If we want to do rationality better, it will probably help to understand how we do it already. How about, for example, the polymerase chain reaction, a fundamental method in molecular genetics? Let’s take a look…
The middle part of my book In the Cells of the Eggplant explains an understanding of rationality that will be unfamiliar for most readers. It’s based on actual observation of how we do being rational.
That might seem like an obvious way to find out how rationality works, but it’s rare. Nearly all theories of rationality are philosophical speculations about how rationality ideally ought to work, based on abstract reasoning about rationality itself, without reference to any specific evidence. I call such theories “rationalisms.” The first part of the book points out numerous ways that rationality does not work the way rationalisms say, and could not even in principle. Such misunderstandings are obstacles to improvement.
The last part of the book suggests ways we can do rationality better, based on the understanding in the middle bit. That’s where the practical value is meant to be. This metablog post isn’t about that, though. It’s about the middle bit: how do we do rationality?
Much of my understanding of that comes from ethnomethodology, which might be defined as “the empirical study of practical action.”1 A major method of ethnomethodologists is fanatically detailed analysis of video recordings. That gives insights where rationalist philosophical speculation cannot.
I believe ethnomethodological investigation can dramatically improve our understanding of science, engineering, and other technical professional activities, in ways that might massively pay off for society. It might even pay off for companies; in a few cases, ethnomethodological studies of business activities have produced large financial returns on investment.2 Unfortunately, there’s been nearly no funding for ethnomethodology in decades, and its potential practical value is mainly unrealized.
The Eggplant relies in part on specific studies in ethnomethodology, each based on fine-grained analysis of videos of technical professionals. Among them, for instance:
- Charles Goodwin’s The Blackness of Black, in which a chemistry professor teaches students how to judge when a reaction has completed, by eye and by hand-feel
- Christian Greiffenhagen’s study of a lecturer proving the Dutch Book Argument, a theorem in decision theory, in “Video Analysis of Mathematical Practice?”
- Phillipe Sormani’s “The Jubilatory YES!”, analyzing the instant of discovery of multiband superconductivity in PbMo6S8
The Eggplant’s explanation of rationality begins with “The Parable of the Pebbles.” It demonstrates concretely key points in the meta-rational understanding of rationality. Most of how rationality works is in this story, although that may not be obvious without the following discussion, which I haven’t published yet. But, it’s a parable—a made-up fable. In fact, it’s a standard, simplistic, armchair thought experiment from the philosophy of mathematics, normally used to derive rationalist conclusions. I had fun subverting it by making the story more realistic, adding details that show why the rationalist theory can’t work and how the alternative does. But this was still just made-up, and I might be fooling myself with philosophical speculation, in the same way the rationalists did.
Ideally, every point in The Eggplant would be grounded in extensive empirical work. Unfortunately, less has been done than I would like (due in part to funding problems). I can’t back everything I believe about rationality with studies from the literature. Partly to supplement the existing literature, and partly to test my own understanding, I decided a few days ago that I had better return to video analysis personally—something I had not done in decades. This post reports on my very first attempt, which went surprisingly well, I thought. Bear in mind, though, that I have no formal training in this field,3 and this is a report of casual amateurish observations inspired by ethnomethodology, not the real thing.
Now here in the future, there’s how-to videos for everything on YouTube—a fantastic resource for learning, but perhaps also for research. “Everything” includes many activities we’d count as rational. The first one I thought of, at random, was the polymerase chain reaction (PCR). Popping that into the YouTube search box found many videos, and I clicked on the first one that looked like a how-to demonstration, not a theoretical lecture:
This is a pedagogical performance, not a naturalistic record of routine activity, so it is not ideal. What someone does when being filmed demonstrating a technique might be significantly different from what they would do in their actual work. However, I went ahead with it because choosing the very first data available eliminates selection bias that could lead to confirmation bias. If I went looking for examples of what I believe about rationality, I could probably find some even if they are rare.
It turns out that some central themes of my story show up in this video, including ones that don’t appear in the parable of the pebbles. I’ll zoom in on some in this post. This is not meant to be anything like a complete analysis; I’m just going to point out a couple things I find interesting. You may want to watch the whole video first, and notice what aspects of rationality show up in it. That’s not necessary, though.
Also, PCR is incredibly cool, and worth understanding for its own sake, so you might like to read the short Britannica explanation, or watch this Khan Academy video. But knowing about PCR is not necessary to understand the points I’ll make here.
Checklists: cognitive prostheses
Before diving into the video, let’s step back for a moment. The usual rationalist theory of action is that you should first reason out a “plan” in your mind; essentially, a program to achieve whatever you are trying to accomplish. Then you “execute” the plan, which means mentally running the program. This theory is impossible a priori, and also empirically false.4
We are hilariously incapable of remembering what we have done, what happened when we did it, and what’s supposed to happen next. We are able to be rational only by leaning on cognitive prostheses: physical aids, external to ourselves, that help keep track. Pen and paper, for example:
The physicist Richard Feynman once got into an argument with the historian Charles Weiner. Feynman understood the extended mind; he knew that writing his equations and ideas on paper was crucial to his thought. But when Weiner looked over a pile of Feynman’s notebooks, he called them a wonderful “record of his day-to-day work.” No, no, Feynman replied testily. They weren’t a record of his thinking process. They were his thinking process:
“I actually did the work on the paper,” he said.
“Well,” Weiner said, “the work was done in your head, but the record of it is still here.”
“No, it’s not a record, not really. It’s working. You have to work on paper and this is the paper. Okay?”5
Imagine trying to write actual program code blindfolded. It’s impossible. In reality, your eyes are constantly flicking around the screen to look at what you’ve already done. Programmers prefer giant monitors so more of the context is visually accessible without scrolling or switching files.6
The simplest cognitive prosthesis for keeping track of what you have done, and what you need to do next, is a checklist. These can be literally life-saving: Atul Gawande’s Checklist Manifesto recounts that hospitals using surgical checklists dramatically decreased medical errors.
At 1:20 the PCR video explains:
As you add the reagent to the master mix, you tick off your working list to make sure that you know you’ve added your reagent, and that you don’t add double the volume that’s required.
These mistakes are easy to make. I have made them both many times myself when doing this kind of laboratory work! It’s the status of such mistakes that makes PCR a formally rational activity, and not a “merely reasonable” one (in the terminology of The Eggplant).7 If you miss or double a reagent, the results of your PCR are incorrect. Making waffles is a merely reasonable activity. If you forget to add oil to the batter, or space out and put it in twice, the results may be bad, or may even be inedible non-waffles, but they can’t be incorrect.8 This is marked linguistically: as in the title of the video, you perform PCR, meaning you do it rigidly, “by the book.” You don’t perform waffle-making, you just do it.9
In the video, you can see her check-marking a box on her paper worksheet after each reagent transfer. By analogy with a programming language runtime, she’s incrementing her program counter, stored externally on paper, after each operation.
Arranging equipment as a reminder
Needing a checklist is fairly uncommon. In most activities, you can usually see at a glance what you need to do next just by looking at the configuration of the objects you are already working with. (This was one main point of the Pengi program I wrote with Phil Agre.10)
In some situations, you can’t see what to do next; in some, it just comes naturally that you can. In some other situations in which it’s not natural, you can arrange already-available objects to track your order of business, without needing a separate device for the purpose. As we wrote in the abstract-emergent paper,
Bicycle repair manuals tell you that when you disassemble some complex subassembly like a freewheel, put all the little pieces down on the ground in a row in the order you disassembled them. That’s because (if you know as much about freewheels as I do) the only available representation of the pieces is “weird little widgets.” They are indistinguishable from each other under this representation. Because quasi-static hardware can’t represent logical individuals, you can’t remember (much less reason about) which is which and what’s connected to what.
This story illustrates several themes of the paper. The manual’s advice is a piece of culturally transmitted metaknowledge that allows you to work around the limitations of your hardware in order to perform activity in a partly planned way. The plan—the order in which the freewheel should be reassembled—is not sufficient to completely determine the activity; you still have to be responsive to the situation to see just how each piece should be put back on. In fact, the plan is not even represented in your head, but externally, as a physical row of objects on the floor.
At 1:35 in this video, he explains:
Now if you are unsure about anything, actually lay out the components in the order in which you take them off. That way is going to make reassembling a lot easier. Also if you’ve got a smartphone, perhaps take pictures as you go along.11
And at 5:02, you can see how he’s laid the pieces out, and he says:
Clean one piece at a time, so you are not disrupting the order at all, as there’s nothing worse than basically trying to put a jigsaw back together when you’re not maybe 100% sure which part goes where!
This tip has saved my bacon several times in various mechanical disassembly/reassembly tasks. If you learn nothing else about rationality from Meaningness, this advice alone is worth the price of admission.
It has probably also saved the scientist in the PCR video a few times. If you watch carefully, you can see that she’s arranged the reagents along the back edge of the ice bucket in the order they get added to the master mix. (That’s the back edge from her point of view—it’s the edge closest to the camera.) So she works her way left to right (from her point of view) through the reagents. That redundancy may have saved her in even this fake, for-the-camera performance of the procedure: she actually fails to tick off the first reagent (“buffer”), at about 0:58!
Something very interesting happens during the next minute. I didn’t notice it until I’d watched the video about twenty times! Maybe you can see if you can spot it:
I’m going to give you a series of clues and then an explanation. You may want to try watching that minute’s worth of video repeatedly as I give more clues? Or just skip ahead if that’s not fun.
You’ll need to know how a pipette works to understand the full story. If you aren’t familiar with pipetting, this clear, two-minute how-to video explains everything you need to know:
Or you can read the Wikipedia article. But if you don’t want to bother, you can get a general sense of what happens without knowing about pipettes.
Watch the video again carefully from 0:58 and see what you see?
Below the video, commenter Nguyễn Xuân Tài observes that “At 1:12 the technician throws the pipette.” Can you figure out why? (Two other commenters reply “Yes she did” but no one explains.)
You might want to watch just the few seconds around that point a few times and maybe you’ll spot it.
Here’s the narration:
The next reagent is magnesium chloride. Between reagents, always make sure you change the tip, so you don’t contaminate your reagents. Magnesium chloride will be added. Pipettes are used to ensure you get the correct volume of each reagent. As you add the reagent to your master mix, you tick off your working list to make sure you’ve added your reagent, and that you don’t add double the volume that’s required. OK. So the next reagent will be magnesium chloride.
What’s anomalous in that?
Watch again, paying attention to what happens at the anomalous moments in the narration?
A detailed transcript
- finishes adding buffer to the master mix tube
- the next reagent is
- ejects the tip, off camera, but you can hear the click
- magnesium chloride
- attaches a new tip from yellow box
- between reagents, always make sure you change tips, so you don’t contaminate
- inserts the tip in the magnesium chloride tube
- your reagents
- ejects tip
- magnesium chloride [momentary pause, voice trails off]: will be aaa…
- slams pipette down on bench, simultaneously picks up a different pipette with other hand
- adjusts volume knob on new pipette
- ticks worksheet, throws down pen
- continues volume adjustment
- pipettes are used to ensure you get the correct volume of each reagent
- puts down pipette, opens new blue box of tips, with some difficulty
- as you add the reagent to your master mix, you tick off your working list to make sure that you’ve, you know you’ve added your reagent, and that you don’t add double the volume that’s required
- attaches blue tip
- [swallowed, nearly inaudible]: okay
- picks up magnesium chloride tube
- so the next reagent will be magnesium chloride
- aspirates magnesium chloride
- ah, ah, magnesium chloride helps the enzyme to anneal to the DNA strand
- dispenses magnesium chloride into master mix tube
- the enzyme used is Taq polymerase, which is added at the end
Now do you get what’s going on?
Repair is a major theme in ethnomethodology. Everything is constantly going wrong, because the world is nebulous, so your actions don’t do what you meant them to. This is nearly always unproblematic, because you know how to do something else that repairs the problem. You just do it almost without thought; and so the pervasiveness of minor breakdowns and repairs is invisible. It only shows up when you do careful video analysis. Then you see that in most activity—including technical, rational activity—repairs occur startlingly often. Repair is a big part of the circumrational work we do to make rationality work, because the world is never quite as any rational theory supposes.
How many repairs can you find in these 49 seconds?
Her actions are part of two different simultaneous streams of activity, each with breakdowns and repairs. On one level, a scientist is “performing” a polymerase chain reaction. The other level is a pedagogical “performance,” by actors who happen also to be scientists, of a staged how-to video. For meta-rational purposes, only the science level is relevant. Here’s my take on some key events in that:
- After adding the buffer, she forgets to tick it off on the worksheet.
- After ejecting the buffer tip, she continues using the same pipette, picks up a new tip, and goes to aspirate the magnesium chloride. The instant before doing so, she realizes she’s forgotten to adjust the volume for the new reagent, which is less than for the buffer. In fact, she will need to use a different pipette to pick up the smaller volume.
- She slams down the large pipette because she’s annoyed by the flub.12
- While adjusting the volume on the small pipette, she realizes she forgot to tick off the buffer. She abandons the adjustment to do that, lest she forgets again, throws down the pen in hurried irritation, and then returns to the interrupted task.
So there’s two skillful repairs in that: switching to the correct pipette after starting to use the wrong one, and remembering to tick the worksheet after forgetting. Both of these would be automatic in the routine activity of PCR, and totally unproblematic so long as you did eventually apply them.
It seems likely that both errors were due in part to feeling self-conscious due to being on camera. From other clues in the video I also suspect this was the umpteenth live take, and she was sick of the whole thing, which makes one more prone to careless mistakes.
Repairing the video
Whereas the errors would be unproblematic in the course of doing science, flubs in a how-to video might be embarrassing. However, I totally missed them until I’d watched it about a million times.
That’s partly due to skillful repair work on the part of the narrator. I’ll point out some of these, although they are not relevant to technical rationality or the Eggplant project. (You may want to skip ahead over this, to get to the conclusion of the post, where I explain why you might care about the rest of it.)
Before reading on here, you might consider the question: why does he use the word “so” in “so the next reagent will be magnesium chloride” at 1:36? What work is that word doing there?
The great thing about YouTube is that you need zero knowledge of video production to make films that vast numbers of people derive significant practical or entertainment value from. This PCR video has had 224,580 views as I write this.
Probably a professional would have filmed it without narration, and added that as a voice-over later. And they would have “repaired” action errors (like using the wrong pipette) by splicing together bits of a few different takes. Here the senior scientist did the narration in real time, and they filmed the whole procedure in a continuous take.13 His repairs concealed her action errors from me the first many times I watched the film, as well as covering up clues in his own performance that something had gone wrong.
- At 1:10, a second after she realizes she’s using the wrong pipette, he also realizes, because she goes to eject the tip without doing a transfer. This is where “… will be added” trails off.
- He’s silent for ten seconds, perhaps considering what to do—should this video take be aborted?
- During those seconds, she fluently repairs the error, and apparently he decides to go with it, without comment.
- At 1:16, probably primed by the salient volume error, he says “pipettes are used to ensure you get the correct volume of each reagent.” This comment is otherwise somewhat unmotivated and out-of-place; logically, it belongs at the beginning of the demonstration, when she first starts using one. It fills space as she gets back on track, though, deflecting the viewer’s attention from “wait, what did I just see?”
- At 1:22, he comments on the worksheet-ticking, perhaps having just realized himself that she’d forgotten to do this earlier. His distracting patter also fills what would be an awkward silence, during which the viewer might otherwise start to figure out that something has gone wrong.
- Why does he say “okay” in a strangled whisper at 1:33?
- He repeats “the next reagent will be magnesium chloride.” I totally missed that he is now saying this the third time. Why didn’t I notice?
- He continues with an explanation of the function of the magnesium chloride; maybe that distracts from the anomaly of the triple repetition.14
So about that “so.” In ethnomethodological conversation analysis, there’s a large literature on “so” at the beginning of utterances. Galina Bolden went through 80 hours of recorded conversations, and found and analyzed the 250 utterance-initial “so”s in them.15 There are several distinct functions of “so,” of course; it can mean “therefore,” for example. The relevant function here is to communicate the intent to return to “a course of action has been interrupted or subverted in some way before coming to a possible completion” (in Bolden’s words). It’s a conversational repair method for an intended activity that has gone off track. This is, of course, exactly what happened with the addition of magnesium chloride.
This use of “so” communicates that the project of the utterance following does not relate to the immediately preceding activity, so that context is no longer relevant and should not be used to interpret what comes next. If the hearer accepts this move, their conceptual attention reorients to the new topic, and it is likely that the interrupting event will be immediately forgotten. In this case, if the viewer of the video is vaguely aware that something has gone wrong, that is likely to drop out of consciousness: especially since the “new” topic is the familiar “old” one, adding magnesium chloride.16
So what have we learned from all this?
- If highly-trained scientists can’t follow a familiar laboratory procedure without occasional goofs, and rely heavily on external cognitive prostheses to keep track of what they are doing, the rationalist theory that action derives from mental execution of programs is refuted.
- If the sorts of phenomena ethnomethodologists identify as typical of reasonable action show up frequently in the first brief video of technically rational practice I looked at, it’s plausible that they are pervasive in STEM fields. (As, in fact, some ethnomethodological studies have found.)
- If a catalog of such phenomena can be extended to cover most rational activity, it’s plausible that we can construct an empirically accurate, non-rationalist theory of rationality, including scientific, technical, engineering, and mathematical practice.
- If rationality works because we make it work—a slogan of The Eggplant, explained briefly in the Parable of the Pebbles—and not because it is Cosmically Correct, then it is plausible that we can make it work better through interventions based on more accurate understanding.
- If rationality typically involves frequent minor breakdowns and repairs, it is plausible, as one example of the previous point, that empirical examination and understanding of their causes is one route to meta-rational improvement of rational practice.
- 1. Actual ethnomethodologists would probably look pained at this definition, and at minimum would want to add numerous caveats, and might reject it altogether.
- 2. The outstanding example is Julian Orr’s study of the social epistemology of Xerox repair people, reported in his Talking about Machines: An Ethnography of a Modern Job. This led to $100 million in savings for Xerox (John Seely Brown and Paul Duguid, “Balancing Act: How to Capture Knowledge Without Killing It,” Harvard Business Review, May-June 2000.)
- 3. I spent three months at Xerox PARC in 1989, casually hanging out with the ethnomethodologists there, from whom I feel I learned an enormous amount by tacit apprenticeship. I’m grateful to Lucy Suchman, Jean Lave, Gitte Jordan, and others, for their kindness to an immature and arrogant computer geek.
- 4. See Philip E. Agre, Computation and Human Experience, particularly Chapter 8; and my Vision, Instruction, and Action, particularly Chapter 2 and section 3.1.
- 5. Clive Thompson, Smarter Than You Think: How Technology Is Changing Our Minds for the Better (2013).
- 6. Since writing this, I have learned that there are fully blind programmers. They use utilities that read a line of the program out loud, sped up enormously. After years of practice, they dart around the code base by ear in the same way that a sighted programmer darts around by eye. They also report developing their memories, both long term (API function signatures for example) and short term (the immediate code environment), well beyond what sighted programmers are capable of. Inspiring!
- 7. Most fields studying “rationality,” including ethnomethodology, use the word to cover both everyday “reasonableness” and technical, systematic, formal rationality. I find it helpful to use different terms for the two in order to pick apart clearly how they relate to each other.
- 8. Whether the waffles are bad, or inedible, is—in ethnomethodological terms—a matter of hermeneutic, accountable interpretation and negotiation. (Your family gets a say, and will definitely have opinions.) Whether your performance of PCR is incorrect is—almost always—an objective fact, in the sense that a group of observers watching over your shoulder would agree.
- 9. Unless somehow you have an audience for your performance.
- 10. Philip E. Agre and David Chapman, “Pengi: An Implementation of a Theory of Activity,” Proceedings of the National Conference on Artificial Intelligence, 1987, pp. 268-272. With The Eggplant, I appear to be mostly just rewriting my PhD thesis thirty years after the fact. This is lazy and stupid, and someone should organize an intervention to force me to think of something new to say.
- 11. The video is about “cartridge hub bearings.” Cartridge hubs are a simpler, better technology that replaced freewheels a couple years after I wrote the abstract-emergent paper. They’re basically the same thing, except with a million weird little widgets in them instead of a billion.
- 12. It would be more ethnomethodologically correct to say that we can see she’s displaying annoyance. We can’t see inside her head, but we can see this as communication. Put in words, she might be saying to the narrator, apparently her doctoral supervisor: “This is the fifteenth time we’ve been through this, and every single time one of us screws something up and we have to start over. I’ve done about as much of this as I’m willing to; I want to go back to being a scientist, instead of playing one on TV.” This is just speculation, in the absence of additional context. That she is displaying irritation is an objective fact, however.
- 13. Actually, there are two splices, at 2:55 and 3:31, so they did know how to do that. I can guess what the second splice is repairing; I have no idea about the first. I also don’t know why they didn’t splice out her errors and replace them with footage from another take.
- 14. Although this was a highly skillful series of repairs, I doubt it was calculated. Such conversational methods operate at the threshold of awareness, like most “merely reasonable” activity. Rationality typically (but not always) requires explicit thinking-through.
- 15. Galina B. Bolden, “Implementing incipient actions: The discourse marker ‘so’ in English conversation,” Journal of Pragmatics Volume 41, Issue 5, May 2009, Pages 974-998. Her paper also includes a useful review of prior research.
- 16. Computer people can think of this by analogy with context switching in an operating system’s process scheduler. Ethnomethodologists will find this analysis of mental events unacceptably cognitivist. It’s sheer speculation (like most cognitivist theorizing). Sorry. I’m humbled by the empiricism of anyone who can work through 80 hours of painstakingly transcribed conversation.