Why knowledge is not enough — Jessie Munton

If all my beliefs are correct, could I still be prejudiced?

Philosophers have spent a lot of time thinking about knowledge. But their efforts have focussed on only certain questions. What makes it such that a person knows something? What styles of inquiry deliver knowledge?

Jessie Munton is a philosopher at the University of Cambridge. She is one of several people broadening the scope of epistemology, to ask: what sort of things do we (and should we) inquire about and how should we arrange our beliefs once we have them?

Her lens on this is in terms of salience structures. These describe the features and beliefs that an individual is likely to pay attention to in a situation. They are networks that depend on the physical, social, and mental worlds.

In a supermarket aisle, what is salient to me depends both on how products are arranged and on my food preferences. Very central nodes in my salience structure (for example this podcast) might be awkwardly linked to many things (multigrain rice … multiverses).

This is a rare and wonderful thing. Philosophy that is at once interesting and useful.


James Robinson: [00:00:00] Hi, Jesse Munton.

Thank you for joining me. Hi James, thanks for having me. Um, so in your book, which you have forthcoming, there’s a wonderful line where you say there is nothing so satisfying to a philosopher as offering a caricature of some set of positions, labeling it the traditional view, and then establishing that own new position as superior to this mythical view that no one holds.

 and that’s just a beautiful characterization of what I see in, in so many places, in so many papers. and I think it’s perhaps a good starting point for what we’re going to discuss today. So epistemology, it’s this word which I think only gets used in philosophy. so Maybe we want to set out the straw man for what traditionally epistemology has been most concerned with, what it is.

And then, and then we can sort of get into your program of maybe some of the lacuna and things it’s missing. Yeah.

Jessie Munton: Thanks James. Also, I wonder if it’s not too [00:01:00] much of a tangent, if I can say a little bit about like what, what epistemology is, or the oddity of this word that like you say, really only gets used in, in philosophy.

Yes. I think it’s really strange that we don’t have a non fancy term for the kind of stuff that epistemology is concerned with. So I guess I think of myself as primarily a philosopher of mind, so I’m interested in how the mind works, and then epistemology feels to me like the normative dimension of that.

So, in particular in relation to sort of when we’re processing information, so there’s kind of good ways of doing that and there’s bad ways of doing that. And actually we have loads of folk terms that we use all the time that are to do with that kind of normative evaluation. So when we call people, you know, stupid or idiotic or often what we’re saying is that they’re not very good at processing and dealing with information and they’re not doing it in ways that we want.

And then philosophers have these kinds of fancy, more technical evaluative terms, like talking in terms of whether beliefs or other attitudes are justified or whether they’re rational and then, have very detailed specifications of that. But I do think it’s curious that, everybody is familiar with the idea [00:02:00] of ethics or morality.

That’s kind of normative dimensions of behavior at large and whether they’re good or bad in some global sense. I think it’s strange that we don’t have a folk term for for epistemology. so yeah, I guess the, the hopefully not too much of a straw man in epistemology that I call traditional epistemology that I’m reacting to is a vision of it is very particularly concerned with, evaluating belief states in particular, um, and the question of whether or not they amount to knowledge, and thinking primarily in terms of kind of justification, whether beliefs are justified or not.

So it’s a way of doing epistemology that’s very oriented towards, or even sort of restricted to, states which are propositional. So they need to be truth apt, so they can be true or false, and so that we can decide if they’re justified or unjustified. And then my concern, I guess, is that leaves out huge chunks of our mental life that don’t take a form that can be readily understood in terms of a set of propositions, which may be true or [00:03:00] false, which can be justified and unjustified.

So we’ve got these kinds of philosophical terms of art, like justification, and they’ve been developed to apply specifically to propositional states for the most part. But there’s all sorts of things which are going to fall outside of that, but which Intuitively, I take to be relevant to that project of understanding how we can deal with information.

Now, of course, you might have a view of epistemology where its subject matter is by definition restricted to that stuff that it has the tools to deal with. So propositional beliefs and the question of whether things justify the question of whether things are knowledge, in which case you’re not going to see that there’s a, there’s a problem here.

So part of what I’m interested in doing, I suppose, is expanding what we conceive of as relevant to epistemology. And I think I see increasingly that I think there’s a lot of people who are interested in doing that with me. I’m in no way on my own in this project, but I think increasingly there’s a gentle divide between people who think of there’s a kind of subject matter of epistemology proper, and there’s like a very [00:04:00] specific kind of epistemic normativity.

that is not going to interact with other kinds of normativity, it’s not going to speak to them in some common language, versus people who are thinking, actually, I think there’s, there’s, no particularly good reason to restrict the epistemic in the way that it traditionally has been, and we need to develop the resources to kind of expand beyond that.

 I wonder if it’s helpful to give an example of what, like, an area where it feels like, yeah, we probably do want to go a bit more expansive than epistemology traditionally has done. Yeah,

James Robinson: sure. First let me say, I really like your definition of epistemology as the normative aspects of the philosophy of mind.

So sort of what we should do as minds, I guess, if someone had pushed me to give a definition of epistemology, I would say something like the study of knowledge. But then that’s like, well, isn’t that everything? Like, isn’t that physics? Isn’t that chemistry? Isn’t that? Yeah. Um, And so that, that really, um, [00:05:00] narrows it down a little bit.

And as you say, I think it’s not so much of a straw man, certainly in my experience. when I did a little bit of epistemology, it was introduced as, the study of knowledge. What is knowledge? Knowledge is true. Justified beliefs. And then, as you say, there’s quite a lot of focus on, well, are things truth apt or not?

Are they even candidates for truth? Or are they things like, I don’t know, is Trump an idiot? Is that something that can be true or false? Or is it just like an opinion that doesn’t have that kind of status or can’t have that kind of status? and then secondly, what is this justification point? And again, massive can of worms, which, Arguably, maybe too much time has been spent on, maybe not, but there’s certainly other interesting questions, uh, to look at.

 So yeah, I, I really love that definition, but yeah, perhaps, yeah, give some, yeah, maybe give, give some color to that. What, what, what are the kind of questions that come

Jessie Munton: up? So, so, so actually you just saying then that view of [00:06:00] epistemology as a study of knowledge, which I think is quite, is quite widespread and would be a common way of, of defining it.

So, I mean, here’s a couple of things that you, that I think that’s never seemed terribly intuitive to me. So there’s this influential strand of epistemology called knowledge first epistemology, which, um, I think it’s been very productive and very helpful in some respects, but I think I’ve never found very natural, the view that epistemology or that what we’re geared towards doing is primarily oriented around knowledge.

So here’s like a couple of phenomena, which you might think are really important that escape that. So one is understanding. So I think with a lot of people, I think of understanding as a state that’s maybe richer than knowledge. So you can kind of know stuff without really deeply understanding a subject matter.

 and so intuitively there’s a bit more going on when you understand something than just knowing a list or a set of propositions about that particular subject matter. It’s something to do with kind of grasping how those fit together or grasping a set of explanatory relations that hold between them.

And [00:07:00] then of course there is a project, amongst kind of knowledge firsters where they want to understand understanding in terms of knowledge states. So they’ll just say yeah, sure there’s more to it, but it’s just special kinds of knowledge states. So it’s knowledge states that relate these other propositions to one another.

 whereas I think I’d prefer to say that we can get a more natural and a more adequate account of what understanding is, if we let ourselves free from the commitment to doing everything in terms of knowledge states and say maybe there’s something to my mind distinctively structural. We’ve actually, we’ve got a PhD student here at Cambridge, Adham Al Shazly, who’s working on an account of understanding as involving a particular kind of structure to the knowledge states that you have that I’m very sympathetic to.

Another phenomenon that I think influences me a lot is I just think often we’re nowhere near gaining knowledge insofar as knowledge is like, it’s a really high epistemic standard. So loads of the time what we’re grappling with is uncertainty and how we manage uncertainty. And the Knowledge First program is going to say, yeah, of [00:08:00] course, that’s completely true, but the right ways of dealing with uncertainty are all going to be understood in terms of our ability to gain knowledge.

 maybe that’s right, but I tend to think that we might kind of free ourselves up a bit here if we don’t always have knowledge as the key goal state, or if we think of this project a bit more expansively as just not always involving the study of knowledge per se. So the kinds of things which motivate me to want to go beyond the traditional paradigm that’s oriented towards kind of belief states and propositional states in particular, are things like, I think the gathering of evidence is a really good example here.

So one way of thinking about whether or not a belief is justified, which I think’s very, very plausible says that it’s justified if it’s like a portion to your evidence in the right way. So I think that fits with how we ordinarily evaluate people around us all the time. So, if your friend has a load of information about how it would be sensible for their child to be vaccinated and then they maintain a belief that it would be unwise for their child to be vaccinated.

In the face of that contrary evidence, you feel like, look, [00:09:00] I’m not sure you’re really, this isn’t really a rational belief. You’re not thinking sensibly about this. So that insight seems absolutely right. But then there’s this extra bit to the picture, which is, where have you got your evidence from? Or have you gone about?

Gathering your evidence and typically evidentialists have not wanted to engage very much with that part of the picture So they’ve kind of wanted to draw a line that says look you’ve got a body of evidence Now, what is it rational and sensible to do in the face of that? And then that’s going to determine whether or not your belief is justified But of course, there are all sorts of terrible ways of gathering evidence that are going to get you some set of evidence That is extremely skewed or partial.

So if your friend has only gathered evidence about vaccination by um talking to somebody that they met in the bakery or something, then that’s not like a great basis on which to go and form the belief. And so their belief might be perfectly proportioned to the evidence they have, but if they haven’t gone about gathering that evidence in a sensible way, then it doesn’t seem like we should think that their normative status is tickety boo.

 but once we get into the business of evaluating, like, how you gather evidence or what evidence you should have, that’s [00:10:00] like an enormous question that traditional epistemology just doesn’t at the moment have. Lots of resources to deal with

James Robinson: This is really an important point. And it, it makes me think of Bayesianists. This seems very, popular now, particularly in kind of, , the rationalist and affective altruism community that That people kind of term themselves a Bayesian and, that’s great, but what you point out is that you can be a Bayesian and update your beliefs, but if you’re going out and looking for the in the wrong sources of evidence, you can still go down some rabbit hole and probably end up with a horribly skewed set of beliefs.

The other thing that I’m just reminded of is,, Paulina Sliwa, previous guest, I think seems to share this view that just getting to knowledge states, there’s more to our epistemic behavior than that. And, and so she talks about, , you know, moral, inquiry, As a process of moving through different [00:11:00] perspectives, and a perspective is something richer than just a set of propositional beliefs, as I understand correctly, it gives you this kind of set of options, points out what are the ways that you can proceed, um, so she has this example of someone who’s been, who’s been raped, but doesn’t think of it as rape initially.

 and then in coming, going through a process and moving through different perspectives, she ends up with the, um, Not only a different set of propositional beliefs about that, but a very different framework for looking at lots of things and for thinking about what to do next, I suppose. , so yeah, I, yeah, I think I, I agree that just trying to, just trying to hold onto this picture of epistemology as, as coming up with a set of true beliefs firstly misses out on the richness of the end point, but it doesn’t explain either.

currently how we, how we [00:12:00] choose or how we get to develop beliefs.

Jessie Munton: Yeah, yeah, I think that’s right. I’m very sympathetic to how Paulina’s thinking about, about those things in terms of kind of perspectives. I think it’s interesting that Liz Camp as well has some very influential work. I think Paulina’s partly influenced by as well of, thinking in terms of.

a kind of perspective or a frame on a particular question and how that might both guide your inquiry and then also itself be a kind of, epistemic goal state to be in the right version of that.

James Robinson: So yeah, What are we aiming for then? Is it, if it’s not propositional, what is the object of study, perhaps better put, if it’s not a set of propositions that we’re trying to evaluate, is it something like Elizabeth Camp or as Paulina’s drawn on?

that work, perspectives or frames, right? Are these the sort of things that epistemologists should be thinking about?, or are there some other ways of kind of assessing the set of beliefs that we have?

Jessie Munton: Yeah, it’s a really [00:13:00] good question. I’m not sure I have a, complete or straightforward answer to it at this stage.

I suspect that my ultimate answer would be kind of pluralist, that I suspect there might be multiple epistemic goods and goal states that sit alongside one another and that don’t always talk to each other. entirely coherently. and actually, I guess I probably hold something like a similar view about ethics.

 but my ethical views are underdeveloped, so we won’t go into those. But so, I mean, one thing I think is important, and this might sound a bit at odds with some of the stuff I was saying, but I think the resolution of uncertainty is key, whether or not we think of that as amounting to knowledge or not.

So I do think that sort of just in virtue of being an epistemic agent and kind of standing within this rational space, you have a standing investment in sort of understanding where you are located in possible space and narrowing that down as far as possible. And actually, so that sounds like quite a kind of reductive element in our epistemology, but I think it can make sense of quite a lot of things.

So for [00:14:00] instance, why should you pursue one inquiry rather than another? I mean, I think that’s going to depend on lots and lots of normative factors, not all of which are even epistemic. But I think a significant part of it, from an epistemic point of view, is that you want to pursue the inquiries which are most efficiently going to locate you in logical space.

So in general you want to zone in on these big questions which are going to get to the heart of something. So I think part of what we care about is sort of inquiries that kind of fasten on to, to big joints that are going to efficiently narrow down the, the options around you. I think that’s part of it.

But I also think that you could potentially have somebody who’s perfectly located in themselves in logical space and knows exactly how things are and they’re still going to be perhaps, Ways in which we could evaluate agents who’ve reached that state against one another and think one of them is better than the other.

So, yeah, there might in addition be things we want in terms of, an appreciation of the [00:15:00] significance of some information over other information, or even just a kind of prioritization of information that’s going to be, more significant or more important than other information. And what that idea of significance or importance is, again, like, do we want that to be purely epistemic, or do we want it to be something that overflows the epistemic to reflect the kinds of broader practical or ethical concerns we have.

Probably the latter, I think, and that means it’s going to be very context sensitive., so, yes, I don’t know how satisfying that answer is. No, I think

James Robinson: that gives colour on a couple of points. I mean, firstly, this idea of,

if propositions are, equally weighted, it doesn’t give us any clue to Why people seem to set fixate on certain things or, or, why of all the billions of propositions that we, , we believe presumably in our daily lives, only some of them occur to us., So I guess traditional epistemology again, bit of a straw man, [00:16:00] but it.

It doesn’t really tell us about what people do with their beliefs or think about how beliefs are prioritized in their, in their minds.

Jessie Munton: That’s interesting. And I think, I mean, one thing I might want to add as well is we maybe don’t just want to think of, um, beliefs necessarily in this way, but also maybe states that we’re not quite sure if they’re beliefs or not, or states which definitely don’t seem to be beliefs or not. So as well as forming beliefs about the world, I don’t know, you might have,.

suspicions and background hunches and worries and anxieties and things like that, which aren’t exactly beliefs, but they’re sort of in the mix. You might also have desires. You might have a kind of effective state. So, sometimes I think we perseverate on particular beliefs, either because, or like bits of information, because.

 perhaps they make us feel a bit anxious or otherwise maybe we perseverate on them because actually they soothe that anxiety. So the, the, role of kind of the valence of these attitudes is interesting because I think positive and negative valence can cause us to kind of [00:17:00] perseverate more on particular beliefs.

And it feels like there might be ways in which sometimes that’s sometimes you, you want certain beliefs to kind of loom large because they are Explanatorily very important and significant other times you don’t because the influences that are causing them to do that don’t seem like they’re the right kind of influence but it’s a job I think to tease apart like when it’s appropriate for something to kind of loom large and when it’s not.

And again, it might be that if something makes you anxious from a practical perspective, it is appropriate that it would be at the forefront of your mind, but from an epistemic perspective, it might be kind of dominating in a way that’s that’s illegitimate. So I guess part I think you’re absolutely right that this.

This question of like, what do we do with the states that we have, be they beliefs or something else is, is really important and something that again, like epistemology hasn’t said loads about. And maybe some of my interest in that comes from thinking a bit about, um, sort of philosophy of psychiatry. And often I think some of the conditions that psychiatry is interested in.

Involve difficulties that arise in people’s lives [00:18:00] because they are not kind of managing the Informational states they have in the ways that the majority of the population do or they’re doing it in ways that cause particular problems For them and so from in my mind that again this is kind of where the philosophy of mind and the epistemology link up is that we need kind of norms that can do justice to Those kinds of phenomena and help us pinpoint the ways in which things might be going wrong.

Yeah, I was

James Robinson: thinking of, Uh, Mr. Dick in David Copperfield, who has this fixation on the, the head of King Charles, I think. So, he’s always trying to write his memoirs and this guy just, you know,

Jessie Munton: King Charles

James Robinson: keeps on popping up and he can’t stop thinking about him. and his kind of, well, you know, King Charles was beheaded.

So the head is, this kind of gruesome object in, in Mr. Dick’s mind, I guess. , And yeah, this is probably something where Mr. Dick’s beliefs about King Charles may all be on the money, right? And all [00:19:00] his beliefs in general can be on the money, but there’s something wrong, right? There’s something pathological, that’s going on.

And to a lesser extent, we all have these kind of mini pathologies, like things that we, we, we care about too much or care about too little. even though, again, many other beliefs may be correct. , We. Yeah, we might give undue attention to certain ones. , and, and you have this kind of nice way of, I guess, codifying this in, in terms of salience structures.

 so perhaps you can kind of take us through what, what that concept is like and how it can arrange all our beliefs and maybe some of these other things as well, hunches and, and, and so on.

Jessie Munton: Yeah, yeah, I feel like I should be better at giving a succinct explanation of salient structures at this point than I, than I probably yet am.

But, but the kind of idea is you’ve got an individual and they’re located within what I think of as a kind of informational landscape. So there’s all sorts of things that they could, um, attend to at any given moment, and attention is really key to this. Attention is very minimal, so it’s not some [00:20:00] very psychologically involved account of attention, it’s just the idea that when you attend to a bit of information, you um, give over your kind of mental focus to it at that moment in a way that then allows you to process it further or to do other things on the basis of it.

So this informational landscape, some of the stuff that’s in it is kind of internal to the mind of the individual. So it might be exactly like memories that you have or beliefs that you’ve already formed because you’ve encountered certain information in the past. Some of it is is external. So, you know, the kind of books that are in a room that you’re in, you could attend to them by visually looking at them at a particular moment.

Or, you know, if you imagine that you’re in a cafe and there’s a conversation going on around you and there’s another one at the other table, you can kind of tune in and out of different conversations and that’s you attending to one conversation. You can then use that information versus you kind of attending to another one.

So I think of all the like uncountably many things an individual could attend to at any one given moment as attendabilia. We can just So just a term to refer to them all and then the idea is that we could sort of imagine This individual is like a [00:21:00] node in a network and there are going to be sort of links so many links But each of those links is weighted in a way that reflects the probability that the individual will attend to that particular bit of information next and then as you attend to that it’s going to kind of shift and this network of items that you could attend to is going to depend on the particular context you’re in, it’s going to depend on the task that you have at a given moment, so it shifts a lot depending on context, but still we can generalize in ways that abstract over perhaps longer time periods over which you have attended to particular items.

And what we can get out of this, so this is what a salient structure is, is a kind of, It really, I think it’s best understood as a kind of model of the mind and the things it can attend to as a kind of network and then some significant aspects of the mind and how it operates are going to emerge at the level of the topology of that network.

So, for instance I really like your David Copperfield example. , so. For that individual, that, that head is [00:22:00] becoming a very dominant node in this network. That it doesn’t matter what task they’ve got or what context they’re in, their mind is continually kind rerouting through that. And then that lets us describe the ways in which sometimes that might be appropriate.

Maybe that is Helpfully allowing them to act in the world. Maybe it’s helping them resolve uncertainty. But sometimes you can get this kind of calcification into structures that are organized around a particular topic or particular preoccupation that are not helping you do that. And I think you’re absolutely right that this is all on a kind of continuum.

Like we’re all doing some of this all the time. So In the book as well I use the example of my oldest son who’s like very preoccupied with football at the moment and like so many topics of conversation come back to football so if you’re talking about countries like his knowledge of countries is really organized around his knowledge of football and that’s kind of harmless enough but there probably are ways in which it’s blocking him from accessing new information that might be rewarding and along various axes to him.

 and then, you know, maybe you’ve got relatives in your life as well who might have, you know, particular beliefs that they really perseverate on. so I think [00:23:00] this could be quite a powerful way of thinking about., sort of the mind in a way that that frees us a little bit from that propositional bias that I was talking about before, because it could be that all of an individual’s beliefs are perfectly accurate, but the problem is just that they’re really perseverating on a small subset of them.

or, you know, you could have agents who score equally in terms of how much knowledge do you have about the world, but one of them has a salient structure that feels like it’s organized in a way that lets them kind of function better, lets them function better as an inquirer as well, and as somebody who wants to learn new stuff about the world, and the other one has a salient structure which isn’t really letting them do that., and so it’s putting, it’s sort of shifting our focus a bit to kind of the, the organization of information., and, and it’s very, it allows that that can arise from a huge number of factors. So some of them might be again, like factors that are internal to the individual, but some of them are just going to be the social context that you’re located in and the ways in which that kind of compels you to attend to some information.

So adverts are a really obvious example of that. Like you don’t have a choice about what adverts are [00:24:00] in your environment as you walk down the street, but that’s going to change what you’re attending to simply because it’s, Giving you stuff to attend to right there. And often it’s doing it in a way that’s deliberately designed to capture your attention with, you know, bright colors and pictures of attractive people.

Yeah, I,

James Robinson: I think that is a wonderful succinct, explanation of what a salience structure is, , or as succinct as, as one can give. , yeah, let me read back some things to make sure that,. I’ve got it correct. So, it, it kind of encodes the accessibility of, of information to us. And that involves both, the physical world and one’s internal world, and even the, the social world, which is maybe some mixture of internal and physical worlds.

 So it’s a quite a, quite a complicated thing, but I think that’s justified because the information that I’m accessing right now is some. Yeah. Influenced by so many things that my computer’s right in front of me. I’ve got my headphones on. I’m listening to you, but there’s the sky outside and I have all sorts of, you know, beliefs and memories about, that and, and about the [00:25:00] objects in my room, which are based on my past history and, and, and, and so forth.

So my dispositions to attend to, or be distracted by, or focus on certain things, however we want to determine it have lots of complex, determinants., And I want to make sure I really understand what the structure is. Is it, so I am sort of located at a node in the structure.

I’m at a point where all these beliefs can sort of, uh, connect to me or, not necessarily beliefs, but just things that can grab my attention, come on my attention, , as if I start to think about something, for example, if I start to look at the sky now, does that move where I am in the structure?

 or does it indeed change the structure , is the structure something that’s kind of relatively. [00:26:00] stable moment to moment. .

Jessie Munton: Yeah, then that’s an excellent question. And it’s one of the things I think I’ve most struggled with is that, um, intuitively, What you’re likely to attend to at any given moment is in so much flux that it feels like it’s hard to identify something that we could call a sort of stable structure behind it.

So the, the, like, specification I give at the moment is I’m thinking of it in terms of a function from tasks and contexts, so a set of attendabilia and, probabilities that you’ll attend to other, To those attend Belia so that the the weighting of the links between the nodes matters a lot because the nodes the links are so There are so many of them that on their own whether or not you’ve got a link doesn’t tell you very much because you can Move in so many directions so then, as you attend to something new, we can think, I think, as long as the task and the context stay relatively stable, if the salient structure is [00:27:00] staying roughly the same.

One thing I think I want to say is that, I mean, , you have changes in what you’re liable to attend to at any given moment. Not all of those then translate up to a change in the salient structure. So a change in the salient structure, I think we want to reflect something that’s like a bit bigger than that.

 something that’s like a bit more entrenched or that emerges over time and in some way or other. So we need to build in like the ability for, you know, what you’re likely to attend to is shifting all the time. But like what we consider the salient structure is an abstraction from that in a way that not every difference in the probability that you attend to something is then automatically going to translate to a change in the salient structure.

James Robinson: Yeah, that makes sense. And I think it one of the examples you give in your book is of Subway transport systems. So like the tube in London, and they are topological structures. Whenever you look at a tube map, it doesn’t give you the distances, but it just shows you how different stations are [00:28:00] connected.

And this is something like, yeah, a kind of a slice of a salient structure in time where. If you’re at one station, you know, you’re attending to something, it makes it easier to attend to other things. So I suppose we can think of the salient structure as not just encoding exactly what one is able to attend to at.

You know, a certain point, but it gives some idea of as the, as the user’s attention wonders, how that will change what they’re likely to attend to next. Just as with the tube, if you’re at , King’s cross, it’s very easy to get a angel., and if you’re an angel, you, you know it’s somewhat easier to get to, I don’t know, Finsbury park or something. So yeah, your, your point in that structure gives some clues as to what the likely next places that you might end up on.

Jessie Munton: Yeah, that’s exactly right. And then the tube is a [00:29:00] much easier example to work with because it’s very stable because it relies very strongly on this particular physical structure on DeLarga, which is we have to build tube tracks and then we get physical trains and we put them on the tracks and off they go.

And so then the kind of options are very limited compared with your mind and what your mind can do. But we can imagine sort of thinking of a transport network that scales up from that, that doesn’t just include the tube but maybe it’s also including buses and then maybe it’s also including kind of cars and it’s also including individuals and their ability to walk to different places.

So now we’ve got like loads of different factors in the mix that are going to determine how accessible different locations are in the city to different individuals. But we can kind of abstract over all of them and then still say some things about how easy it is for a person to get from x to y. Given this like very complex structure, which is emerging from multiple causes.

So I guess that would be closer to the case of the mind because you’ve got this like multiplicity of causes and you’ve got a lot of flexibility and shiftiness in there, and you don’t just have sort of one physical structure [00:30:00] that is determining what these accessibility relations are.

James Robinson: One of the things I like about salient structures, is that it, it allows you to think it allows you to bring to bear some of the machinery from, network science.

So my objection is to say, well, salient structures are nice things, but why, you know, what do they offer me in terms of. Thinking in terms of perspectives or frames, right? there’s some stuff that’s foregrounded and, and so on. Why do I care about thinking about links between things?

But when I actually think about in terms of links and nodes. We can draw on this, this, this fairly rich, set of, findings within network science. ,So yeah, perhaps kind of give us some ideas of, of the

Jessie Munton: utility of that. Yeah. And I think we have to be, we have to be like cautious here.

So, in a way, networks are really, really cheap. All you need for something to be a network is that you have nodes and you have some kind of. relationship between them. And so, I mean, we can treat almost anything as a network. And then the question [00:31:00] is, what usefulness do we get out of doing that? Or like, what, what are we gaining through thinking in those terms?

For me, a big thing that we gain is that we can think then more structurally about the mind. Like I was saying, we can think in terms of the sort of topology and how that might matter. And so maybe a good example for for my purposes for this is to think about prejudice and the different forms of prejudice can take and how I think thinking in terms of a network might help us capture some of those nuances and differences.

So there’s an extensive debate in philosophy around like what prejudice is and what it takes to be prejudiced. And there are important ingredients in the mix here that aren’t the primary focus of my concern there. So you might think you have to really dislike people in order to be prejudiced against them, or you might think you have to behave in certain ways.

But there’s also, accounts that say, well, the key thing about prejudice is you hold certain beliefs about the group against which you’re prejudiced that are kind of nice beliefs to hold, their content is like negative content. But I think just thinking about that kind of what’s the, what’s the mental [00:32:00] attitude you have to have, that side of the picture, I think there’s lots more sorts of mental attitude that can count as a form of prejudice that don’t necessarily involve kind of explicit negatively valence beliefs.

So, Or, or that do, but we might want distinctions between that. So here’s your kind of classic case would be you’ve got a negatively valence belief. Not only that, but it’s also a kind of, it’s serving as a big hub in your network, so that lots of traffic gets routed through it in the way that we were saying might happen with the David Copperfield example or the football example.

So it’s important that it’s not just that you hold this belief, but that it’s become this kind of important pivots in how you’re thinking about the world or how you’re interpreting new information or how you’re deciding what other information to search for. Like lots of stuff is being kind of rooted through this really core belief at the other side.

We can imagine a situation in which it’s not actually that you hold. negatively valence beliefs about the group in question, but it might just be that, um, kind of [00:33:00] not obviously negatively valence or maybe even positively valence beliefs also have that role of being a kind of hub or a sort of connector of some kind.

So, if you just think it’s so wonderful that women have these caregiving abilities and that, you know, when your colleague is pregnant, you are just. I think that’s just brilliant and that’s like the main thing about your colleague at the moment is that they’re pregnant So that’s kind of when you’re thinking about your colleague That’s become this sort of hub that lots of stuff is being routed through again That’s a problem because in a work context your colleague doesn’t just want you attending to their pregnancy They want you to attend to what they’re up to at work and these other professional concerns that they might have Another way we could think that prejudice might arise might not even be that there’s like some one particular belief that things are organized around, but that, when we step back and we look at these structures of accessibility and how you’re inclined to move between particular beliefs.

Race or gender or whatever category it is ends up looking like a kind of organizing principle in [00:34:00] how you are attending to other things. and so that’s very broad and it might take a lot of different forms, but again, it’s like a something that might look like a form of prejudice, but that doesn’t necessarily involve you having some explicit propositional belief of the right kind.

 Yeah, I think

James Robinson: that’s, those are all really wonderful examples. So the colleague example, one might have lots of true beliefs that, one’s colleague is very good at their job and, um, they did a great, did some great work on this last project, but if just. You fixate on, oh, they’re such a great mom, right?

There, there’s a kind of prejudice in, encoded there. and then, yeah, similarly, if I, I suppose if, as you say, if we organize our beliefs along, racial lines, um, Yeah, we miss out on so much of the nuances of other people if we’re just thinking of them. Oh yeah, they’re, they’re from [00:35:00] this culture and they, you know, I know lots of things about this culture and I’m going to foreground all my knowledge of that when I go and talk to them and I’m not really going to care about their individual, cares and preoccupations.

 I may be on point on all of those things in terms of having only being correct in my knowledge, um, but just the way that I’ve organized it, , leads me astray. , and you’ve mentioned here, yeah, the idea of hubs, which is kind of bringing in some of this, yeah, machinery from, from, from network science.

 One of the points that I enjoyed in your book was you pointed out that we don’t want, we don’t want a kind of completely random, salience structure and that that’s, you know, we understand what randomness looks like in, in, in, in network science. You just kind of have every node connected to other.

Every other completely kind of fairly randomly. And sometimes one meets people who have that, seem to have a bit more of that structure. They will just kind of make these really wonderful random jumps. And actually that [00:36:00] can be a really useful thing in society. I kind of feel like maybe that’s a kind of creative trait.

 but then there’s people who have, you know, very, very focused areas of knowledge. And, uh, when they think about, uh, I don’t know, cars or something, they. They just think about the machinery of it and the mechanics and how it all works, which actually might be kind of too structured, right? They might need more links between that hub and other things.

They might need to care about, I don’t know, the environmental effects of cars, for instance. so you need some kind of balance between having, , very concentrated, isolated, nodes, but also having connections between all your different areas of knowledge and yet not too many connections that everything becomes entirely random.

And there is this kind of notion of small world networks, which one can think of, you know, probably the hubs and spokes models of, of, airports is maybe a good example [00:37:00] where, kind of a regional airport will collect, connect to, um, a hub. So, the East Midlands Airport will connect to Heathrow and so forth and a few others, but it’s not going to fly you to Boston or New York.

 but then you can go through these big hubs, which will take you all across all the way across the Atlantic, and then you’re kind of into a new area of the, of the network where you can take lots of, short hops if you want to. so yeah, I think that, by the way, I love your comment that we need to be careful about applying network science to everything.

, but I, I think that it is, it is useful here. , I don’t know if there’s any kind of more. Mathematical that we can get with it or, or, or, or if just this kind of level of analogy is, is useful enough.

Jessie Munton: Yeah. Like, um, sorry, there’s lots of things I would say in response to what you’re saying, uh, yeah, so, so the small world network, I think it’s really helpful to bring that up.

So a big thing that I think, I mean, the [00:38:00] normative framework that I’m then applying to salient structures, we kind of want two things out of it. One is that you want. information that is relevant to be readily available. And I have a particular kind of epistemic understanding of what relevance is. And the other thing you want is flexibility.

So you don’t just want, like you’re saying, you don’t just want information that’s relevant to one problem. You need to be able to move between that. And the key thing about small world networks is they tend to combine some of those virtues. So you can move easily between kind of centers of information about a particular topic.

I think, I think another good example of small world networks is like. University towns, I think, definitely have this structure. So, within the university, they’re very, very, there’s a lot of very dense connections there, so everybody knows everybody else. Within, uh, in Cambridge, within Addenbrookes Hospital, there’s a lot of connections there.

Everybody knows everybody else. And then you don’t need to have very many people at the university who know somebody at the hospital in order for you to be able to quite easily connect any given individual with, at the hospital, with any given individual at the university. You know, you can add in some schools and you’ve got a kind of a [00:39:00] particular structure of a community there.

 yeah, in terms of getting, mathematically precise, so I feel at the moment like I want to set up a kind of bear trap outside the math department and wait for somebody who does a lot of network science to come by and fall into it and then I can take them away and bully them into, helping me do some of this stuff in like more mathematically precise ways.

One thing in particular that I want to make sense of a bit more at the moment is like how do networks aggregate across individuals to produce social salient structures? And that’s something that I would really appreciate a network scientist helping me. out with. so I’ve been thinking recently about, um, ignorance and forgetting, and I’m wondering about the ways in which ignorance and forgetting at the individual level scale up in such a way that we could talk meaningfully about a social group forgetting something.

I think we can do some of the work we want to at the individual level, some of it, not all of it, in terms of salient structures. And I’m wondering if we can talk more meaningfully about group salient structures. But for that, I think I do need to get more into the nitty gritty of the, of [00:40:00] the maths of the network science.

James Robinson: My last conversation was with Geoffrey West, who is a physicist who look, who’s, who’s looked at the importance of, of networks in terms of, firstly in terms of biology, but then in terms of cities, companies and other kinds of emergent entities.

And he, and he has like a managed to mathematize it really well with his, his colleagues., and actually leads me to. A slightly tangential question. One of his findings is that in cities the kind of pace of life increases and it’s actually really, it’s incredibly consistent how across a large range of features it increases in the same way.

So for example, the number of patents per head, uh, doesn’t just. Double if you double the size of a population, if you double the size of population of a city, it goes up by, uh, 2. 15. So you get this kind of additional 15 percent for free. [00:41:00] Uh, and the same with a number of restaurants, right? You don’t just get, um, more restaurants as you get, as you, as you double the size.

 you get more restaurants per head by 15 percent and so on. And so as you scale up cities, double and double and double again, you get these massive, kind of benefits in certain senses. You also get, problems. You get more crime per head. You get more, um, AIDS cases, probably more COVID cases,, at least a faster rate of them. And where this all leads me, one of the places it led me to thinking was, all these You sometimes just seem to suggest that salience might be a kind of zero sum game, which makes sense, right? If something is more salient to me, then it’s got to be drawing away my attention from something else.

But I also wonder, thinking of Geoffrey West’s work, if, if actually the sort of, um, The amount of salience itself can scale as well. Like in a city, if I’m in Times [00:42:00] Square, I’m getting so much stimulation. Whereas if I’m just sitting in, I don’t know, an empty room at home, I, I may actually think about fewer things.

 there may be, yeah, just less for me to attend to. and I, I don’t know if that’s something that you, you, you thought at all.

Jessie Munton: That’s really interesting. Yeah. I guess one way of reflecting that might be I mean, so I think I, I do think that salience is a kind of nil sum game that any that, well, it’s attention that really is like attention to one thing comes at the cost of attention to another thing.

but there might be ways of still accommodating how it can seem as though in some cases. Your attention is spread over a wider variety of things, or, um, Yeah, that’s interesting the idea you might just have more of it in some situations. I mean, I guess having just more of it is compatible [00:43:00] with it still being kind of a nil some thing.

 you just have a larger quantity to distribute over the different things, but attention to one thing is still coming at the cost of attention to another thing. Um, but another thing we might want to include in here is maybe it matters. Um, Not just what you are attending to but what you’re kind of likely to attend to so you might think that attention Cut or salience at least certainly comes in degrees And you might think that sort of attention can do something similar.

So this is like a debate within people who work on um sort of attention within cognitive sciences, is it just like there is attention and there is not attention or can you also be doing something that’s like a bit of attending at any given moment, in which case maybe you could spread that bit of attending over multiple different stimuli, uh, in a way that kind of reflected a little bit what Perhaps you have in mind?

Another thing that might be happening in these very, like, informationally dense environments is just that we are switching our attention an awful lot. So at any given moment, we’re just switching between attending to these different stimuli. And then that creates a kind of effect where we [00:44:00] feel like, I mean, we are, in some sense, attending to a lot of things, but in a less sustained, a more fragmented way.

James Robinson: It’s one of the things. That probably illustrates why this is going to be so tricky to, to actually cast in mathematical terms. But I, I really like your idea of casting a bear trap, um, as I think for the right person, this would be a fascinating problem, but yeah, these networks seem so dense and, you know, possibly dynamic.

Um, and so there’s this kind of interaction with how long you attend to something as well. Um, Which, yeah, so they’re kind of, that might have some effect in terms of like how, what I’m trying to think is like how you could connect this to kind of outcomes in, in, in the real world. Like, um, you know, could you, you can certainly look at [00:45:00] the physical repositories of knowledge, like libraries and stuff, and, and even some of the kind of virtual ones, like the internet and so forth.

Like those things, you should be able to get some kind of model of how their structures much, much harder, obviously, to do the internal models. Um, but even those external models, like, I guess the problem is the external models kind of mean nothing unless you have, unless you mix in these internal models, because you know, one can go into a library and just ignore all the books and focus on the tables, right.

Or, you know, um, so yeah, it’s, yeah, it’s a really, it’s a really intriguing. And, uh, and tricky problem, I think.

Jessie Munton: Um, yeah, I think that’s right. And sometimes we might want to draw, um, so for the most part, I’ve just been amalgamating all the different causes that trigger your attention to go in one direction or another together.

But for certain purposes, we might want to separate those out. So you could have two people who just never, ever read, um, books by. Africana [00:46:00] philosophers, and in one case that might just be because there aren’t any in the library, and in the other it might be because there’s loads in the library, but they just don’t care, they’re not interested.

Those seem like significantly different states, and so for certain purposes we might want to say, what’s the subset of causes here, like is this something that’s Due to an environmental restriction that in some sense, we don’t want to include those nodes in the network. Um, or is it due to something that seems like it’s coming from some kind of internal motivation?

I think, I think there’s going to be all sorts of distinctions like that, which matter for various purposes, which aren’t immediately reflected just in the salient structure itself. Yeah,

James Robinson: I think that brings up an interesting question, which I know you’ve looked at, which is, you know, Well, to, to what extent we can change these salient structures, you know, particularly if, if one’s beliefs can all be correct and, and, and yet one’s still biased because perhaps, you know, it could just be simple, as simple as, uh, as your Africana suggestion, uh, or example suggests, it could be as simple as one having zero interest [00:47:00] in a particular cultural, um, group or, um, gender even, right?

Um, and Yeah, like, how do we, what do we say

Jessie Munton: about that? Yeah, it’s a good question, because a lot of this stuff feels like it’s very involuntary. And there’s a tension, I think, too, between recognizing the role that you know, social context play in determining what you attend to and then wanting the individual to take responsibility for some aspects of their salient structure, because you can’t control your physical and social environment always.

so in terms of how we change things, I mean, I, I think it, there is some stuff I think that the individual can do that you can try and become a bit more aware of how your attention goes and what you’re attending to. You can also be more, Mindful of what kinds of, um, material you’re kind of exposing yourself to and the ways in which that’s influencing how you’re then attending to people.

[00:48:00] So, um, You know, I grew up in an age where there were lots of magazines that had information about celebrities in where they’d like literally be kind of circling women’s cellulite in them. That’s definitely going to influence your salient structure. It’s going to influence how you attend to your own body and it’s going to influence how you attend to other people’s bodies.

So there’s probably a good reason to avoid exposing yourself to too much stuff of that sort. I mean, there’s probably lots of reasons to avoid it. But one reason is it is going to influence your salient structure and you can take some kind of responsibility for that. I think there is like lots of stuff that you can do at a broader social level that would influence people’s salient structures for good or for bad.

Um, I was really interested, you know, having come through well over half a decade now of reading a lot of children’s books. You can really see in that how we encourage certain patterns of salience by drawing children’s attention to certain basic aspects of families and how they’re set up. Children’s books are like massively, massively heteronormative.

Um, And also like in the kinds of [00:49:00] gender roles that are assigned to women and to men and lots of the books that you come across and, um, I think there’s lots of room to kind of be a bit more conscious about, you know, you might not think it’s something you’re explicitly telling a child, but you’re setting up a structure of kind of anticipations that they’re going to have when they encounter families in the real world when they encounter people in the real world and sorts of things that they do.

And it’s tricky because some of that stuff just just reflect. you know, true generalizations about how the world currently is. Some of it I think goes kind of beyond that, um, into, um, setting up these kinds of like very basic expectations or sort of sense of this being a fundamental kind of structure in the world.

I think there’s stuff to be done. to be done there. I also think it’s maybe something to think about when we are doing implicit bias training is it’s maybe helpful to talk explicitly in terms of salience and what’s going on at that level. Um, and also when it comes to thinking about what’s a [00:50:00] responsible way of communicating with statistics.

So your statistic might be accurate, but what What impact is it going to have beyond the accuracy of the statistics? And what stuff are you making salient by pulling out a particular statistic in a particular context? Um, there’s something that you need to be thinking about when you’re thinking about am I using this statistic?


James Robinson: Yeah, I think those are, those are lovely. Those are lovely examples. Um, I find it kind of ironic. This is the second, um, video. podcast that I’ve, I’ve, um, recorded. I mean, probably a lot of people will be listening to this just through headphones and not watching it. But, um, you know, one of the reasons I’ve decided to add video is just that, you know, it’s such a, people love YouTube.

It’s such a good way of finding podcasts. I kind of, in some ways, I like the purity of just speech. Because it’s removing some of that salience. Like you were saying, like, you know, the, the human body [00:51:00] is like a massive object of attention for every human. And, um, we just don’t seem to be able to get around that even often when we want to, right.

Even if we just want to listen to ideas, like we like to look at faces and. Uh, and you know, maybe that’s not a terrible thing, like the face does express some meaning, but I think it can often be a distraction. I’ve got thinking recently about on the augmented reality.

So we have the Apple vision pro, which is, you know, could be used to reinforce salient structures to, To calcify them, to use a term that you, you so aptly earlier. So for example, if someone’s really into football, like your son, right, he could just, he could wear an Apple vision pro and it could sort of football ify everything for him, right?

It could repaint the walls in Chelsea’s colors. He could, graft your, uh, you know, some famous [00:52:00] footballers. I don’t know enough about football. Let’s even mention a Chelsea player’s name, but you can graft their face onto his friends or something like this.

I have a vision. So, um, but these, these technologies could also be used in the other direction to correct for some of the, the things that we find too salient. this is like very sci fi. I don’t think. This is going to be an early use case of these technologies, but one can imagine people giving job interviews.

And, um, Kind of redrawing all the candidates to look the same. And, and so, um, we, we lose some of the biases that, that we, that we, we find so hard to shake off. And actually I think it is, you know, it does seem just very hard to shake off. So much of the, uh, or to, to, to adjust salient structures in so many ways.

I think there’s a kind of hysteresis firstly, like there’s lots of evolutionary reasons why we find certain certain things salient, [00:53:00] but it occurs to me that a chance comment can update one salience structure in a way that’s very difficult to update a friend of mine. Said, oh, high arched eyebrows, Joseph Fritzl’s got them, so has Gary Glitter.

And now, like, I don’t think he believed that they were, like, evidence of evil, and I certainly don’t.

Jessie Munton: Now when I look at high

James Robinson: arched eyebrows, they’re just slightly more salient for me because of that. Yeah. Yeah. And I can’t really unlearn that. Um, yeah.

Jessie Munton: Um, yeah, I mean, I think, I think, and you’re absolutely right about like, I dunno, philosophers have a long history of sort of denying our embodiment and because the framework that I’m offering is very much focused at the level of kind of information.

I worry that it feeds into that a bit and you’re right. They’re like our physical embodiment makes certain things salient to us. And it’s an aspect of other people that is inevitably salient to us. And. [00:54:00] Um, you know, you don’t want to end up in some situation where it feels like you have to bash yourself over the head with some hammer to change that in a way that you can’t.

I think one thing that’s really helpful is that you don’t necessarily have to change the salient structure, but you can have a metacognitive awareness of the role that it’s playing and how you’re thinking about things and what you’re absorbing from the situation. So, of course you’re going to notice what the people that you interact with look like, but maybe you can kind of challenge, you know, how you’re responding to that.

Um, thinking about your example of the, of the glasses, like, so I’m on Instagram and I follow like a bizarre selection of accounts. Some of them, I guess, are quite sports oriented, and there’s definitely a kind of presentation of the physical body there that, that is sort of valorizing in explicit and implicit ways of a kind of very athletic physique of a particular fairly circumscribed kind.

And then I follow some photography accounts at really push back against that and are quite invested in, encouraging and appreciation of the physical beauty of a much wider diversity of bodies. I think it’s really valuable following both of those. I think following both of them makes me [00:55:00] more aware of what’s going on in each of them.

 it’s interesting that I guess in that case, one thing I want to say is I feel like the ones which Encourage me to attend to the ways in which a wide diversity of bodies can be very beautiful. I feel like they’re putting me in touch with reality in some substantial way, but I don’t yet have a good way of saying, like, what, when does a salient structure do that?

And, you know, when does a salient structure, it feels like it’s akin to being misleading. But what do I mean by applying that term misleading in the context of a, of a salient structure?

James Robinson: Yeah. I, I think that’s a much healthier way. Uh, instead of trying to blinker ourselves with, um, apple Vision Pro or some other headsets.

Just to add some richness to our information diets, seems much more positive way of, of, um, uh, combating this. And as you say, just being aware of even having these salient structures, I think it’s really useful. And again, the, the physical embodiment of them, you know, changing the books that you have, or, we’re just being more mindful of [00:56:00] the programs that you watch.

I think those are all kind of useful learnings for me from this. I guess. You know, it’s nothing completely new, right? Everyone knows that it’s important to like, choose their information diet. , but, I think it, it does give, it does give an inkling of how,, the items in our diet connect to each other or the sort of the items in that our universe of information connect to each other.

Another thing I’m interested in is, search engines, which I know you’ve, you’ve looked at , and

one thing I like about search engines is that they’re these kind of public objects. So they’re pretty much the same for everyone. So they kind of create some. Baseline similarity in salience structures, sort of similar to how Walter Cronkite, everyone used to believe what he read on the news, right? At least [00:57:00] search engines, if I type in something and you type it in, we’ll probably get a similar set of results.

There’s some magic in the background, which may be depending on location and other things, it might, reorder. Or, or certainly if we were, we’re typing the same query in Spanish. We’d get a, a different set of results, but by and large, um, they’re fairly fixed structures. And they’ve had some demerits, but I think I’d overall say that they’ve brought benefits to society, search engines, social media I think is different because.

That is, targeted to each other very individually. So if I click on a link in Google, it doesn’t update its algorithm, for me alone. It doesn’t, I think, I’m not an expert in this, but I don’t think it’s more likely to, let’s suppose, I don’t know, I search for like fluffy dogs and I click on a poodle. I don’t think it’s more likely to think that I’m into poodles because of that.[00:58:00]

If I do the same in a social network, The algorithms do try to update and figure out what is the stuff that I’m really into and change my information diet, like tailoring it to me. And that seems like it would, have the effect of, I guess, kind of narrowing or reinforcing certain aspects of my salient structure.

Probably those aspects which, are Most attention grabbing in some way, or most easy to manipulate, I guess, like social media has these has this kind of meta goal of just making us all more manipulable, right? If it it’s primary goals are to, I know, sell stuff to, um, advertise things, etc. And to make us take certain decisions.

And it’s meta goal to do that is, Make us more easily to manipulate. So, um, encouraging us into a very, very kind of niche set of views makes it [00:59:00] easier to identify what things we will find salient. So yeah, I. I’m putting a lot of stuff on the table. I’m going to put one more thing, which is, I think the next evolution in the, the way that we access information is large language models like chat GPT.

,And I think that could go either the way of search engines or the way of social media. So again, they, there is for, for chatGBT, and similar, there are competing, , benefits or competing, how to put this,

, it gets points from us from tickling our, cognitive biases, right? And, like if it presents the information to us that we really find super salient and we just, really love digging into. [01:00:00] We’ll use more of that tool. On the other hand, it does also get points from us from, helping us develop better models of the world.

Cause at base, we are epistemic agents use a very philosophical term. We’re all interested in having a good, broad understanding of the world., At some level, it just may not be that in each individual task, that’s what we’re looking for. Like we, we might just be looking for fluffy dogs, right? , But.

They, they ought to see some benefits from encouraging us to broaden or, or, improve our salient structures in certain respects so that we do get better epistemic outcomes., so I guess, my, my feeling is like from an individual visit, one might like to have one’s cognitive biases tickled by an LLM from across lots and lots of different visits and lots and lots of different queries.

One might learn to appreciate that it’s actually better that sometimes the LLM throws in things that we hadn’t thought of at all, [01:01:00] that we’re just not aware of and kind of broadens our knowledge base. Um, but I don’t really have any intuitions on which of these models might win out. Um, so yeah, a lot of, I plunked a lot of ideas on the table.


Jessie Munton: Yeah. James, I’m interested in your confidence that search engines aren’t providing us with a more personalised service that might mean that individual’s experience of them differs. Do you have particular grounds for that?

James Robinson: Not very good grounds, um, but I would say that they’re less, certainly they’re less tailored than social media.

I definitely agree with

Jessie Munton: that, but I think it’s interesting that it’s, search engine companies are not very open about how much, you know, personalization there is at that level. And I suspect from the readiness with which Google gives me stuff to do with philosophy, That it’s not doing that for everybody.

But I might, maybe I’m wrong about that, but, but I think it’s significant and it’s something that there ought to be a lot more transparency around than there is. And in fact,, I think there ought not to just be transparency around it. I think you ought to be able [01:02:00] to just change settings or, you ought to have a slider that kind of says to how, how wide open do you want this set of results to be?

And how much do you want it to be tailored to the particular things that we think that you’re interested in at this given moment? So. In the context of search engines, so I said before that I think a really important thing we want from salient structures is we want flexibility. We want to be able to do lots of different things.

We want the information that we need in order to be able to do those things to be available to us. I kind of think the same about search engines that like, You want lots of different sorts of information to be available. We also intuitively want that to be information that is relevant to the thing that we’re searching for.

But there’s this big question, which is like, what does flexibility really involve in the context of, um, search engines and like, how open do we want that informational landscape to be? And I think my kind of judgments about this is a little bit all over the place. So part of me thinks with you that we want this landscape that’s quite open and we want this to be a shared tool that.

people are getting the same experience of, are getting the same [01:03:00] information from. That seems kind of like, well, transparency seems important here and that sort of seems like it’s part of transparency. At the same time though, I worry that that is going to flatten things out in a way that is to the disadvantage of people with more niche interests or whose informational concerns are not those of the majority.

So I feel like if you’ve got some very particular subset of interests and you’re going on Google, then Yeah. You want Google to give you stuff that’s relevant to that particular subset of interests. I also have the sense that we want, Google other search engines to make sure that we’ve got a really wide variety of information available that as you say, exactly my like challenge, the sorts of biases that we’re coming in with, all the kinds of views that we’re coming in with.

 then there’s another thing in the mix, which is like, to what extent do you also want it to make you aware? Not just of information that’s good and accurate, but information that’s not great, but the other people in your community are probably accessing because they find it a satisfying way of.

consuming the news and things like that. So I think that search engines have this kind of meta role to play a lot of the time where we don’t just want to know about the topic, but we want to know about [01:04:00] what kinds of information are available about the topic and who they’re available to. And again, that’s maybe a parameter that you should just be able to vary.

And that would be an amazing search engine. I think that it was like, do you just want the most accurate information or do you want to know what most people are learning about when they learn about this particular topic? So then when it comes to like large language models and what we want from them I guess like my ideal would exactly be that you give the user much more control and much more transparency by letting them vary those things to, to, to what they want to do.

I think there’s actually a particular Danger with large language models that I’m interested in at the moment, which is the way in which we’re encouraged to interact with them as though they’re people. Um, and so when I’ve thought about the epistemology of search engines, I’ve not thought about it in terms of testimony.

I don’t think that’s an appropriate model. So there’s all sorts of work in epistemology about, like, the appropriate norms around testimony. When you have one person who’s talking to another person and they tell that person some things, it’s like, your most basic interaction between two people.

That’s, that’s [01:05:00] testimony. And one thing that’s going on there that seems important is there’s this. interpersonal relationship there between two people. And part of what’s happening when I tell you something is I’m sort of telling it to you as me. And I’m saying, I’m telling this to you as one person to another, and that gives you some grounds to trust it because it’s that sort of an interaction.

I think large language models want us to think that we have that kind of a relationship with them. And I think it’s important that we don’t because that relationship is something that automatically kind of softens your skepticism or your wariness around certain things. Like we want to be, conversationally cooperative with the people that we talk to.

We don’t want to disagree with everything they tell us. We trust them to some minimal degree that just lets that conversation even get off the ground. And if we didn’t, you know, be hard to even do anything. But I don’t think we should have that trust in place with large language models. I think we should be very circumspect.

We should be thinking all the time about where they’re getting their information from and what kinds of things they’re not telling us. I think that’s crucial. It’s like, what information is missing from this conversation that I don’t know about? Um, and I think it’s helpful to not think of them [01:06:00] as kind of a person with the authority and the, and the moral significance that, uh, that, uh, that interacting with a person brings to a conversation.

James Robinson: Yeah, I completely agree with that. I think there is a, a lot of danger associated with the, anthropomorphization, if I can say that word of,, And even people who are very careful in their language, about them. So Steven Pinker, for instance, you’ll even manage to catch them out and find places where they refer to them. You know, it said this, right? I’m like, well, that’s not something that you’d say about a search engine, just the the saying of something is, is to us a very. Act. , so yeah, it’s, it’s really hard to catch them out. Incidentally, just one small thing I think they should do with LM’s is just like make them and I’ll put all the results in comic sans.

So we take it a little bit less seriously., I do think they, they do at least offer that offer that opportunity to tailor one’s results and say , you can just say, I am interested in these things, this is the kind of person I am. , [01:07:00] yeah. Tell me about X. Right. Explain, the theory of relativity as if I was a toddler.

Right. They will do a really, they will do a really impressive job at that and kind of drawing on what is salient to a toddler and reframing, something that you want to know in, in those terms. But I also wonder if they’re maybe too good at that because we might just all end up with this default setting, which is give me the information the way that I like it.

and, and even worse, give me only the kind of information, that I like., but yeah, yeah, completely remains to be seen how. How that plays out I

Jessie Munton: agree with that. And then it’s funny in a way that maybe we want to begin to build in, like we reach a point where these tools are very good at giving us information.

And part of what we want from them is occasionally to frustrate or annoy us by not quite giving us the information we want, but still to give us information that’s valuable.

James Robinson: Yeah. Another thing that came to my mind is I really like this idea of using search engines and also LLMs, actually even more LLMs as a way [01:08:00] of uncovering the kind of common.

Opinion in a certain way. Like, what is it? I think search engines are really good at capturing one’s, like societal level preoccupations. And so, we’re old enough, I think, to remember when Google introduced autocomplete. So you’d start typing something and then it would say, okay, you know, I am having problems with.

And then it comes up with a list of all the things that you could be having problems with. And that’s essentially a representation of the stuff that, that people most commonly type into Google and therefore the things that, people do have problems with, and that is like incredibly useful or yeah, really interesting information.

 and LLMs., also do this. There’s a good example from, uh, Isabel Boemeke, who’s, who’s this nuclear power, influencer, probably the only nuclear power influencer. On her blog recently, she looked at, she started asking chat GPT about nuclear power, nuclear power, and [01:09:00] it became really clear that all its answers were very biased against it.

It was like, oh, it’s really, you know, it’s pretty unsafe and so on. And it’s really hard to dispose of the waste. And it’s like, well, has any, you know. Has anyone ever died from, uh, nuclear waste disposal? Are there any deaths associated with that? Well, and it, and it answers no, but it can’t resist kind of imbuing the opinions, which is, are encoded into its training set, which are overwhelmingly, negative.

 even though like certainly within scientific communities, people are really keen on nuclear power., it’s, it’s very hard to get it to answer in a positive way about that.

Jessie Munton: Yeah. I think another thing that’s in the mix here with both search engines and LLMs is. And maybe this kind of comes back to the themes that we started with. So we naturally think of them as giving us information, and then there’s a question which is, well, what information do we want them to give us?

But I think that that process of giving us information can have these really broad ramifications that go sort of way beyond what we might ordinarily think is associated with it. So the [01:10:00] nuclear power example is nice in that respect, that I would say partly you might think, well, this is a slightly misleading subset of information.

There’s other information which we need in the picture to have a rounded impression of the, value and impact of nuclear power, but it also changes your concept of nuclear power. I mean, presumably part of this is happening because people do have a kind of bogeyman concept associated with it a bit in some contexts.

And then that’s further encouraged if you have these, tools for sorting and prioritizing information, which are prioritizing some misrepresentative subset of it, it’s not just that you end up not knowing the stuff you might want. to know, but your actual concepts that you’re using begin to change under this kind of an influence.

James Robinson: Yeah, there’s a very, uh, this is a

tangential example, but I think it Maybe illustrates again, how just the appearance of something on one’s cognitive landscape can have weird consequences. So in the paper that this morning was, [01:11:00] uh, headline Shakespeare expert overturns flytip a myth about playwrights father exclusive. John Shakespeare’s Muck Hill fine in 50 to 52 was a waste disposal toll rather than a punishment.

Research says, and, I, I noticed this today and I immediately thought about this conversation we’re going to have because I’m like, well, what is the point of that piece of knowledge, right? That Shakespeare’s dad was not a fly tipper, right? Is that even interesting to Shakespeare academics or Shakespeare aficionados?

Well, it is because it’s a headline and, and I read it and I, I was kind of like, oh, yeah, it’s sort of interesting, but it doesn’t, it’s not going to change my appreciation of his plays, but it does. Redraw my salient structure in a certain way. Like now I’m probably more inclined to think of the immortal bard when I go to the rubbish dump or even just take my bins out at night, right, even though there’s no kind of real justification for that.

Jessie Munton: Yeah, but yeah, I think it’s helpful you mentioning newspapers, because actually the sorts of things I’m interested in, of like, what information should be prioritised, that’s a question that’s been in newspapers since forever, they’re having [01:12:00] to make decisions all the time about what goes on the front page and what goes on all the other pages, and what even doesn’t get into the newspaper at all.

And I think another discipline where this stuff is there is also in library science and deciding how you organize libraries, like, what are the relevant categories and what are you promoting by having a category division between these two things rather than another two things. So I think these questions have been there all along, and we’ve tended to pursue them outside of the context of the mind directly, but I think it’s helpful to think about them more directly within the mind.

James Robinson: Yeah, yeah, I wonder if you have any, Ways in which this has kind of practically,, changed how you live your life. Has it made you more aware of what sources of information you take in? And, , or are you kind of like, like many philosophers I talked to you? Like, Oh, this is interesting stuff, but Uh,

Jessie Munton: no, I think it does impact me in certain ways.

I mean, I don’t know if this counts as an impact. I’m surprised [01:13:00] how much stuff I hear going on in the news. And I think there’s a problem with how this is being reported. And it’s not a problem that’s to do with the accuracy of it. It’s a problem to do with what we’re making salient by reporting it in this particular way.

So some of the coverage of stuff to do with trans women, I think, is like often making salient threats. that comes from a particular direction without like necessarily saying things that are inaccurate, but that’s having a really big impact on how that conversation goes. Or I think, coverage of immigration is often like very, very invested in making threat very salient, even when the groups concerned are extremely small.

 so, so, I mean, I, I guess I fall back on it increasingly as like a way of analyzing these things, I think is valuable, but that’s not really, I mean, an impact in sort of my, my particular life. I think, I think it has made me think. differently about how I use social media in various ways, or, yeah, what things am I passively letting just come into my informational environment in ways that are going to change my salient structure in ways that perhaps I don’t want.

 I guess I see it most broadly. I don’t know if this comes out of the stuff on salience or just how I’m thinking about epistemology in general. I think there’s [01:14:00] always a kind of mandate towards curiosity in all of this, like, a mandate towards trying to expand what you’re exposed to or the sorts of things that you consume., I, I think it, I hope it does make me more aware of that and more ready to invest in it in terms of perhaps buying books on weird subjects you might not have thought about or trying to think about what are the options that I’m not. Even engaging with it. I guess one thing that is important to all of this, I think, is that thinking in terms of salient structures lets us think more about the kind of negative space in our mental lives.

So what are the things that I’m not engaging with? What are the kind of lacuna that are there? I think that’s really crucial. Um, and, and so, you know, having just one ear listening out for some of those things and then trying to tune in when some of that’s coming onto your radar, I think is valuable.

James Robinson: Yeah, yeah.

I think that that’s spot on. I mean, certainly ignorance. If we, if we bring up the, if we prop up the straw man one more time, like ignorance has not really been studied [01:15:00] directly in, in, in epistemology, like where it’s come up is just where you have a belief, but it fails to be true, right? You, you believe something wrongly, but it’s not looked at all that negative space.

I guess, just for many evolutionary reasons, we’re not very good at looking at negative space. We fixate on things. I find it so fascinating that you can put a tiny dot in a huge canvas, and although the space is eclipsed. Right? Um, you will spend almost all your attention on that dot, right? It’s not even that your space is like slightly unequally, your attention is slightly unequally weighted across the space.

Like it’s, it’s completely like overwhelmed by that. And that’s just so, um, yeah, it’s so fascinating. Um,

Jessie Munton: and I think, I mean, I, yeah, and I don’t want to downplay the extent to which there has been work on. ignorance, [01:16:00] or it is there to some extent. But I think it’s true that there is this aspect of traditional epistemology that to the extent that you’re interested in applying terms like justification and knowledge, you need something there to apply those terms to.

And with a canvas, like you say, we don’t tend to notice the negative space, but the problems Much more acute in our own minds insofar as the negative space is exactly the stuff that you don’t know. And there’s a kind of, you know, meta negative space. You know the negative space is there in some ways because it’s not, it’s, it’s the stuff that you don’t know or that you’re not thinking about or that’s not making its way into your mind.

So I think that’s like a point at which thinking in terms of structures is really helpful that they will reveal to you where you have negative space a bit more potentially.

James Robinson: Yeah, I think it’s, it seems to me like it’s another lens on a very old problem of exploration and exploitation, um, which is, yeah, I don’t know originally where that, that, that term or that, that phrasing has come up, but I know it’s popular within the kind of machine learning.

[01:17:00] community, like should your models , really focus in on certain features that they, they think are improving their predictivity, um, or like versus the kind of compute time that you could spend on looking at entirely. different features, which, initially may not really seem to be doing much.

 But, but perhaps we, you know, just need a little bit more information. And I mean, that’s such a hard problem, right? Why, what books to read, like what blogs, what people to listen to. We’re in such a rich. Informational landscape. We carry around these devices, which kind of potentially give us access to all the world’s knowledge. , But we can only take such a thin slice of that. But maybe, maybe the answer is we do need to take. Slices across various different dimensions, discard some, but, you know, nonetheless don’t just lean into the things that we, [01:18:00] we naturally inclined to.

Jessie Munton: And it is tricky, like, you know, our time is limited.

There’s way more books already probably that I would like to read than I’ll ever get to read in my lifetime. and so, you know, that can feel like a mandate towards exploiting the stuff that you enjoy rather than exploring. On the other hand, I think there’s all these ways in which we’re fairly. Um, it’s kind of spend thrift with our mental resources and I don’t know, I consume a lot of stuff passively or that I’m not that invested in.

And maybe if I thought a bit more about it, I could make better use of those resources.

James Robinson: Yeah, absolutely. Um, yeah, I don’t know if you have any kind of final thoughts. I, I know there’s like. There’s, there’s a lot more of your work, which is really fascinating. I advocate people to, to look up, you know, you’ve looked at, um, perception and, um, kind of optical illusions and how we learn from, from those.

 and all of this does overlap, I guess, with, with your work here on, on, on salience as well, which is really fascinating. Um, so. [01:19:00] I don’t know if you want to add either on that topic or, or anything else as a kind of final, final

Jessie Munton: thought. I don’t know. I feel like talking about perception might get us into a whole, a whole nother substantial conversation.

 So I, I think I’ll, I’ll leave that there for the moment. Brilliant.

James Robinson: Yeah, this has been such a, such a lovely conversation. I, I do concur that this model does actually unusually for a, Philosophical thesis, change my way of thinking about not, not just change my way of thinking in philosophy, but actually change, the way that I go about and generally consider things.

So I find it a really useful framework. So thank you. Oh,

Jessie Munton: good. Yeah. And maybe if any, network scientists are listening, who wants to help me out, they can, get in touch. Yeah,

James Robinson: I hope so.

Jessie Munton: Brilliant. Yeah. Thanks James.