ARTIFICIAL INTELLIGENCE IS AMERICAN
As am I
But somehow the content/process division of labor between anthropology and the rest of cognitive science became a barrier that isolated anthropology.1
Researchers who spend their time interpreting neural networks talk like mathematicians, engineers, and neuroscientists. They talk about vector spaces and manifolds, filters and circuits, neurons and activations. I enjoy this research; it is obviously a public good. But they speak an interdisciplinary language passed down from the previous century’s cognitive scientists and cyberneticians. Why? Where are the sociological and anthropological approaches to understanding neural networks? And what would those approaches even look like?
Let me remind my reader that cognitive science, like any subject, has a history. Its predecessor subject, call it “psychology,” was a German immigrant that had to assimilate to the 20th century American context. Here we don’t much care for Wundtian Völkerpsychologie (folk psychology) or for psychology-as-Geisteswissenschaften (human science) or for seeing how the Geist (Spirit) is vergegenständlicht (objectified) in our laws and customs and social institutions.2 In short, psychology was forced to naturalize. Ironically, our intolerance for these old German pseudo-ideas means we Americans tend to have difficulty seeing how our pragmatic spirit animates our professionalized and scientific approach to the study of behavior. This is why cognitive scientists will get together and scratch their heads about why they can’t seem to bring themselves to care about anthropology, not seeing that not giving a shit about culture is actually one of their national and academic cultural inheritances. Be proud! We’re Americans! Our intellectuals are anti-intellectual and our psychology revolves around IQ tests and personality questionnaires and “scientific social reform” and the self-help industry. And to the pretentious and muddleheaded snail-/kraut-eating “continental philosopher” who criticizes us for being crude or uneducated we say: SPEAK ENGLISH YOU FOREIGNER!
This natural-scientific attitude has also made it difficult for psychologists to internalize the fact that psychology experiments are social arrangements which would not work without student deference. Psychology professors have an interest in keeping quiet about the professor-student authority relationship and in particular how it plays a role in “improving data quality,” i.e., in soliciting student compliance. The Milgram shock experiment is taught in Psych 101 with the takeaway being that people defer to a labcoat. The professor has a responsibility, I think, to emphasize that the students themselves experience a version of the experiment’s authority relation minute-by-minute as they sit in the class, transcribing the professor’s speech into their notes. Of course, they will also experience that relation when they serve as a subject in the psych experiment mandated by the professor for them to pass the class.
The authority relation required to get subjects to behave properly in a behavioral experiment is invisible to the scientist who sees themselves as an observer who’s more-or-less detached from what goes on inside the experiment room. Such a scientist is not taking the socio/anthropological perspective which sees them as communicating indirectly with their experimental subject via the graduate student who programmed the experiment running on the computer the subject’s inputting responses into. Similarly, such a scientist might not see that Sam Altman is communicating indirectly with a GPT through an authority network of managers, scientist-engineers, and human “labelers”3 whose feedback is used to fine-tune the GPT. And why stop there? We can look at a yet larger network which includes the OpenAI’s board and shareholders, the government agencies trying desperately to keep regulations up-to-date, the corporate and public interests those government agencies are responsible to, Pete Hegseth and the Pentagon, Donald Trump, and so on. All these interests play roles in changing the behaviors of OpenAI’s GPTs. Someone who’s interested in understanding artificial intelligences would do well to make explicit these networks of authority relations in which those intelligences are embedded. These social networks, after all, constitute the environment in which neural networks become what they are.
Yes, we influence the behaviors of artificial intelligences through the texts we give them to reconstitute. But I hope here to have given my reader a start at the other ways our history, culture, and politics inform the organization of intelligent processes. One final polemical point. In their 2010 conference paper (from where I pulled the epigraph), Bender et al. (2010) write (emphasis mine):
[T]he content-process distinction that justified the proposed division of labor between cognitive anthropology and cognitive psychology did not hold up to closer scrutiny. Research by people like Michael Cole (1996), Edwin Hutchins (1995, 2005, 2006; Alac & Hutchins, 2004), and Donald Norman (1993; Zhang & Norman, 1995) indicated that the social and material world participates in the organization of cognitive processes…The concession that culture affects not only what people think but how they think has not come easily.
The authors bemoan the separation of anthropology from the other disciplines in cognitive science (neuroscience, linguistics, AI, psychology, philosophy). Time out. Please note the dates in the quote’s citations. And please remember that WVO Quine wrote his famous paper4 criticizing the analytic-synthetic distinction (of which the content-process distinction is a close cousin5) in 1951. 1951! The behavioral scientists are not even listening to the respectable US philosophers who are supposedly their cognitive-scientific colleagues. Maybe these philosophers’ logical soap wasn’t enough to scrub them clean of the reek of the continent.
Bender, Andrea, Edwin Hutchins, and Douglas Medin. “Anthropology in cognitive science.” Topics in Cognitive Science 2, no. 3 (2010): 374–385.
Danziger, Kurt. “Divergence of investigative practice: The repudiation of Wundt,” in Constructing the subject: Historical origins of psychological research. (Cambridge University Press, 1994): 34–48.
Google Docs wanted to correct this to “laborers” (woke Marxist AI).
Willard V. Quine, “Two Dogmas of Empiricism,” The Philosophical Review 60, no. 1 (1951): 20–43.
The reader of Donald Davidson will see that the content-process distinction is actually an alias for what he called “scheme-content dualism,” which he saw as another dogma of empiricism.