“You can’t pick winners in drug development” rhymes with a cluster of memes that are popular in the zeitgeist today:
- “Complicated things can’t be understood from first principles”
- “Collecting a lot of data without models is better than building models”
- “People don’t engage in abstract reasoning much, they do things by feel and instinct”
- “Don’t overthink it”
- “What it means to be human” refers to what distinguishes us from machines, not what distinguishes us from animals
Once you clarify any of these claims down to a specific proposition, sometimes they’re true. But there is a general sense that you can get social approval from saying things whose upshot is “Thinking: it’s not that great after all!”
Sarah Constantin on Less Wrong
I had quite a bit of raw material to start from this time. My brain has woken up for spring and is spamming out all kinds of thoughts of wildly varying quality. I’m still struggling to concentrate on anything for more than a couple of days at a time, so I haven’t got very far with any one thing. But I did manage to write up two posts based on some writing in last year’s newsletters. One is a review of Matthew Crawford’s The World Beyond Your Head, and the other is a follow-up based on some random paper I dug out of the references about kids playing SimCity.
This is something really useful that I’ve got out of the newsletter: there’s a hard deadline at the end of the month so I have to finish something up even it’s incomplete or not very polished. Then later I can go back and use it as a first draft for a blog post, and maybe add in some more ideas I’ve had in the meantime.
This month I’m going to write up a few of the less terrible streams of thought I’ve been having. I’ll start with a few thoughts inspired by Turkle and Papert’s Epistemological Pluralism and the Revaluation of the Concrete, and then I’m going to talk about Anki for a bit, inspired by Michael Nielsen and Andy Matuschak’s new quantum computing essay with built-in spaced repetition. Then I’ll finish up by making the bold claim that thinking is actually good. It’s all a bit on the navel-gazing side this time, with a lot of poking around trying to understand my thinking style, but hopefully there’s some more generally applicable stuff buried in there somewhere.
Turkle and Papert
I talked about this paper in a blog post a while ago. I was really struck at the time by how well their description of ‘bricolage’ fitted me, and if anything the feeling is even stronger on rereading. It’s weirdly diagnostic for me. Pretty much everything they describe as being characteristic of the bricoleur style fits me very precisely. I’ll quote a few that really resonated, starting at ‘true in a horoscopey kind of way where lots of people would agree’ and progressing to ‘how did you get inside my head???’:
Bricoleurs are also like writers who don’t use an outline but start with one idea, associate to another, and find a connection with a third.
Haha, yep, that fits my writing (and thinking) style very well. I do make outlines sometimes, but I do it by pasting a bunch of things I want to cover (which are often big blockquotes of someone else’s text) into a text editor in whatever order vaguely makes sense to me. Then I start swapping them round as needed, and fleshing them out at whichever point looks most interesting to me at the time.
Anne does not write her program in “sections” that are assembled into a product. She makes a simple working program and shapes it gradually by successive modifications. She starts with a single black bird. She makes it fly. She gives it color. Each step is a small modification to a working program that she has in hand. If a change does not work, she undoes it with another small change. She “sculpts.” At each stage of the process, she has a fully working program, not a part but a version of the final product.
Again, yep. I wrote about this in my kaleidoscope post. I would never have been able to survive the ‘send your punch card off to the queue’ early days of programming.
Alex wanted to draw a skeleton. Structured programming views a computer program as a hierarchical sequence. Thus, a structured program TO DRAW SKELETON might be made up of four subprocedures: TO HEAD, TO BODY, TO ARMS, TO LEGS, just as TO SQUARE could be built up from repetitions of a subprocedure TO SIDE. But Alex rebels against dividing his skeleton program into subprocedures; his program inserts bones one by one, marking the place for insertion with repetitions of instructions. One of the reasons often given for using subprocedures is economy in the number of instructions. Alex explains that doing it his way was “worth the extra typing” because the phrase repetition gave him a “better sense of where I am in the pattern” of the program. He had considered the structured approach but prefers his own style for aesthetic reasons: “It has rhythm,” he says. In his opinion, using subprocedures for parts of the skeleton is too arbitrary and preemptive, one might say abstract. “It makes you decide how to divide up the body, and perhaps you would change your mind about what goes together with what. Like, I would rather think about the two hands together instead of each hand with the arms.”
Now we’re getting into the ‘weirdly diagnostic’ stuff. I’m slow to compartmentalising in programming, for the same reasons: I want to ‘see what things mean’ at a glance, and I want to retain the freedom of re-carving in a different way for as long as I can. (Though I notice it’s context-dependent. If I a procedure feels like ‘boring implementation bullshit’, I hive it off to its own function as soon as I can and kind of resent having to think about it at all. If it’s something that feels conceptually important, I really resist separating it out.)
I have the inclination even more strongly in maths, often to an annoying degree where I’ll find myself repeating the same calculation by hand over and over every time I think about a topic, because I can’t face losing track of what the pieces mean. Like, in a lot of my explorations of the Wigner function recently, it would be useful if once I understood some property I could just ‘package it up’ and accept the final result, but instead I have to keep unpacking it, because the property on its own doesn’t drag enough of its own meaning along with it to be any use to me.
When programming, bricoleurs tend to prefer the transparent style, planners the opaque, but the program’s authorship is a critical variable in this preference. Planners want to bring their own programs to a point where they can be black-boxed and made opaque, while bricoleurs prefer to keep them transparent. But when dealing with programs made by others, the situation is reversed. Now, the bricoleurs are happy to get to know a new object by interacting with it, learning about it through its behavior the way you would learn about a person, while the planners usually find this intolerable. Their more analytic approach demands knowing how the program works with a kind of assurance that can only come from transparent understanding, from dissection and demonstration.
This last one is a very specific observation and quite strange on the face of it. But it actually fits me very well, and maybe helps explain something I’ve never understood before. There are a lot of situations where I’m very happy to take things as given and don’t have the slightest curiosity about how they work, and other situations where I obsessively HAVE TO KNOW THE ANSWER and will not let it drop. I think the difference really is something like authorship. I’m happy to start with a load of other people’s stuff as a kind of raw material for my collage (like the blockquotes I paste into my blog posts), but for anything I build using those materials I need to understand how it’s built up from the building blocks.
Anki for conceptual understanding
This month Michael Nielsen and Andy Matuschak released an essay on quantum computing with integrated spaced-repetition flash cards, and I spent a while working through that. I don’t know much about quantum computing in particular, but all the basic QM setup is familiar to me, so I already knew the answers to most of the cards and it’s not going to be the best test of spaced repetition for me. But I’m interested in the basic mechanism, so that wasn’t really a problem – in fact I could concentrate more on it than if I also had to learn lots of new material.
I’d skim-read Nielsen’s previous long post on Anki, and his follow-up giving an example of its use in linear algebra, and had become vaguely curious about its use beyond basic memorisation. I even downloaded Anki and added, like, five cards, in a very pathetic stab at trying it out myself. But I didn’t really see the point, and quickly gave up.
My main interest this time is the possibility of using it to train a skill I’m weak at. I’m very good at getting a vague, sloppy big-picture idea of an area quickly. And correspondingly poor at going back and sharpening it. So I tend to end up with this annoyingly porridgey sort of understanding of a field. Also, a lot of the understanding I do end up with is not very verbal, so it’s hard to communicate. (In real time, I mean. I’m articulate in writing because I can spend the time to fit words to it.) It’s a recurring frustration of mine that I’ll go to a physics workshop and hear people produce these very fluent on-the-spot verbal explanations of things, with precise details included, and meanwhile I’ll have something to contribute like ‘buhhhhh it’s some sort of manifold’, which doesn’t exactly show off my brilliance.
Not too sure of this, but I’m playing with the idea that a major cause of this is that I’m good at picking up associations, bad at learning rules. By which I mean that I’m good at the sort of learning that involves seeing a bunch of examples and noticing common patterns, but not the sort where you explicitly learn a rule and apply it to situations.
(Related distinction: I seem to be good at procedural learning, bad at declarative. I’m generally happy with a piece of knowledge once I know how to use it, and don’t bother to put the work in to make it explicit. Maybe this is the same as associations vs rules, I don’t know.)
Anyway, the ‘good at associations, bad at rules’ thing fits with a lot of things I’ve observed about myself. A couple of examples:
- When I read maths and physics textbooks I ignore the theory text as much as possible and just go through the worked examples.
- I really hate applying a rule that I’ve learned but never seen used before. Learning e.g. a board game is unpleasant because of this. I have a lot of trouble remembering the rule in the first place, but even if I do manage to remember it it feels viscerally wrong because I have to carry out this weird contextless action I’ve never carried out before. It’s a little better if I’ve seen someone else apply the rule already, but still not great.
I compensate for this well a lot of the time – associational and procedural types of understanding are actually really powerful! – but I’ve really started to notice the points where it falls down. The process just inherently produces a somewhat sloppy end result, and I’ve started to realise I want more articulation. It seems like a fairly fundamental feature of my cognitive style, so there are probably limits to trainability, but it seems to be worth at least trying to tackle this head on.
One good example for me would be the command line (this is actually the first thing Nielsen successfully used Anki for – he discusses it in the long essay). I’ve used it for for ten years now and my knowledge is still atrocious. Mostly because I’ve never ‘learned the command line’ in the sense that Nielsen does – learning what the commands do, what the common flags are, etc. I’ve just learned a jumble of ‘things that work’, from copying and pasting or from watching other people, or from just pressing the up key until I find a relevant previous command, and this gets me through a lot of typical situations. If I have to do something remotely outside my normal experience, though, I find it extremely difficult to repurpose that knowledge, because it’s too closely tied to the situation I learned it in.
Anyway, back to Anki. Nielsen specifically mentions sharpness and recombination as advantages of using Anki:
One benefit of using Anki in this way is that you begin to habitually break things down into atomic questions. This sharply crystallizes the distinct things you’ve learned. Personally, I find that crystallization satisfying, for reasons I (ironically) find difficult to articulate. But one real benefit is that later I often find those atomic ideas can be put together in ways I didn’t initially anticipate. And that’s well worth the trouble.
I noticed reading the quantum computing essay that it was gently prodding me towards a more explicit kind of understanding (even for the qubit/linear algebra basics that I thought I knew already). After all, I knew I’d get tested at the end of each section, and the tests would involve being able to reproduce some kind of explicit verbal/pictorial answer that can be printed on a card. I made myself actually reproduce this, by writing it down if necessary, rather than use my normal trick of lazily sort of throwing some approximate nonverbal felt sense at the card and calling it an answer if they matched acceptably.
I’ve undervalued this kind of sharp verbal understanding in the past; the underlying felt sense feels more ‘real’ and like it’s where most of the meaning is coming from. This has previously made me write off things like Anki as ‘just memorisation’. Nielsen says this is quite common:
Many people treat memory ambivalently or even disparagingly as a cognitive skill: for instance, people often talk of “rote memory” as though it’s inferior to more advanced kinds of understanding.
I could definitely train this skill outside Anki as well. In fact, straight after working through some of the quantum computing essay I went into work and found myself putting more effort into actually learning some commands I needed, rather than putting the minimum work into just being able to reproduce them in one situation.
I plan to try one or two Anki experiments next month, starting with using it for notetaking for a paper. I’ll probably use the van Enk paper – despite banging on about it on here for the last year, my knowledge is vaguer than I’d like. I don’t know whether Anki will end up being the specific tool I want, but something in this area could be very useful for me.
Thinking Is Good, Actually
I’ll have to write a proper blog post about this eventually, but here’s a few notes as a start. There’s a weird irony to a lot of the stuff I’ve been writing recently. My blog grew out of a tumblr that I originally got to rant about the crappiness of a lot of maths teaching, and how topics are so often presented in this weirdly abstract, contextless way, with no attempt to explain what any of it means. And also about how incoherent the emphasis on rigour is, where proofs are treated as these incredibly formal things when really the argument depends on a lot of contextual details of what level the audience is at, what they’ll accept as a proof, etc.
I got out some of my frustrations with this in my metarationality post, in the ‘how we think we think, vs. how we actually think’ section. I’m not especially happy with that post as a whole, but that section was definitely therapeutic. A lot of my maths degree was incredibly dry and abstract, and I had absolutely no idea why we were doing it. This put me off whole fields for years. And so I’ve spent a lot of time in the last ten years on this weird personal crusade against rigour and precision. Mostly in a rather self-sabotaging way, because, um, rigour and precision in mathematics turn out to be quite useful. Thinking clearly and explicitly is useful. Who knew??
In the last year or so I’ve started to finally appreciate this. And so I’ve started using this blog I got to rant about rigour to instead start exploring the idea that rigour is useful after all. To the point where I spent a lot of that Crawford review attacking him for making exactly the same mistake I’ve been making, in a slightly different context.
I can credit a lot of this realisation to David Chapman, who has written approximately one billion words to get the single insight that ‘things are nebulous but also patterned’ into our thick heads. While he’s mainly aiming at the sort of pattern-obsessed STEM nerds who have trouble with nebulosity, it also turns out to be useful to nebulosity-obsessed confused STEM contrarians who are fighting their own pointless war against pattern.
And then Norris on Derrida seems to have finally tipped me over the edge. I find this hilarious, given that Derrida is every scientist’s favourite punching bag for woolly, incomprehensible postmodernist nonsense. But actually he’s good at pointing out the useful features of structure.
Anyway recently I’ve been noticing these occasional mild feelings of being positively inclined towards precision, instead of my usual ‘ugh why is this so pedantic’. I’m going to keep pushing gently on that, and I’ll see where I end up.
I’ve decided it’s time for another internet diet, so I’ll be off Twitter, my RSS feed and all my other standard distractions until the start of May. In theory I can spend as much time on the internet as I do in a normal month, as long as I stay away from any habitual distraction loops. In practice I normally spend a bit less time there, and in the time I do spend I normally dig up a few interesting things away from my usual paths. The ideal is to get bored enough that I end up working through a few things I’d never bother with if I had easier ways to waste time. That Norris talk was one example from last time. I had no interest in Derrida and have no idea how I ever found it, but somehow I ended up reading it anyway and have been chewing on it ever since.
I’m going to try at least one Anki experiment as described above. If I go for the van Enk paper, I’ll also combine it with writing up the old notes I made on it for my other physics-focussed website. I did start writing up some other discrete phase space notes for that site, but as with everything else this month I got distracted. So maybe I’ll also manage to get back to that.
Oh yeah, should maybe point out that although I’m off blogs and such I’m still happy to hear from any of you by email 🙂