A friend recommends a scientific paper.
At this time I am a computer science student thinking of quitting computer science, because I live in California, and love computers but have grown exhausted by Silicon Valley.
The paper is called ‘Building Machines That Learn and Think Like People’, and it is sixty pages long. It turns out to be about artificial intelligence, and cognition, and also brains, and babies, and what it means to be human. There are diagrams of Segways. It contains a surprisingly lengthy discussion of igloo construction. I read it all in one go at my dining table, and over the coming months, I will mention this paper to friends and friends of friends until they tell me to stop.
On page 9, the paper slips in a line about what the authors refer to as ‘common knowledge’, the fundamental facts of the world that even young children know -- when piles of blocks will stand and when they will fall, or what it means to smile. Of these things, the paper says,
These core domains cleave cognition at its conceptual joints.
The phrase is startling, poetic. It conjures a world that might be carved into beautiful and knowable pieces, if only we understood where to look.
At the time I spend most days in classes looking at slides that say things like, ‘this neural network used 128 layers of neurons.’ None of us know much about neurons. This is different. This is real science, I think.
Fig 1. The paper, of course.
I fly out for a graduate school interview with the MIT Brain and Cognitive Sciences Department, where I have applied because an author of the paper teaches there. I do not know much about cognitive science. I am not even entirely sure, by this point, that I know much about computer science. During interviews I mention the paper. It is possible that I mention the phrase, ‘cleave cognition at its conceptual joints’. It is possible that I claim to want to cleave cognition at its conceptual joints. One of these conversations takes place at a crowded beer hall, with that paper author, who will later become my advisor. In my memory, I shout that I want to build machines that learn and think like people. It is a confusing time.
Remainder of March until the last week of April.
A period follows in which I consider abandoning science to study fiction writing.
I call the professor from MIT; he tells me that in his undergrad he considered writing musicals.
I read the paper again.
I want to cleave cognition at its joints, I think.
I call my creative writing professor, who has written me recommendation letters to many creative writing programs, and tell him that I need to go to MIT.
I arrive in Cambridge. It is oppressively hot. The air conditioner in the apartment hums when turned on. One of my many roommates begins sleeping out on the porch. During MIT orientations, I meet the other new students, and we attend one presentation after another on the many fascinating mysteries of the brain.
Around then, a weird thing happens. I start seeing variations of the phrase “cleave cognition at its joints”. I start seeing them everywhere.
How it starts: a newly minted professor gives a talk in our systems neuroscience class about language. He clicks to an introduction slide.
Fig. 2. Excellent slide courtesy of Idan Blank, who humored me by sending this over.
A few weeks in, a former student from the department gives a presentation. Afterwards I dig up and read a paper on his research, which includes the line,
We believe that the modular disentangled structure in CRL biases it to cleave the problem distribution at its joints, yielding this 10-fold reduction in sample complexity.
In October I enroll in a developmental psychology class.
Our professor assigns two papers. One is titled ‘Carving the World for Language’. Another is titled ‘Carving Events for Learning Language’.
We are asked to read a seminal paper and a series of responses drawn from across cognitive science. Together this collection mentions carving, or joints, at least five different times.
Fig. 3. The collection. Also, that B.F. Skinner reference doesn’t even seem to exist on the Internet.
I attend a student-lead philosophy meeting. We read a philosophy paper from the 1990s.
The phrase starts to take on a weird alien sheen, like an obscure band I’ve listened to once and now can’t stop hearing everywhere. Can science actually have joints? What does it even mean to carve anything?
I tell another friend, a graduate student in chemistry, about the strange affinity between cognitive scientists and carving. “Wow,” he says, “I love that phrase.” He is making a presentation for his lab meeting and adds it to one of his slides.
Fig. 4. A bit of the fateful slide.
I wonder, Is this a meme? Am I witnessing a meme happening in real life?
It seems extremely critical to understand exactly how this phrase and its carve-happy brethren began slicing their way through so much of cognitive science, an importance that will later seem suspiciously inversely proportional to my understanding of how to do research.
Google mostly returns Q and A posts from anonymous answerers, many of which link circularly back to each other. Most responders seem to obliquely agree that the phrase comes from Plato, but the original source is surprisingly difficult to find.
“Hi Tom. Cutting up any body is difficult, but it is a process made relatively easier if one incises at the joints, as a butcher typically would do. Thus a concept, area or object that is not in nature separated can be done to some degree for a satisfactory physical or conceptual result by approaching it at a naturally softer point.”
This is alarming, but admittedly, conceptually clear.
The class ends. I go home to New Jersey. Around Christmas, we serve an actual roast chicken. We carve it, mostly at its joints, and I try not to think about cognitive science.
I do actual research and forget about the whole thing until a workshop on writing for this very blog. When I pull up an old list I have made of possible topics, there are four boring ideas and one all caps note that says, “CARVING???”
Basically last week.
I sit down to write. OK, I think. Plato. Start by finding Plato.
After a whirlwind tour through the Perseus Digital Library of Classical Texts and its counterintuitive user interface, I finally land on this, which translates to:
That of dividing things again by classes, where the natural joints are, and not trying to break any part, after the manner of a bad carver.
and which answers the question of origins, but doesn’t actually explain how or when when cognitive scientists began quoting Plato en masse.
I email one of the professors from the developmental psychology class. He is also one of the authors of the original paper, on building machines to learn and think like people. I ask, does he have any idea how all the carving got started? He replies within minutes. At last, I think.
“What an excellent idea for a blog post!
I'm afraid my answers will disappoint you."
In his email, he includes a link to a graph he has searched of its usage, and concludes that he has no clue where he heard the phrase first.
Fig. 6. This is not a timeline that starts with Plato.
Carving cognition at its joints seems to be nowhere and then, suddenly, everywhere. He says he thinks he might have heard the phrase for the first time back when he started grad school, in the lab where I work now.
This evening, while writing this post, I manage to trace the phrase back to an old address to the American Philosophical Association, and a somewhat well-cited psychology paper, and a newer encyclopedia entry, and even an old bibliography of the DSM-III, but eventually I start to realize that there might be no real single beginning, that it could be like trying to find the tipping point of a tide. At some point I find myself back with the original paper, scrolling through all sixty pages, and I realize - as I consider returning to search for the inflection point of the phrase - that I don’t actually care.
Flipping through the old paper is surreally comforting. It’s been a long time since I’ve opened it again; reading it is like spending an unexpected hour with a distant but familiar friend. I go back to page 9. How couldn’t I? I go back to Plato. I sit for a while with the phrase, with both phrases. It is strange to read them again after what I’ve come to think of as our history together -- so willfully silly that I’d forgotten, I realized, how they were also profound.
Even now, I read them again, and they were still beautiful to me.
They still carry the ring of real science. Just as lovely and hopeful, somewhere between mantra and dream.