Episode post here. Transcription by Gavin Leech.
Matt Teichman:
Hello and welcome to Elucidations, an
unexpected philosophy podcast. I’m Matt Teichman, and with me today is
Sam Enright, editor-in-chief of The
Fitzwilliam, which is an online
publication about Ireland, policy, philosophy, and literature. He is
also a non-resident fellow at the policy think tank Progress
Ireland and an Emergent
Ventures fellow. And he’s
here to discuss lifelong learning. Sam Enright, welcome.
Sam Enright:
Thanks for having me.
Matt Teichman:
I would definitely put you at the top of my
list of people who are really good at learning. But before we talk
about some of the ways that you like to learn, maybe we could just
address this question of: is it a good thing to learn? A lot of us
assume the answer is yes. I certainly assume that; I love
learning. But I don’t know—is that wrong? Is it a waste of time to
be learning?
Sam Enright:
[ LAUGHTER ] I think the question of the returns
to effort of learning about different fields is a really non-trivial
and interesting one, where by default, maybe people assume that linear
model of: you on average will get a lot greater wisdom and insight
about a field by reading about it more, discussing it more with
people. But also, an effect of learning more about a certain topic can
be to make you more insulated in your views. I believe it’s a
reasonably consistent social science finding that people’s views are
more correlated with one another at higher levels of knowledge.
Matt Teichman:
Yeah, you definitely find that in academia.
Sam Enright:
Yeah.
Matt Teichman:
You get these little bubbles where no one
challenges anyone’s basic assumptions, because there aren’t enough
outsiders learning about the conversations that are taking place in
the particular bubble.
Sam Enright:
Yeah, my friend Sam
Bowman, who I
think you might also know, had this line to me before which was: “It
seems like our options are either to be open-minded ignoramuses or
extremely polarized, politicized, knowledgeable people.”
Matt Teichman:
Yes.
Sam Enright:
[ LAUGHTER ] And on the whole, he would prefer
that we be highly ideological, knowledgeable people.
Matt Teichman:
Mm-hmm.
Sam Enright:
But is somewhat depressing, but that’s—
Matt Teichman:
—it’s kind of a sad choice though, isn’t it?
Sam Enright:
Yes. [ LAUGHTER ]
Matt Teichman:
I feel like you kinda just described college
versus PhD school right there. Like, college is the time where you’re
excited to learn about all new things; you’re kind of a neophyte. And
then, doing your PhD is the time in which you’ve been conscripted
into the service of a field. You’re no longer working for yourself;
you’re working for the field, and now there’s ideology to contend
with. Not as much ideology, perhaps, to contend with, when you’re
shopping around the buffet of college classes.
Sam Enright:
Yeah. I have an economics professor friend, and
he was commenting to me recently that one of the things he likes about
economics is that the returns to learning about it as a field
are—it’s quite straightforwardly diminishing returns. You can get
the majority of the value that people will get, let’s say, of a whole
degree, in the first one year or two years. You can probably get a
decent fraction of the value from those years within a single
class. Even within that, if you can really convince people of the
usefulness of certain economic ideas, you can get a lot of it across
in a couple of hours.
He was commenting upon how history, which is another field that he thinks about a lot, seems to basically be an area of linear returns, where he feels glad that he has many different mental models from different cultures and time periods to think about. For example, modern political problems. But he suspects that the returns to this are totally linear, where it’s just better and better the more history you know, but there’s no general lessons of history.
Matt Teichman:
Hmm. Are there fields where it’s the
reverse—where there’s no payoff for the first few years you study
it, then suddenly it just gets exponentially better, once you’ve put in
a certain amount of time?
Sam Enright:
Do you think philosophy might be like that?
Matt Teichman:
Uh, no. I think it’s not like that. [
LAUGHTER ]
Sam Enright:
Okay.
Matt Teichman:
[ LAUGHTER ] No, more like economics, I would
say.
Sam Enright:
Okay. Interesting. I think economics is flipped
the other way.
Matt Teichman:
I mean, I don’t really know. This is just me
talking through my hat, but anecdotally, it seems like the biggest
transformation that I’ve witnessed happening in people is when they
take their first couple philosophy classes. Suddenly, now they’re
getting a little more careful about making hasty generalizations, and
they’re thinking about, e.g., “Well, okay, any time I say anything,
what’s a possible counterexample?” And, you know, “Gotta make sure all
my i-s are dotted and my t-s are crossed when it comes to making a
case for something.” In my experience, people get those kinds of
payoffs from their first few philosophy classes, and then after that,
it tends to just be getting further and further entrenched in your
area of specialty.
Sam Enright:
On the flip side, there do seem to be some
areas, particularly ones that are quite vague, or sort of slippery in
their concepts, where you can bang your head against it for a really
long period of time, and then there will actually be some moment of
very sudden insight. Maybe meditation, and learning about certain
spiritual practices, and so on, is one of these areas that has
negative returns for the first few hundred hours that you do it.
Matt Teichman:
Mm-hmm.
Sam Enright:
I can’t really speak to it, but people claim
that there’s subsequent major gains.
Matt Teichman:
This kinda gets back to my initial question
about, like, is there anything where the returns on learning it are
always negative? It just keeps getting worse. I’m inclined to think
there must be some such things, but I don’t know. Maybe it’s always
good to learn, and the difference is just a matter of the rate at
which the good things happen, when you start learning something.
Sam Enright:
Yeah. I worry a lot that certain of the topics
that I look into, I will never have the time to approach the thing
with depth, and I could be accused of just basically learning trivia.
Like memorizing basic details of a certain thing, or names and
years. This kind of stuff. Maybe there’s something to that. I find,
interpersonally, there’s a lot of returns to, for example, learning
basic things about the histories of different countries. Like, I just
got back from Taiwan and China, and the bar for knowing anything about
Chinese history among Westerners, even that visit there, is so low,
to the point that you have a lot of interesting conversations and
interactions with locals, by even knowing the bare minimum. And I hope
that I know a lot more than the bare minimum about these things.
At a certain point, I just realized that I also kind of enjoy learning trivial details about things. That’s a quite peculiar personality characteristic. I know there’s a lot of people for whom those precise details, or memorizing dates, is actually somewhat unpleasant. But the raw experience of recall I find relatively enjoyable. My grandmother used to say when we were kids, “Knowledge is no burden,” anytime that we asked why some specific piece of information was useful. [ LAUGHTER ]
Matt Teichman:
Why do I have to do my homework? Well, it’s no
burden. Do you think some things are easier to learn and other things
are harder to learn? How should we think about that issue?
Sam Enright:
This seems like a very Socratic sort of
question. Like, you might see a question like this in Plato’s
dialogues. It’s interesting to think about whether the characteristic
that different fields are differing in is that some are made up of a
larger number of individual concepts or atoms, versus are the atoms
of understanding very complex in themselves?
Matt Teichman:
I’ve definitely experienced this teaching
philosophy and linguistics versus teaching computer science. In
computer science, the sheer volume of fiddly little details that it’s
easy to mess up and get wrong is quite striking, compared to in
philosophy and linguistics, where you are basically trying to get the
high-level stuff right, but you don’t have to fuss over tons and tons
of little arcane details of e.g. how does this computer system work
and what is the exact arithmetic property of this versus that etc.
Sam Enright:
Yeah. I wonder: mathematics is one obvious area
where it seems like the concepts themselves are very complex. But I
wonder if this is somewhat of a cognitive limitation about the amount
of time or energy humans are reasonably able to devote to it. Like,
if you actually spent long enough doing this thing, then you would
notice these higher level patterns, or you would crystallize certain
patterns or ideas at a higher level, in a way that would make the
whole concept easier. And I wonder, is there some sort of pressure
that those crystallized understandings are—maybe there’s some kind
of cognitive argument that they should be approximately similar in
complexity, across disciplines. I mean, if you even just think about
how calculus is sufficiently easy now that secondary school students
can learn it, compared to how it was so complicated that only the
greatest minds in the world could do it in the 17th century. You
wonder: what is the limit on this process?
Matt Teichman:
Right. Is graduate-level mathematics going to
feel easy to us in another 100 years?
Sam Enright:
Right. Somebody that might be good for your
podcast is, David Deutsch, the
physicist. He has this book called The Beginning of
Infinity. It’s
about many things, but one of the topics that it’s about is this
concept that humans have passed some kind of threshold of being
universal thinkers, or universal solvers. I think he would argue in
favor of this view that there’s no limit on this process: we can keep
simplifying, and keep simplifying, and keep coming up with more useful
concepts, to abstract things down, and there’s no limit on what is in
principle human-understandable. I don’t agree with that view
necessarily, but that’s one very strong statement of it.
Matt Teichman:
Do you think that this phenomenon, whereby
certain things that used to be hard are now easy to learn—there’s,
like, a trade-off there, where maybe 200 years ago, there were certain
things that people found easy to learn that we now find hard to learn?
Or do you think it’s cumulative, in the way that Deutsch is
suggesting?
Sam Enright:
So, when I think about things that are more
difficult to learn now, I might think about—
Matt Teichman:
—like using a loom? I don’t know. Like, what
did people do a long time ago?
[ LAUGHTER ]
Sam Enright:
I would think about a lot of practical
skills—certain skills relating to foraging and knowledge of
agriculture and being able to identify certain plant species. When you
read kind of anthropological accounts of different hunter-gatherers,
they often mention how they’re able to tell the difference between
certain plant species so easily, to the point where even teenagers are
able to do it. But in the West, only some of the most advanced
botanists in the world would be able to make such fine
differentiations, because learning those precise skills is very
important to these tribes.
I’m just not sure what is the appropriate level of skepticism to apply to those claims. I think anthropology is one field that has suffered from a lot of political bias in our choice of how to represent these relatively uncontacted groups. And also, even those groups—is this just making a point about how human society is quite culturally contingent, and there’s cultural variation across these groups? Maybe there’s other tribes that are absolutely terrible at this.
Matt Teichman:
Yeah, absolutely. This is something we’d want
to try to verify with empirical evidence. But nonetheless, I guess it
does have the intuitive ring of plausibility, from the armchair, that
given the right cultural background, where we’ve figured out how to
teach something, and that skill is something that’s in the air, you
would think that most people would find such a skill, given that
background context, easier to learn. One example I often come back to
is that there’s this certain on the beat but also loose style that
‘70s drummers used to play in, which you kind of don’t hear after the
‘70s. You can hear people fairly closely replicate it now, but at the
same time, you put on any funk record in the ‘70s, and it’s got this
just very distinctive groove to it. Drummers call it perfect pocket.
Sometimes I wonder whether there can just be skills that are in the
air, at a given time.
Sam Enright:
Yeah. Certainly, a lot of these come to mind
with relation to architecture, and physical construction of
infrastructure projects. The Empire State Building went up in, I
think, 18 months and was under budget and early. Collectively, as a
civilization, I’m not sure we actually have the knowledge to build
infrastructure projects with the speed and efficiency with which they
were previously constructed. You see a lot of these things in the
physical environment, like many countries now building so few airports
and stuff. It’s not clear, even if it were a national priority, that
they would actually be able to do this. I think it would be very
interesting and would maybe have civilizational consequences if there
needs to be an unbroken line of people engaging in a certain skill.
Like it has to be taught father to son, like—
Matt Teichman:
—because it’s kind of a craft knowledge.
Sam Enright:
Yeah, literally or metaphorically. And if this
ever stops—if your country stops building important civic
infrastructure for 40 years, you’re just never going to be able to
start doing it again at the levels of beauty, and efficiency, and
aesthetic considerations, and so on, as you were able to do before.
Matt Teichman:
Or anyway, if it were to happen again, the
same planets would have to be in alignment that were in alignment the
first time it happened.
Sam Enright:
You could retreat to a weaker claim, which is
that you just need there to be an unbroken line somewhere in the
world, because you can always import laborers from another country. Or
just in the extreme case—you know, Italians are quite efficient at
building rail infrastructure—you could just hire them to build your
country’s train system. But of course, you would wonder about how much
there’s region-specific cultural and economic differences between
places that would mean that the skills were not fully transferrable.
Matt Teichman:
So we’ve talked a bit about whether some
things are easier versus harder to learn, and we’ve talked a little
bit about: are there things that are worth learning, and things that
are not worth learning? What are some examples of things you’ve
learned recently and what were some of the techniques that you used to
learn them?
Sam Enright:
I read a fair amount about, uh, Taiwanese
history in anticipation of going to Taiwan, reading several books,
papers, and making a lot of flashcards associated with them. I’m a
total evangelist for spaced repetition as a system of more
effectively memorizing.
Matt Teichman:
Can we give a little background on that? What
does this term mean and where does it come from?
Sam Enright:
The historical origin that people usually refer
to is the psychological concept of a forgetting curve, which is that
your level of retention of a certain topic seems to decay
approximately exponentially. You can imagine it getting reset if you
have to recall the thing again up to, let’s say, 100% retention in the
extreme case when you’re relearning it each time. But every time that
you are reminded of that thing, or have to actively recall it again,
the decay slows down. My understanding is that there’s this
psychologist called
Ebbinghaus, and he
verified that this does actually seem to follow an exponential curve
for human memory, and also that the exponent seems to change, such
that the memory curve decays more slowly, over time, with a larger
number of active recalls.
This implies that the optimal spacing out for how often you should review something is spacing it out farther and farther apart, longer away from when you originally learned it. So maybe you learn some concepts, and then you review it one day later, and then you review it one week after that, and then one month, and then one year. There are various software packages for creating flashcards and other memory-based prompts, where you see how well you can remember this thing, you give feedback to the system on how easy or difficult you found it to remember, and it uses this information to, with software, space out the cards in an optimal way for retention.
By far the most popular application for doing this is called Anki. The one I use is called RemNote, but it’s really only superficially different from Anki. I think the field of educational psychology and pedagogy has unfortunately attracted some pretty low-quality research, and I’m not sure how much in general we have learned from this field. But I think one educational intervention (or just pedagogical intervention) where the evidence base for it, as I understand it, is extremely strong, is spaced repetition.
Matt Teichman:
So the idea is that you review the material
frequently at first, but then the frequency with which you review the
material starts to stretch out.
Sam Enright:
Yes.
Matt Teichman:
And is it always like that or does the
software adjust the frequency based on how difficult you say you found
the recall task?
Sam Enright:
Yeah, exactly. You can give feedback to the
system that you got the card entirely wrong, that you found it
different, that you found it easy to recall, medium level, and so
on. I’ve been doing this for long enough that I have cards going back
to all of the classes that I took in secondary school, when I
was 15. So the cards from that are so deeply ingrained into long-term
memory, that I only see the cards related to them like maybe once
every four years. It’s on a super long cycle—
Matt Teichman:
—because it asks you less often the more
correct answers you give?
Sam Enright:
Yeah.
Matt Teichman:
Okay.
Sam Enright:
I’ve been getting it correct long enough that
there’s now multiple years between the flash card being reviewed. But
I’m still somewhat regularly reviewing all of the course material for
like classes that I took when I was 15 and 16.
Matt Teichman:
So it tries to pound you on whatever it seems
to it that you’re weakest on?
Sam Enright:
Yeah. Exactly. This feature of spacing out your
revision of things more and more is one reason why these systems are
much weaker for exams, or studying to the test. The thing that it
really excels at is long-term memory, especially of any class you’ve
taken, or specific project you’ve worked on.
Matt Teichman:
Because maybe studying for a test is more
conducive to cramming, where you remember everything for one night,
and then you instantly forget everything forever, because you don’t
have to take the test again.
Sam Enright:
Yes. Precisely. In general, one of the things
I’ve always been struck by, but I’m struck by even more as I get
older, is how little society is set up to cause you to have any kind
of active recall toward the things that you have read, or the things
that you’ve previously done. The average person—even very curious
person that reads non-fiction books—they’ll read this book, and then
never do anything again with that book. They won’t actively try to
recall it; they’ll never try to test their knowledge of it. It
probably won’t really come up again in a way where they’re really
grilled on whether they probably understood it. And similar with
classes: even people who do very well on the exam just take it and
completely move on.
Matt Teichman:
I remember I was at a bookstore once, I saw
the novel Caleb
Williams
by William Godwin, and
I remarked to my friend: “Ah, this is such a wonderful book. I love
it.” And he was like: “Oh yeah? What it about?” And then it was like:
you know, I don’t remember the story of the book at all. I just
remember I loved it for some reason. [ LAUGHTER ]
Sam Enright:
Yes.
Matt Teichman:
It was a humorous interaction. I would imagine
this technique maybe works better for the type of learning that
involves remembering a lot of one-off facts, and maybe is less
relevant for the type of learning about how to solve certain kinds of
puzzles. Do you find that that’s the case, or does this technique
really go beyond just remembering details better?
Sam Enright:
Spaced repetition is certainly not a substitute
for learning a deep understanding of a topic, or developing central
intuitions or problem-solving ability, and I have really never seen
somebody claim otherwise. There’s some people that are skeptics of the
powers of spaced repetition, but I think they’re attacking a straw
man, to a certain extent. Certain characteristics of using these
spaced repetition systems, though, can be a real aid to
problem-solving intuition. One of them is that I always write my own
cards, and I think this is very important, because so much of the
knowledge is you deciding what things that you’ve learned about are
actually important, and trying to summarize them correctly, and
rewriting those cards as you gain deeper knowledge and appreciation of
a subject.
Matt Teichman:
Right. Whatever answer you wrote down to a
question when you were just learning something might not even be as
accurate as could be. Maybe your future expertise would be useful in
redoing those reference answers.
Sam Enright:
Yeah. I have a lot of mathematics cards. Many of
them are definitions, many of them are important theorems, but the
definitions—it’s not even necessarily that the definition itself is
that important to me. It’s more like there’s a lot of meta information
in the fact that I even thought it was important to recall this to
begin with.
Matt Teichman:
Oh, that’s interesting.
Sam Enright:
Like, I only made a card of this definition
because I remember getting stuck on a problem that was related to this
and I want to recall the feeling I had in the moment I was working on
that problem. And that’s why I created the associated card. I don’t
think it would make sense to just read a Wikipedia page on a technical
topic from start to finish, and card what you think are the more
important topics within it. Because the recall should be related to
something that was actually important to you, in your journey of
trying to understand the subject.
Matt Teichman:
You know, this sort of gets to another
question I’ve been having about this, which is: we’ve been talking, I
think, about active recall—I think you used that phrase—where
you’re explicitly asked a question and then you sit there and make
your head hurt trying to remember the answer for it. But I’ve found
often that, for certain kinds of skills, it’s really helpful to have
this subconscious recall.
One example that I would give would be speaking a foreign language. You’re never going to have a conversation in a foreign language by sitting there and trying to explicitly remember the word for this and the word for that. It’s much more of the kind of thing where your reptile brain is going to make you remember whatever you need to say the thing you need to say. So another question I have about this technique is: does it seem to have equal impact on the times when you decide you’re going to actually try to remember something and the times when something comes out of your mouth and you don’t remember why, exactly?
Sam Enright:
I can’t speak to the effectiveness of spaced
repetition for something like language learning because I don’t know
any foreign languages. But I have been shocked by just how many areas
this is a really effective aid toward.
Matt Teichman:
You opened by talking about history, which was
a little bit surprising-to-me example.
Sam Enright:
Yeah, I mean, one area of spaced repetition
that’s interesting to me is: there’s a guy called Andy
Matuschak, who you may know through the
Emergent Ventures group. He has really wonderful writings about his
attempts to learn different fields and is kind of a hero to me in the
earnestness with which he learns new disciplines really seriously,
reading biology textbooks, even though that’s not in any way
professionally—
Matt Teichman:
—and he’s written some books specifically
written in order for you to apply this method, right? Like the
Quantum Computing book.
Sam Enright:
Yeah, yeah. He has an amazing quantum computing
mini textbook called Quantum Country, with
Michael Nielsen; and that is a book
that incorporates spaced repetition into the text, where you put in
your email and there are certain flashcards and questions spaced
throughout the text, and then it will periodically email you after
you’ve read the thing to test whether you have actually remembered
certain key points of information that you learned at different points
of the text.
One of the things that Andy talks about with spaced repetition is the extent to which it can help you find more joy in learning. For me, the frustrating thing about learning a new topic is expending all this effort to just remember something that you think you know deep down, or just forgetting the name of something. It’s very frustrating to just mix up the definitions in math. It feels somehow unfair that you’re getting penalized for this, for example, if you mess this up on a test. And that is precisely the most unpleasant area of learning, and I think spaced repetition precisely helps with that the most. So I found that even though people find it quite alien using flashcards in so many different areas, I actually found it’s made the process of reading about new topics to be dramatically more fun.
Matt Teichman:
Do you ever find yourself wishing that you
could forget something? I don’t know. I think Nietzsche has a version
of this argument. The main reason we’re able to have friends is
because we don’t remember every single fight we’ve had with our friend
in gory detail. There’s certain uses that your memory being like a
sieve can have for you. I wonder if you’ve ever run across techniques
to help with forgetting? Or is that just a ridiculous idea that’s not
worth pursuing?
Sam Enright:
Interpersonally, I think this is a very
interesting concept, that relationships are dependent upon a certain
level of forgetting. Like maybe one way in which marriages can be
successful is: there’s a certain rate of people literally forgetting
about all of the ways that your partner has hurt you.
Matt Teichman:
Or at least you want to forget about what it
felt like to be angry in the moment of the fight.
Sam Enright:
I know there’s some fiction that has explored
that theme. I don’t really see it as much in learning technical topics
and trying to improve your recall. The argument that people often give
is that you need to trust your brain to figure out what is
important. If you don’t, say, take any notes in watching a lecture or
listening to something, one benefit that there might be of that is
that your brain, by necessity, has to prune down all the things that
you heard and compress it into its most important parts.
Matt Teichman:
Which, paradoxically, might be easier to
remember than the raw transcript of everything that takes place.
Sam Enright:
Yeah. There’s definitely some risk that if you
took a lot of studious notes on a certain topic that you would
highlight things that are quite unimportant, or you would have these
relatively trivial details. But I think this argument presupposes a
quite unusual degree of trust in the brain’s ability to be aligned
with your very long-term interests. I don’t trust my brain in almost
any way.
Matt Teichman:
It can sabotage you sometimes.
Sam Enright:
Yeah. If I have to do anything at a specific
time, I put it in a calendar, because there’s no way I’m trusting my
brain to remember that.
Matt Teichman:
Absolutely, yes.
Sam Enright:
I try not to use electronic devices in certain
rooms, or before certain times, because I don’t trust my brain to not
get incredibly distracted by this thing. I find it very odd that
people have this huge level of cynicism or skepticism that, by
default, their cognitive processes will be aligned with their
long-term interests. But then suddenly, when it comes to attempts to
learn more systematically about a certain topic, they will suddenly
have a much greater level of reverence for the brain’s respect for
them.
Matt Teichman:
So in Episode 126 of
this podcast with Agnes and Ben Callard, in which we answered a bunch
of listener-submitted questions, one of the questions we got was: how
do you keep learning philosophy after you’re done with your degree, or
outside of an education context? All three of us, Agnes Callard, Ben
Callard, and myself, immediately jumped on the answer: start a reading
group. Nothing, in our three opinions, at least at the time, is a
substitute for getting some interesting, meaty text, reading it in a
group of people, and then just talking through what’s going on, and
trying to make sense of it collectively. You’ve done a lot of work
organizing reading groups in the UK. How have you found the tactic of
planning reading groups to fit in with some of these other methods
we’ve been discussing for learning?
Sam Enright:
As you mentioned, I have organized a reading
group for quite some time, and I’m generally such a big proponent of
them. I think they’re incredibly, incredibly underrated. I’ve run a
monthly reading group for over the last three and a half years, and
it’s been quite eccentric, the different topics that we’ve
covered. I’ve built up a group that (I hope) trusts my judgment in
assigning really high-quality material that will be interesting.
Some topics that we’ve talked about before: we did a life of Napoleon reading group, we did a industrial history and a history of steam engine group with Anton Howes, our mutual friend, who’s an economic historian. I recently ran a Shakespeare group. We read The Merchant of Venice with Henry Oliver, who also is a great and quite accomplished writer. Sometimes it’s obvious how large the benefits are to learning in person, but even then, I am still failing to price it in, like how much more I learn about a certain text, discussing it with my friends, and also how rare it is to do this in any kind of systematic way. Even the reading groups that exist in the cities that I’ve lived in have been pretty specialist ones related to a particular department in a university. Just a general reading group to try to read some text more deeply seems to be really disappointingly rare.
Matt Teichman:
You know, a question I often get is: is a
reading group the same thing as a book club?
Sam Enright:
I think that the term reading group connotes a
greater degree of seriousness. And also, if you’re looking for really
high-quality engagement with ideas, I actually think that books are
often not the best source material. In the whole time I’ve been doing
these reading groups, I believe we’ve assigned three books, maybe. A
lot of them are edge cases where it’s some paper, but I’ll also
optionally say that people can read an extended book version of
it. The way I look at it, if you can just read a certain piece of
material and understand it perfectly the first time, you probably
don’t need a reading group for that thing. The area where reading
groups are incredibly useful is some kind of abstruse, technical
material where you’re gonna struggle reading it by yourself.
In particular, I’ve picked up this reading group over the years through pretty idiosyncratic means, like meeting them online, or on Twitter. A lot of them are both a lot older than me and more experienced in programming and computer science. So I’ve been utilizing this recently and running a series on classic papers in the history of philosophy, but also history of science and mathematics. We recently read and discussed the Alan Turing paper from 1936 on computable numbers, where he proves the undecidability of first order logic. This paper is really, like, unreadable if you are just trying to get it by yourself.
Matt Teichman:
It’s a good example, yeah.
Sam Enright:
But I found this incredibly helpful set of
annotations that explained a lot of the background related to it. And
also, several of the people that were coming to this group are
professional computer scientists. We just had this really wonderful,
lovely conversation about it for a couple of hours. I read it once,
quite a few months ago, in deciding whether to assign it, and then I
read it again, in preparation for the meetup, and I wrote a series of
discussion prompts, as I always do, to kind of keep us vaguely on
topic. This really was just, bar taking a particularly well-organized
class on this thing, which I don’t really have time to do, and also
would be quite challenging logistically, I don’t know of, what other
better way I could have engaged in to learn this topic.
Matt Teichman:
I’ve found the same thing actually. Often if
you want to study something where nobody’s teaching a class on it,
maybe the class literally doesn’t exist in the world, it’s much easier
to throw a reading group together on it.
Sam Enright:
I will say, if you run a reading group in the UK
or Ireland, please let me know. Email me, because I think there’s
almost no limit to the sort of topic that I would go to a reading
group for, provided it’s sufficiently serious. There’s various reading
groups and book clubs that I’ve seen advertised in different places
for topics like poetry and literature that I would certainly be, in
principle, interested in going to a meetup for. But if they don’t
insist upon people actually doing the readings or being serious about
it, showing up at a particular time, and actually discussing that
thing, then I find the whole thing kind of falls apart.
Really, you’re walking kind of a tightrope in the beginning to set a norm where you probably should not even come if you have not read the exact material that we will be discussing. But also, obviously, you still wanna be open, and welcoming, and so on. So I think once you have a group like this established, it’s quite precious; you should really hold onto this. I don’t know if Agnes Callard’s has been long running or if she herself assembled this. But yeah, I’m super, super pro-reading group. [ LAUGHTER ]
Matt Teichman:
I think the exact format can vary a lot as
well. Gregory Sadler
comes to mind, as somebody who offers online philosophy courses. His
response to the fact that attendance falls off rapidly in online MOOCs
was to literally assign no reading for any of the classes. So
everybody just shows up with zero prep, and then they just read the
thing over Zoom in the online class. So I think there can be a lot of
different kinds of formats, that can be on offer, in the tons and tons
of reading groups everywhere utopia that we’re envisioning.
So as time marches on, more and more of my students have been talking to me about how they use generative AI to supplement the learning process. What are some ways that generative AI can contribute to this lifelong learning process we’re describing?
Sam Enright:
I find it incredibly helpful to ask Claude,
which is my LLM of choice recently, clarifying questions about
anything I’m reading. Tyler Cowen, on his blog, recently, has been
arguing that the value of reading has actually just shot way up, and
people are not really responding to this in the rational way of
reallocating more of their time and attention to this.
Matt Teichman:
Right, because traditionally, the annoying
thing about reading a book is that you can’t ask it questions.
Sam Enright:
Yes.
Matt Teichman:
But you kind of can with generative AI, in
some sense.
Sam Enright:
I just found suddenly, even in the last six
months with the release of Claude 3.5 (and more recently Claude 3.7),
the outer limit of technical material I’m able to read and actually
understand has increased quite dramatically. Like, even just knowing
important terms and results that you can then ask a load of clarifying
questions of, you might actually learn more from your conversation
with the LLM than you would from the original material. Another one
that I think is less widely used is telling your current understanding
of a topic to an LLM, and then asking it to grade it, or saying
whether your responses to it are correct. This is another case where
I’m not sure if it’s the first order effect, or maybe somewhat down
the list of most important effects of this, but I just find this makes
the whole learning process so much fun. I feel like a kid again, in
terms of—
Matt Teichman:
—it feels more interactive, right?
Sam Enright:
Yeah. With the combination of how we have much,
much better tools for supplementing long-term memory now, with spaced
repetition, plus being able to constantly have dialogues about
everything you’re learning, I feel like the value that I’m getting
from just reading a book has increased over 2x over the last year and
a half, which is pretty extraordinary. The rational response to this
would be to—
Matt Teichman:
—2x is many x-s—
Sam Enright:
—yes, reallocate a lot of your attention
toward reading from what you would otherwise be doing.
Matt Teichman:
Are you concerned about hallucinations, or do
you have methods of dealing with that when you spot one?
Sam Enright:
Obviously, it’s certain topics where LLMs are
more likely to hallucinate than others. I wonder whether the optimal
rate of hallucination is actually above zero, like possibly quite a
bit above zero, where if the LLM were only giving completely accurate
responses, it wouldn’t give you as many opportunities to test your
understanding of things or, like, scrutinize arguments.
Matt Teichman:
Oh, that’s interesting. So if you see
something you think might be a hallucination, you could be like: “Wait
a minute. What about this part here?”
Sam Enright:
I mean, I had this exact sensation yesterday. I
was doing this online maths class and asking a bunch of clarifying
questions about a certain question, and I realized that there was a
hallucination in the LLM output. Then I was querying the LLM about
its own hallucination, and then I brought up other ones like
Perplexity and ChatGPT, and I was getting it to grade the other LLM’s
response, and the other two agreed that Claude’s response to this was
a hallucination. Then I got further confirmation of this. And it’s
like—
Matt Teichman:
—yeah, you can triangulate, pitting the
various platforms against each other.
Sam Enright:
My understanding of it is definitely better than
if the LLM had had absolutely zero hallucinations to begin with. I
don’t know if that is scalable, or if that’s just kind of a—
Matt Teichman:
—so just put enough in there to keep you
honest—
Sam Enright:
[ LAUGHTER ] Yeah. I don’t know if that’s
scalable, like whether that’s quite an idiosyncratic example or if
people are finding—
Matt Teichman:
—it seems like it could be a very delicate
equilibrium point to try to hit. Like 70% hallucination, okay; that’s
clearly too much.
Sam Enright:
—yes. [ LAUGHTER ] Yeah. If there is an
optimal rate, it’s certainly a very low one. I wonder whether there’s
a kind of analogy here with the way in some of the most productive
conversations I’ve had, not everything you say in the conversation is
strictly speaking true. Sometimes you say things that are a stronger
version of what you actually believe.
Matt Teichman:
Yeah. People really underestimate the amount
that happens. Professors will just say stuff that’s, whatever, common
knowledge, in the air, but may in fact not be correct. But it’s part
of the classroom session anyway. People remember it.
Sam Enright:
Yeah. I definitely do that with students, like
stating a stronger—
Matt Teichman:
—sometimes the theater of it all just takes
over.
Sam Enright:
[ LAUGHTER ] Yeah. And if you’re arguing in good
faith, this is a really powerful argumentative technique. I wonder: is
there some analogy there with LLMs, um, a non-zero rate of
hallucination actually possibly being optimal.
There’s a really good post from Fernando Borretti that you could link to called How I Use Claude, where he’s talking about all the different ways that he integrates it into his workflows. How his primary use case for it was learning about new topics that he was curious about, and querying science and maths books that he’d been reading. His personal experience was that hallucinations were somewhere between basically not an issue anymore, for particular learning to, if anything, very slightly helpful.
Matt Teichman:
Sam Enright, thank you so much for coming on
Elucidations. I’m going to go straight to my hotel room and start
prepping some flashcards from our conversation right now, which I will
then get quizzed on.
Sam Enright:
[ LAUGHTER ] All right. Thanks so much. This was
great fun.