. “Adaptive Nodes Enrich Nonlinear Cooperative Learning Beyond Traditional Adaptation By Links.” Scientific Reports 8(1):5100.

Sardi, Shira, Roni Vardi, Amir Goldental, Yael Tugendhaft, Herut Uzan, and Ido Kanter. 2018b. “Dendritic Learning as a Paradigm Shift in Brain Learning.” ACS Chemical Neuroscience 9:1230-32.

ScienceDaily. 2018. “The Brain Learns Completely Differently than We’ve Assumed Since the 20th Century.” ScienceDaily. Retrieved October 16, 2018 (https://www.sciencedaily.com/releases/2018/03/180323084818.htm).

Shrourou, Alina. 2018. “Dendritic Learning Occurs Much Faster and In Closer Proximity to Neurons, Shows Study.” News Medical: Life Sciences. Retrieved October 16, 2018 (https://www.news-medical.net/news/20180830/Dendritic-learning-occurs-much-faster-and-in-closer-proximity-to-neurons-shows-study.aspx).

Underwood, Ted. 2013. “Wordcounts Are Amazing.” The Stone and the Shell. Retrieved October 17, 2018 (https://tedunderwood.com/2013/02/20/wordcounts-are-amazing/).

Limits of innateness: Are we born to see faces?

Sociologists tend to be skeptical of claims individuals are consistent across situations, as a recent exchange on Twitter exemplifies. This exchange was partially spurred by revelations that the famous Stanford Prison Experiment (which supposedly showed people will quickly engage in behaviors commensurate with their assigned roles even if it means …

Beyond Good Old-Fashioned Ideology Theory, Part Two

In part one, I examined two recent frameworks for understanding ideology (Jost and Martin) and explained how both serve as alternatives to the good old-fashioned ideology theory (GOFIT). Ultimately, I concluded that Martin’s (2015) model has specific advantages over Jost’s (2006) model, though the connection between ideology and “practical mastery …

Durkheimian Sociology and its Discontents, Part II: Why Culture, Social Psychology, & Emotions Matter to Suicide

In a previous post, I argued that despite its importance and “classical” status, sociologists have not contributed to the study of suicide as much as they could. While Anna Mueller and I have yet to posit a general or formal theoretical statement on suicide, in this post, I attempt to …

Durkheimian Sociology and its Discontents: Why its Time for a New Sociology of Suicide

Since Durkheim showed that certain social structural factors, external to the individual, had a strong positive relationship to variation in suicide rates, sociologists have maintained the argument that suicide is caused by social forces and, therefore, is a phenomenon squarely in the domain of sociology. Yet, western medical professionals (Marsh …

Culture, Cognition and “Socialization”

Culture and cognition studies in sociology are mainly concerned with the construction,  transmission, and transformation of shared stocks of knowledge. This was clear in the classical theoretical foundations of contemporary work in the sociology of culture laid out in Parsons’s middle period functionalism (Parsons 1951) and in Berger and Luckmann’s …

What’s Cultural About Analogical Mapping?

Analogical mapping is a cognitive process whereby a particular target is understood by analogizing from a particular source. For example, Lakoff and Johnson (1999) have observed that people often reason about love metaphorically as a journey. In a previous post I discussed some experimental evidence supporting the claim that activating …