In 2018 the University of Wollongong hosted a four-day conference dedicated to evolutionary perspectives in the philosophy of mind. Naturally Evolving Minds boasted an impressive line-up of philosophers of mind and biology including Peter Godfrey-Smith, Kim Sterelny, Dan Hutto, Shaun Gallagher and – most impressively of all – Ruth Millikan, well into her eighties and still willing and able to travel to the other side of the world to defend her theory of teleosemantics. On another day I might write about Millikan’s talk (which can be viewed along with the other headliners here), but this post is about the contribution of Paul Griffiths, philosophy of biology at the University of Sydney.

The content of Griffiths’ talk remains unpublished as far as I’ve been able to make out. I believe his argument is insightful, important, and of deep relevance to philosophers of biology in general and those interested in teleosemantics in particular. At the heart of his presentation is the claim that the infamous “disjunction problem” raised by Fodor against Millikan’s teleosemantic theory of content is really three different problems that have been confused for each other.

As well as clarifying issues facing teleosemantics, Griffiths raises methodological worries for philosophers of biology who rely too heavily on oversimplified examples of evolutionary processes. His exhortation to take the biology seriously has risen to greater prominence recently, with his Sydney-based group publishing “Are Biological Traits Explained by Their ‘Selected Effect’ Functions?” together with commentaries in the Australasian Philosophical Review. That paper argues that certain applications of the selected-effects theory of function rely on unrealistic simplifications of evolution by natural selection; likewise, in the 2018 presentation Griffiths critiques philosophers’ tendency to assume that evolutionary examples need not pay attention to biological detail.

Because I think Griffiths’ claims are worthy of attention, and because the audio in the recorded presentation isn’t great, I’ve gone ahead and transcribed the whole thing. Underneath the video of the talk below you’ll find my (attempted) transcript. Some text is uncertain because the audio quality is poor in places. In italicised square brackets I’ve indicated slide numbers, missing text, descriptions of interlocutors whose names I don’t know, and citations to papers mentioned. There’s a reference list at the bottom. Speakers’ names, along with claims I think are particularly significant, are bolded in strong text.

If you spot any mistakes or improvements, please email me and I’ll update the transcript accordingly. I think this talk deserves to be more widely known.

The video

The transcript

00:00 [Slide 1 omitted]
Paul Griffiths: So, going way, way back, in a long-ago galaxy far away, when Jerry Fodor invented the disjunction problem, the problem he raises for the program that Ruth [Millikan] was just outlining is,

00:16 [Slide 2: Semantics and selection]
“Darwin cares how many flies you eat but not what description you eat them under.” [Fodor 1990:73] So in other words, when you’ve determined what evolutionary purpose is associated with a representation, you have a range of choices about how to describe that representation, which will infect the content with a degree of indeterminacy, and so we don’t get a theory of content.

00:36
So Fodor points out that you might say the fly [sic] is sticking its tongue out because it’s seen something small, dark and moving, it’s seen a fly, it’s seen some food, or it’s seen something that it would be fitness enhancing to ingest. And there’s a whole literature in this period in the early 90’s in which everybody in this room seems to have written a paper about.

01:03
Kim Sterelny: Those of us who are old enough.

01:06
Paul Griffiths: All of the grey-haired in the room have written a paper about. And various people have tried to solve this problem. I’m not sure whether it was Richard Goode and I who started calling them “Target of Selection” arguments […?…target of selection arguments]. What you do is to say, look, sure, you can describe this in many ways but only one of them captures the process that’s really driving natural selection here.

01:35
So you could say, look, I think Larry Shapiro has a nice quote where he says something like, there’s no kind of evolutionary optimisation principle that talks about how great it is to eat small, dark, moving things. So instead that’s just wrong, and so Shapiro says, actually, the function of this representation is to represent food. Kim [Sterelny] disagrees and says “fly”. Karen Neander says, no, the function is to represent small, dark, moving things, and then you get other proposals.

02:12
If you think about it, and there are some of these proposals, you can put “and” or “or” between any of the options, and get a whole bunch of- I recommend, if you’re looking for a cheap publication, you’ve got an “and” and “or” operator, you’ve got [five?] philosophy papers, you’ve got another paper, it’ll probably come out in Philosophical Quarterly.

02:33
What’s going on here is the application of Elliott Sober’s selection-of/selection-for distinction, which I’m sure we’re all familiar with but I’ll quickly go over it.

02:44 [Slide 3: Selection of versus selection for]
I don’t know if maybe it was the young Aaron Sober or somebody who had this toy. You shake this toy, you hold it, turn it upside down, shake it, all the balls go to what’s currently the top; turn it up again, shake it, the balls fall through, but only the smaller green balls get to the bottom. And one wants to say – in the Wittgensteinian mode, one is inclined to say – that they got to the bottom because they were small, not because they were green. And Elliott [Sober] cashes that out saying, in this case this is a machine which selects for size, it doesn’t select for colour. How do we know that? By evaluating some counterfactuals. If the balls’ colour was held constant and their size was changed, you wouldn’t get the same outcome; but if the size was held constant and the colour was changed, you would get the same outcome. So we just evaluate some counterfactuals, we’re thinking, as Elliott says, this is just the distinction between causation and correlation. Nothing complicated; nothing to see here.

03:45 [Slide 4: Scepticism about the distinction]
There’s a lot of scepticism about this distinction. A lot of it is scepticism about whether or not we can legitimately make and evaluate the counterfactual statements. There was a paper I came across, a 2012 paper, arguing that these counterfactuals are not well-formed and can’t be evaluated. I haven’t cited it because I think it’s totally mistaken – and why should I confuse more people?

04:14
There’s an awful lot going on in Fodor’s complaints about that idea that you can make sense of the distinction between the target of selection and its side effects. There’s all sorts of arguments mixed up in there. One of the key arguments is basically a claim that you can’t make or evaluate – several arguments, which are criticising counterfactual evaluation, saying it can’t be done in the right way. There’s a very nice, classic paper I think by Elliott Sober basically explaining, I think, why, if you take that scepticism seriously, you’re really being very sceptical about a lot of basic ways we think about causation and the laws of nature. It’s not a little problem about evolution. And I agree very much with that view.

04:59
There have been I think more interesting, some neo-Kantians, who think that you can’t actually – in I think better ways than Fodor – argued convincingly that you can’t – I’m not convinced, but it’s convincing – that you can’t eliminate the implicit taking-the-point-of-view of a designer in the [same?] situation. And their arguments are not that distinct – so, a 2002 paper by Peter McLaughlin at Konstanz, that sort of describes this as a neo-Aristotelian view […] [positive to the?] good, but actually in the 2001 book [he’s a?] neo-Kantian. So these sort of worries about whether you’re entitled to your overall views about what the system’s trying to achieve. I’m not going to talk about that kind of scepticism, but just to note, there are all these kinds of sceptics floating around.

05:51
I want to present the selection-of/selection-for distinction as – biologists don’t talk that way particularly, but they think that way all the time. Sometimes they talk that way. But this is a sort of philosophical, abstract gloss on a very straightforward piece of reasoning which you do all the time when thinking about biology. And this is my favourite example.

06:11 [Slide 5: Which property was causally efficacious?]
And it’s important that it’s my favourite example. We’ll come back to that later. A lovely bit of work by a guy at the University of Vienna. So, there are lots and lots of leg-waving frogs. Leg-waving frogs are really cute. Instead of croaking to say, “other frog, go away”, or whatever it is, they hold their back leg up and they wave it around to send semaphore signals. And they often have beautifully coloured webbing between their toes, a kind of secondary adaptation. And they’re really cool!

06:43
And they’re really cool partly because they’ve evolved several times on different continents. So a lovely example of parallel evolution; something important’s happening here. There is an observation which is that you find this behaviour evolving in frogs that live on fast-moving watercourses. They live in environments with lots of rushing water.

07:05
Here are two hypotheses you might have: anybody who’s listened to a frog croaking must have had the same thought we’ve all had, which is: If I was a happy little frog-eating organism, that would be a really good way to find frogs and eat them. So you might think that one of the great things about leg-waving is that it doesn’t give away your location. The frog on the next rock who’s trying to invade your territory can see the semaphore signal, but the heron 50 metres upstream has no idea that you’re waving your legs around. Whereas if you were sitting there going “croak, croak, croak” very aggressively, the heron might hear you.

07:42
The other hypothesis might be that if you live next to a noisy waterfall, and you spend your time trying to make people go away by croaking, well you’re kind of onto a losing game. “What was that you said?” So maybe the thing to do is to wave your legs around and have brightly coloured webbing.

08:04
There are some nice comparative biological tests of this. And the universal view is that leg-waving frogs evolved because you can’t hear the croaks over the noise of the waterfall. Now I take this to be really clear what’s going on: so you say, as a matter of fact, these signals are quiet, but they’re also noise-proof. They were selected for being noise-proof, there was selection of quiet signals but there was no selection for them being quiet.

08:37
So what we’ve got is two distinct empirical causal hypotheses about what caused gene frequencies (if you want to be gene-centred) to change over time in these populations of frogs around the world. And the claim is that the causal explanation involves the fact that one signal can be detected in a noisy environment and its competitor can’t be detected in a noisy environment. And there’s another hypothesis which says one signal can be detected by distal predators, and the other signal can’t be detected by distal predators. The first hypothesis has been tested by experiment and observation, and proven to be true. And the second one has been shown to be false. Very straightforward!

09:19
And all I want to do is to elucidate what we mean by saying these are causal competitors. What it means to say they’re causal competitors – I think this is quite interesting – it’s that it’s actually always intellectually coherent to say, “Well, maybe it was six of one and half-a-dozen of the other.” It’s entirely possible that if there were no predators, the gene frequencies would change as a result of selection for noise-proofing, but if there were a lot of predators the gene frequencies would change faster, because in fact the thing was protecting against predators as well as being able to be detected by conspecifics. So you get composition of causes. It’s totally coherent to think both.

10:00
That’s just vectors of causation coming together. They are causal contenders in the literal sense that something’s going somewhere at a particular speed, if this one’s pushing it harder, that one can’t have been pushing it that hard. And if this one’s doing all the pushing, that one wasn’t doing any pushing. That’s the sense in which they’re causal competitors. Very, very straightforward, and very causal.

10:26
I think, and I think if you look at everything Elliott’s written about selection-of and selection-for, he thinks it’s just a way of distinguishing between correlations and causes, and perhaps relative importance of causes in a straightforward “composition of causes” sense. Very, very straightforward.

10:41 [Slide 6: The disjunction problem revisited]
All right, let’s come back to the disjunction problem. Here’s a hemoglobin molecule. We might say, “Does blood contain hemoglobin because it’s rich in iron, or does it contain hemoglobin because it binds oxygen?” Or we might say “Does it contain hemoglobin because it binds oxygen, or because it enables respiration?” “Is it there because it enables respiration or is it there because it’s fitness-enhancing to have it there?” And so now we’re looking at the classic dinsjunction problem. Notice that we’re looking at something very, very different from the leg-waving frogs. We’re not looking at alternative causal explanations of why we find hemoglobin in quite a lot of organisms. In fact, what we’re looking at is something like different levels of [?] presenting exactly the same explanation.

11:35 [Slide 7: A means-end hierarchy]
So I’m viewing this mainly as a means-end hierarchy. So we say, look, this molecule contains a heme group in order to bind oxygen in order to enable respiration. And this mechanism in the frog fires in response to small, black, moving things in order to fire at flies in order to ingest food. So it’s a means-end hierarchy. And you could say that it does all of this in order to maximise the representation of its lineage in future generations. So you can stick a “fitness” as an overall one.

12:12 [Slide 8: Levels of explanation]
So in a paper way, way, way back, during one of the prequels to Star Wars I think, Richard Goode and I wrote a paper drawing attention to the relationship between the disjunction debate and discussions of levels of explanation in biology. And that’s kind of an obvious thing to do. We weren’t the first people – the first person, I think, to make that point about the hierarchy is Karen Neander. We were interested in taking that hierarchy seriously.

12:43
So here’s a kind of methodological rant: this kind of stuff is so informal that it’s kind of crazy reasoning about its consequences, “Does that make sense in evolutionary terms?” This is not even trying to be serious about giving an explanation of this stuff. And if you give an explanation of this kind of throwaway intuitive in a lecture example, and then worry about its consequences, goodness knows which of the bits of sloppiness that you started with are responsible for the consequences. It’s not an interesting way to think about this problem. So you have to think about this as merely gesturing towards actually trying to publish a paper about why frogs wave their damn legs. And the kind of content you might put in such a paper.

13:30
So, we think that to take that means-end hierarchy seriously as biology, each of the levels in the hierarchy has got to correspond to an explanation that a biologist would take seriously. “Ah yes, now I see why it is so.” The highest-level explanation, where you say, “All these things were selected for enhancing fitness”, actually we’re really talking about what– Elliott Sober has the phrase “consequence laws”. So you characterise the trait and its alternatives merely by assigning them relative fitness values, or a function – say it was in frequency-dependent selection, it’s a function that gives you the fitness of that trait relative to the others. And that’s fine, because if somebody says, “Wow, why does this trait persist?” And you say, “It persists because it’s frequency-dependent selection; if there were any more of them then they’d be less fit and if there were any less of them they’d be more fit and that’s why you find this number of them and not more.” And that’s an explanation of why we see that trait, where the trait is characterised purely abstractly as a biofitness function. That’s the only thing you need to know, to plug into the explanation. And that’s a good explanation! In fact, the most interesting explanation [that some people…?] “look at this [exciting? …] of frequency dependence.”

14:56
Another phrase of Elliott Sober’s, “source laws”. So fitness has consequences, where does it come from? We might go to a kind of evolutionary ecology perspective and talk about how the trait interacts with the environment so as to enhance fitness. So, what does it do which leads to the organism being able to have babies or not get eaten or whatever it might be. And again, we might actually get interested– so my example about frogs, we might be interested in why frogs wave their legs, we might be interested in why this species of frogs waves its legs. And so we might need to give an explanation which features the actual taxa. So it names specific genealogical objects in evolutionary history, and – just plugging what I think is one of David Hull’s greatest papers, that everyone always forgets, a great paper called “Genealogical actors in ecological roles” [Hull 1987].

15:47
And the other thing is we might want to give an explanation of how the ecological interaction is mechanistically mediated in a specific taxon. So take a particular frog, a particular fly, whatever it is, and we want to say: “How does this organism solve the problem?” And these are all great explanations, and you can find explanations of all these and no doubt other forms.

16:13
So in order to have a function description you need to back it with a serious evolutionary explanation. And how many function descriptions are legitimate is not a matter of how many you can run off your tongue! It’s a matter of how many serious explanations are there. But we’ll come back to that. And also whether they’re good explanations. We all disagree a lot about which explanations are good. In biology we’re always arguing about which explanations are good, which ones drive out others; that’s where the real action is.

16:48 [Slide 9: Is fitness the sole target of selection?]
Here’s a particularly bad view of where the real action is. “Is fitness the sole target of selection?” So back in that 1995 paper, Richard [Goode] and I argued that targets-of-selection arguments iterate until– so somebody says, “Natural selection doesn’t care about catching small, black, moving things, it cares about catching flies!” And then someone says, “Natural selection doesn’t care about catching flies, it cares about catching food!” “Natural selection doesn’t care about catching food, it cares about doing things that are good for you! If food wasn’t good for you, nobody would eat food!” And sometimes food isn’t good for you. Ephemeroptera, the adult males never ingest food, because it would be bad for them and reduce their fitness. So there’s no reason why food has to be fitness enhancing. Sometimes it is, sometimes it isn’t.

17:42
In that case you’d say: “I’ve solved the problem, Sober’s selection-for. The only thing that there is selection for in the toy is getting to the bottom. There’s no selection for being small, there’s only selection for getting to the bottom. So what’s wrong with that? Why’s that an important argument? It’s an important argument because lots of smart philosophers think it’s a good argument!

18:03
So I’ve chosen a smart philosopher instead of a dumb philosopher – there are many dumb philosophers I could have chosen who have said the same thing! This is Steve Stich, in The Fragmentation of Reason [Stich 1990], talking about exactly Ruth’s topic. He says, “Natural selection doesn’t care about truth, it cares only about reproductive success.” Which is the argument we’ve just seen. And in various papers with John Wilkins I’ve argued that’s a bit like saying natural selection doesn’t care about respiration, it only cares about fitness [Griffiths & Wilkins 2015].

18:34 [Slide 10: Is fitness the sole target of selection?]
So truth-tracking and fitness-tracking are not alternative hypotheses. You can see that because they fail the causal composition test. So if somebody says, “Oh, so you mean it’s not selected for corresponding to how things are, that representation is only selected for enhancing the organism’s fitness.” Now try and formulate the hypothesis that it evolved to maximise an organism’s fitness, but it would have evolved faster if it had also corresponded to reality. There is no such hypothesis. You can’t put the two together as vectors that drive gene frequencies, with one driving it at a certain rate and the other speeding it up. It just is completely incoherent.

19:14
Here’s the way I explain this to students. Here’s a perfectly sensible question: “Did you bus to campus or walk to campus to come to the conference this morning?” And here’s a seriously dumb question: “Did you bus to campus or travel to campus this morning?” So it’s literally a Rylean category mistake to say that tracking truth is an alternative to tracking fitness. You may very well reject the view that organisms want to track truth; there are respectable arguments for that. For example by Dennett, who I think gives what’s the strongest version of them a few years ago. Ryan and Dennett, I think. [Probably McKay & Dennett 2009.] There are respectable reasons to reject the hypothesis, but rejecting the fact– “You didn’t use that means to the end. You merely achieved the end!” The only way that makes any sense is by nominating an alternative means. You can choose different means to the same end; you cannot substitute the end for a means. Unless maybe it was magic.

20:06 [Slide 11: Summary to here]
So, summary to this point: the for/of distinction is a distinction between distinct, competing causal hypotheses about episodes of natural selection. It can’t be used to discriminate between more or less abstract descriptions of the same empirical hypothesis about natural selection. However, there are other ways to dismiss a claim about the target of selection. Those other ways are perhaps more relevant to real disputes about content and indeterminacy. So I’ll briefly talk about those.

20:40 [Slide 12: Redescriptions]
The first one is redescriptions. Suppose I make the following claim: “Leg-waving was selected for communication in the environments discussed in Hödl and Amézquita (2001).” Is that true or is that false? Or I might say, “Hemoglobin was selected for having Paul’s favourite example of differential binding affinity.” Is that true or is that false? Well, maybe the philosophers of language in the room may have a better view of this, I think this is just sensitive to the old de dicto/de re distinction. They’re both true de re, they’re both false de dicto. This also shows up in debates over Fodor with several people. The nomic force of a statement can be lost by substituting co-referring terms, but not if you read it de re, because then they’re just picking up the thing you were talking about anyway.

21:40 So I kind of think a lot of this is going on in the literature. And I’m pretty sure it’s not philosophically very interesting. This is a good way to slip in an argument; pull the wool over an eye. I’m not sure that deep work on exactly how to formulate nomic statements so as to not get into Quinean problems is really going to do the work. But there is a lot of that going on. Again, what you’ve always got to do is ask that question about causal composition. Are we really talking about distinct empirical hypotheses about what pushed things? If not, we’re not using the distinction. One take-home about what’s going on is not only Fodor but also other people, critics of the distinction who say, look, you can’t discharge the intensionality, seems to me they’re basically confusing – coming back to intensionality – and so they notice that there are intensional contexts here, and think, “Oh my God, mind must be involved!” But no, in fact, it’s nomic force that’s involved, that is generating the intensional contexts.

22:49 [Slide 13: Reduction and explanation]
So – ah! I’ve chosen the wrong book cover, how could that have happened? Anyway. Alex’s [Rosenberg’s] other life featured here. So we’re going to talk about reductionism and therefore of course we have to have an image of the reductionism book. So are all levels equally causally explanatory? So this is where I think there’s something really interesting going on. You might well be quite sceptical that reducing– replacing an ecological explanation which generalises across all taxa that are subject to the same process of convergent evolution with a blow-by-blow story about the actual history of the specific taxa, is genuinely causally explanatory. You might say look, that’s– and I don’t particularly want to endorse this view, but we all know people who have the view, “Look, that’s just a narrative. You’ve given a narrative account of what happened, but it’s not the causal explanation.” So what we need to think about: we’ve got all the stuff that we’re wanting to talk about– I think Richard and I in that 1995 paper presented this in terms of program and process explanation. Obviously, now we’ve got comparative and contrastive; there’s a whole literature on how to think about this stuff. So we’re all aware that there’s a whole bunch of philosophical views according to which moving up and down those kinds of means-end hierarchies doesn’t always preserve whether you’ve got a real gutsy objective causal explanation. And of course Rosenberg’s view is going to be that it’s description all the way down to something physico-chemical. So all of what we normally call explanations in biology are in fact descriptions, and you don’t get any actual causal-explanatory force until you go reduce to the absolute bottom level.

24:41
If you haven’t read the novel, the novel is actually an attempt to make the view of the world devoid of anything but physicalism, in which narratives are a delusion, they don’t correspond to anything real. It’s actually an attempt to– he’s trying to do a Jean-Paul Sartre here, it really is the view from Darwinian Reductionism sensuously embodied in a novel. Good!

25:15
The point I want to make is that suppose that I say, “Here are two function descriptions, each backed by a more or less abstract selective explanation, and I like the first one but I don’t like the second one. I don’t like the second one because I actually don’t think that you get– because supervenient causation isn’t really causation, I don’t think that high-level one is really causally explanatory.” We are not even in the same ballpark as the selection-of/selection-for distinction. And yet that’s often expressed by saying, “There’s selection for the high-level property but not of– I mean, only selection of the lower-level property” or vice versa depending on your prejudices. There’s nothing wrong with that it’s a really coherent reason for rejecting– picking one over the other. But it’s got nothing to do with what was going on in the selection toy. It’s got nothing to do with what was going on with the leg-waving frogs. It’s completely orthogonal.

26:15
Again, note that it fails the causal composition test. It’s not the best example, but it’s fine: “Hemoglobin was selected for binding oxygen.” So I say, “I think early in the history of life, hemoglobin was selected for binding oxygen, but it went to fixation in the population much faster because it also enabled respiration.” Again, that’s again just incoherent, it’s a category mistake. So that’s not the right way to– so, we’ve got really something completely different, and that’s why I’m inclined to think if anywhere there is a right answer to these, “Is it a fly or food or small, dark, moving thing,” it’s got to come out of your general commitments about causal explanation in the special sciences, which is going to tell you that some of these levels are admissible or inadmissible. I haven’t really written much about that, because there’s one immediate conclusion, which is, “Wow, that’s a vexed debate.” There are so many competing– Marcel Weber’s going to have one view, I’m going to have a– Karola and I are going to have a completely different view. Rosenberg will have a view. Alan Love will have a view. You can pull out a whole bunch of really smart philosophers of biology all with different views on which of those levels yield genuine causal explanations. And there’s going to be a take on the disjunction problem corresponding to each of those views. Wouldn’t that be fun to work out! And argue for one or the other.

27:40-28:00 [Looking for a slide]

28:00
So I’m pretty much finished but there was one point I wanted to make. This stuff about not being able to – and this is again a kind of a methodological rant – there are some quite strong claims get made about how you can’t evaluate counterfactuals when things are nomically linked. So somebody will say, “Look, you just can’t make sense of Sober’s claim that – or my claim, sorry – it was selected not for having a heme group but for binding oxygen because there’s a nomic connection [that you can’t have that heme group without] having that binding affinity with oxygen.” That’s a really simple case because in fact – I had to put this in for Peter Godfrey-Smith – if you’re an octopus, you don’t have hemoglobin, you have hemocyanin, and you have two copper molecules in the middle there instead of a heme group, and you bind the oxygen with copper molecules. And there’s various insects as well, and you get nice green blood as well, which is cool. Or, I don’t know, Peter is it green? I always assumed it was green.

29:05
Peter Godfrey-Smith: Greeny-blue, though I’ve never actually seen it in the flesh.

29:09
Paul Griffiths: That is so cool. I would like to have greeny-blue blood. That would be great. I’ve never look embarrassed.

29:20
Hemoglobin and hemocyanin. So in this case you see how you evaluate that counterfactual. You basically say, “In fact, if you take a terrestrial organism like an insect–” so I believe there are some quite substantial reasons about why, to do with pressure, about why the hemocyanin’s quite good for an octopus – if you take a terrestrial organism and it can have either one of the many hemoglobins or one of the many hemocyanins, and you basically say, “If the historical contingency had happened to give this lineage this set of genes and not that set of genes to deal with this problem, would they have still been able– sorry, and they had solved it using another molecule, would they have been equally successful?” Well, yeah, sure as hell they would have been equally successful. They would have moved oxygen actively to the tissues, it wouldn’t have made any difference. So you can evaluate counterfactuals like that.

30:12
So, the methodological rant is this. What you tend to get is people saying, “Oh, you can’t evaluate that counterfactual because they’re nomically linked.” Instead of thinking the real question should be, “Suppose that I was actually a biologist who was puzzled by this. I wasn’t sure why that trait evolved.” And that really does happen. Birdsong – Ruth has used the example of birdsong. The functions of birdsong are still quite obscure! Not because the hypotheses are obscure, but if you look up a good review article on the functions of birdsong, and you’ve got a [?] debate. Not about what is it doing in the chickadee at the moment, but how did it evolve, what were the big pressures that led to its evolution. It’s quite hard; there’s a lot of controversy about comparative claims about why, what was the big driver in its evolution. Is it territorial defence? Is it bringing the female into breeding condition? Is it demonstrating mate quality? All the different things you might be doing. So it can be quite hard.

31:20
Look at the ways in which somebody does what would actually correspond to evaluating the counterfactual. You’ve got all of these ingenious indirect ways of asking some sort of question that corresponds to, “Were you to keep the upstream effect which is leading up towards fitness, but ring the changes down here, would it make a difference?” And there are a lot more ways of doing that than you might think. They can be quite indirect. They can involve– I was at a great, an amazing – this talk blew my mind – a theoretical chemistry talk on the third phase of water, another phase of water which would sit between, I think it was between liquid and freezing, if it hadn’t frozen by then. These guys are building serious quantitative models of possible worlds in order to think about real-world questions in chemistry. So all I’m saying is I think what the philosophers think about how you evaluate counterfactuals often are just not thinking long and carefully enough and looking at the way that scientists will do stuff and might in that case– “Hang on, isn’t that kind of crazy?” No, it’s not crazy, it’s weird, it’s indirect, but it actually gets at the causal– it’s a way of thinking about causation in complicated counterfactual ways.

32:46
So the point I wanted to make about the hemocyanin is just that I really think we can in fact use selection-of/selection-for in a lot of the cases where people claim it’s simply logically incoherent to imagine the two properties being distinct. So we might only be left with something like logically coextensive equilateral/equiangular examples pretty much anything else you might want to make sense of a counterfactual.

33:20 [Slide 14: Conclusions]
So, conclusions! Selection-of/selection-for can only distinguish between properties that feature in competing causal explanations of the same episode of change due to selection. We can also reject a selection explanation as not being causally explanatory, but that’s not an application of the distinction. So if I’ve got two causal explanations which are both humdinger, stamp-of-quality ISO-9001 certified causal explanations, but one of them is false, that’s the selection-of/selection-for distinction. If I’ve got a causal explanation with a 9001 ISO stamp, and another one with a fail stamp on it, that’s not the selection-of/selection-for distinction. That’s just a bad explanation, and the reason that that’s not its function is not because it competed with this one, but because what you said wasn’t true. You said it happened because of this, and the “because” is a non-sequitur. So I think that second one, second option, actually offers us some opportunities for making some progress on the old disjunction problem issues.

34:30
But here’s the depressing conclusion: I think that the value of the distinction for worrying about detailed questions in semantics – “Is the content this or that? Is that really content in this case, or quasi-content, or whatever?” – is going to be very limited, because actually I think the big philosophical disputes about content, whether that counts as content, is there content here, [and what we were talking about] earlier, are really unlikely to be causes– to be disputes about the actual causes of evolutionary change. There’re unlikely to be people arguing– so I think there are some disputes we were having back in the 90’s that we can make progress on by getting, committing to firm views about the admissibility or inadmissibility of different explanations within a hierarchy of reductive explanation and supervenient causation. There are some things that we can clarify there. But I doubt whether the sorts of things that people, when people worry about whether the teleosemantic program can recover a certain aspect of the semantics of natural language, that those are the disputes that are really underlying what they disagree about. There probably are, in fact, disputes in semantics rather than disputes in [evolutionary] biology.

35:41 [Slide 15: Acknowledgments]
OK, and acknowledgments to all my group, friends and collaborators, particularly John [Matthewson and John?] Wilkins of course, who worked on one of those papers I cited. So that was a short talk and we have lots of time to discuss.

35:54 [Applause]

36:04
[Chair]: so we’ve bought ourselves some time before lunch.

36:10
David Kalkman: I wonder what the consequences of what you’ve said are for teleosemantics. So for me, two points of view that [aren’t?] completely unavailable. The first is, so in the case of indeterminacy you were talking about [what to do with?] levels, rather than competing causal explanations on the same level. And you said, what we would regard as a good, real cause of selection would depend on our explanatory interests, […] what, if any, consequences does that have for the grounding of content which seems sometimes to be more of a metaphysical project. And the other, to think about how teleosemantics fares is, when you come to more paradigm intentional states like linguistic signs or beliefs and desires and […] representations, the content of those doesn’t seem to depend on one’s explanatory stance in the biological sciences. Sort of like you mentioned at the end. So you might wonder whether when you talk about the content of some subpersonal representation, [whether] that’s really continuous with the content of more paradigm intentional states. Because you just don’t have the same sorts of indeterminacy in more paradigm cases.

37:38
Paul Griffiths: That’s really helpful. David Kalkman – thank you, that’s helpful. People were saying who they were when they asked questions at the beginning, which was really great, maybe we could go back to that. So, David, good question. I don’t want to endorse an “explanatory interests” sort of pragmatic take on how those things– I think that if there’s a population-genetic explanation of why a trait– of why the dynamics of competing traits in the population had a particular course, that’s a really good explanation. And I also think that if there is a historical actual-sequence explanation involving particular taxa interacting, that’s a really good explanation. I don’t see them competing any more than – my favourite – Tinbergen’s four questions compete. How does it work, what is its etiological function, what’s the sequence of events that led to it being here. These are– they’re just answering different questions.

38:36
And in order to give some notion of– I’m inclined to think– if you answer all the questions, there are some funny things– there are things like, I’m very fond of Arno Wouters’s idea of viability explanation. Some of the stuff that Brett’s [Calcott?] done. There’s probably more flavours of explanation around. So I’m not suggesting it depends on your explanatory interests. So when I talk about indeterminacy, the kind of thing that Richard Goode and I said was, it’s not that the function is indeterminate, it’s that there are a lot of true, more or less abstract descriptions of the function.

39:16
Because in fact there’s a single– we’re not disagreeing about– there’s no disagreement at the level of, if we had the gossip magazine level of detail about evolutionary history, so we knew which hedgehog had done it with which hedgehog, and what the other hedgehog thought about it, so we have everything – these hypotheses are not disagreeing at the level of that. They’re disagreeing about what story, what sort of extraction of the explanatory facts to make from that. And they’re not disagreeing in the sense that– I mean, the bits that they extract are complementary, they fit together in a way that’s part of a richer total picture. So I guess one view, and I think Ruth was sort of hinting at this earlier, would to say, look, when you’ve told me all of those stories, you’ve actually given me a univocal story about what the function is. Does that make sense? So you’d say there’s only– “well that then gives you a univocal story about content.” That’s the problem.

40:17 [Chair picks another questioner]

40:22
[Second questioner]: Thank you, yeah. Another question about the implications for teleosemantics. I was wondering if a lesson to draw from your comments is rather that the disjunction problem for teleosemantics is overblown, and we should just say something to the effect that the point of teleosemantics is to offer us some kind of a direct reference account, which is to say that there’s a correspondence between representations and reality. Representations refer to reality. So, if teleosemantics wants to tell us that representations refer to something like the conditions under which the representations perform their function, and as long as those conditions are things that take place out in the world, then we have our direct reference between representation and represented. So, why should it matter so much how we describe that function? Why should it matter if it’s described more or less abstractly? Just as long as the condition in which the function is satisfied is a condition in the world.

41:41
Paul Griffiths: Great question again. So my reply to the anonymous questioner is as follows…

41:44
[Second questioner]: [?], sorry. My name’s [?].

41:52
Paul Griffiths: So that actually helps me make a slightly more substantial answer to David’s question. The take-home message from this: the disjunction problem is three problems. One of them is kind of silly: it’s the one about redescription. One of them is the selection-of/selection-for distinction. People can understand it a bit better, and they’ve made some pretty major philosophical errors like the one I attributed to Stich, by not understanding a simple distinction. And the other one is even more interesting: it’s not what Sober was talking about, but it is actually worth worrying about when you think about it. Why did we fall into this error of calling this “the” disjunction problem? Because we thought about it at the level of, “Isn’t there some kind of thing, it’s got four legs, anyway, suppose that its function was this.” And you can’t think about that stuff at that level of wooliness. You’ve got to take explanation seriously. So what is a functional explanation? It’s one of THESE explanations. Spell it out. So that’s really why you– it’s the utter– you couldn’t see that they were different because you were looking through a lens covered in Vaseline by using those kinds of thought experiments. The moment you’re actually spell anything out, it becomes transparently obvious that they’re not the same thing. But it’s not like that stuff about leg-waving frogs was complicated.

43:07
So that’s the methodological take. Now on the natural language, the stuff that I’ve done with these guys, the people [in red?] up in the top left, who’s sadly not with us any more, we’ve only been interested in thinking about representation in gene regulatory networks. And in gene regulatory networks, we really want to talk about what is the pushmi-pullyu content of this distal molecule drifting over here and latching on here and switching something on. And at that [point?], one just says, once you’ve given me all the levels of explanation I fully understand how that thing came to being what’s going on. We’ve got all the facts, you can talk about it in various ways in English, and that doesn’t really matter a whole hill of beans. If you think that something important in the biology turns on how you describe it using your native language, you’re probably just confused.

43:57
On the other hand, however you describe it, once you’ve said all those things about its function, [and] they have quite substantial consequences, for example, for […] the distinction between pathological and non-pathological state on the network. Is that network functioning or malfunctioning? A signalling network can be sending– are they “false” signals? Well you know, “false” is again one of these very… something’s gone wrong in that signal. That signal is not working as it was meant to work. This thing’s bust. We can get that.

44:28
When it comes to moving over into thinking about semantic natural language, I guess– and this is, I know, really petty, what I was kind of getting at with that last point was that they’re the kinds of things that serious philosophers of language worry about, things like hyperintensionality– hyperintensionality is just my favourite example. Probably whatever they’re talking about there is going to be extremely difficult to relate in any kind of detailed way to the biological causal narratives that we’re saying are grounding– So I’m inclined to think that… One response to this, actually, is to say that we can give a solid naturalistic explanation of simple kinds of representation, and that some of the things that we– and this is completely waving my hands– but some of the really subtle things we want to say about differences in meaning are the kinds of things that we only get to– you know, that we kind of– I’d say they’re sort of social constructions, the fact that these meanings are distinct is only understood in terms of communities of speakers, and conventions, and so on and so forth. Like sort of McDowellian or something.

45:40
Anyway, so I’m quite sympathetic to the view that this stuff works really well for simple representations, and then there’s a bunch of other stuff where you’d have to be more sceptical. But that’s really just hand-waving, I haven’t written anything about it or thought about it deeply enough.

45:55
[Chair]: Alright, I think we have time for maybe a few more questions.

46:00
Dan Hutto: Hi, I’m Dan.

46:03
Paul Griffiths: Sorry, do you work here? [Laughter]

46:05
Dan Hutto: Sometimes. So I just want to kind of come to the end of the old paper from ‘95 [Goode & Griffiths 1995], to see how much, if anything, has really changed in your thinking since then. Because of course, one way of putting it, […] you were trying to answer the canonical Fodorian […] you might have got to the end of that paper and thought, you’ve kind of come up with a conjunctive solution. You’ve put all the things together and said, “That’s the content.” And if you do that, I take it what you’re saying is you can do some serious work to parse that out. Perhaps it’s not impossible to choose between levels now, is that the idea?

46:41
Paul Griffiths: Yeah, look, I think that was– I would be more inclined to say that when you are talking about– the paper, the most advanced paper that I’m [?], it’s not got an author on it, but it was written by our group leader who I’ll talk about in a sec, [?], is Brett’s paper, Signals that make a difference [Calcott et al 2020]. So we want a quantitative measure of information content in gene regulatory networks, where we can still subject that quantitative flow of information to a correctness condition. We can say that there was a functional or dysfunctional flow of information. And potentially, I forget whether this is in the published paper, potentially we can measure the difference between a fully functional flow and the flow we actually get. So how many bits of difference are there between the flow where the system is working perfectly and the flow of information through the network where it’s not; OK, using causal information theory.

47:45
So, at that level, I really think the right thing to say is, I can give you these several different function descriptions, if they’re all good, they’re all illuminating what’s basically the same multi-level causal story about how stuff got into place. And if you then get puzzled about, “oh but if you say this, one would have to say that”, you’re sort of assuming that gene regulatory networks speak English, which has got really no connection with what’s happening down there. So you’re using these English-language ascriptions of function as ways of pointing to aspects of a causal-explanatory narrative. But you shouldn’t– you know, as I say, if you start generating weird puzzles about what the meaning is, probably you’re just taking the language you’re using too seriously.

48:41
The problem with that is that that sounds great for the kind of stuff that Brett [Calcott] was doing. It perhaps doesn’t sound quite so great for some of the things that philosophers of language are exercised about. But I’m not an expert on that, I’m not a philosopher of language.

48:55
[Chair]: follow-up.

48:58
Kim Sterelny: I’m Kim Sterelny. This is back to Dave’s [Kalkman’s] question. I wonder, even if you had the full kind of gossip history of your hedgehogs, you’d still have some indeterminacy of function, because you’ve got to distinguish between the episodes which were selective versus the episodes which are drift-like. It’s not obvious that there’s going to be an unequivocal and objective way of drawing that distinction. So, at the very least, you’re going to get some indeterminacy around the fringes as to which episodes you count as part of the selective story, and which ones you don’t.

49:34
Paul Griffiths: Great question again. I love– thanks for asking that. That is in fact what [Kimiko?] and I are currently worrying about and working on, and hopefully we’ll be working on it fairly substantially. And the right formulation of the question in my view is part of the solution. The formulation of the question is: given the dimensions and degrees of indeterminacy that will exist for exactly that reason, do you get enough determinacy to answer the questions you need to answer for some specific scientific purpose? That’s the way we’re framing it. So you’re absolutely right, but is that a problem? (Kim: it doesn’t have to be a problem.) Paul: Yeah, it totally depends on the details.

50:12
[Third questioner]: What about where you’re– what about where the distinctions are in fact fuzzy sets? Paul Griffiths: well that’s what we were just talking about. Evolution by natural selection is a process– populations, defining the limits– so somebody says, is that one population of wallabies or two, where there’s two [rocky knolls?]? And you’re looking at gene flow between them. So here’s the, again, really important: it is, you’re never going to make progress on a problem like that without taking the biology seriously. Because in fact population geneticists who are much smarter than I am – people who are really world-class mathematical minds – are interested in, are there workarounds for that? Are there tests that are OK under certain conditions? If you don’t think about it at that level, you’re just playing games. You can say, “oh there’s vagueness here, that might be a problem”, but then you just have to stop, unless you take it seriously. It might be a problem, it might not be a problem. It might be the case that we can’t do it unless there’s more determinacy– sorry, unless the facts out there draw sharper distinctions than they actually do. But you can’t answer a question like that, you can only raise it, without actually having some serious scientific models to talk about.

51:38
[Chair]: In that case I think it’s the perfect time to break for lunch. So we have a break now for an hour for lunch. So let’s thank Paul. [Applause]

References

• Calcott, B., Pocheville, A., & Griffiths, P. E. (2020). Signals That Make a Difference. The British Journal for the Philosophy of Science, 71(1), 233–258. https://doi.org/10.1093/bjps/axx022
• Fodor, J. A. (1990). A Theory of Content and Other Essays. MIT Press.
• Goode, R., & Griffiths, P. E. (1995). The misuse of Sober’s selection for/selection of distinction. Biology and Philosophy, 10(1), 99–108. https://doi.org/10.1007/BF00851989
• Griffiths, P. E., & Wilkins, J. S. (2015). Crossing the Milvian Bridge: When Do Evolutionary Explanations of Belief Debunk Belief? In P. R. Sloan, G. P. McKenny, & K. Eggleson (Eds.), Darwin in the twenty-first century (pp. 201–231). University of Notre Dame Press.
• Hull, D. (1987). Genealogical actors in ecological roles. Biology and Philosophy, 2(2), 168–184. https://doi.org/10.1007/BF00057961
• McKay, R. T., & Dennett, D. C. (2009). The evolution of misbelief. The Behavioral and Brain Sciences, 32(6), 493–510; discussion 510-561. https://doi.org/10.1017/S0140525X09990975
• McLaughlin, P. (2001). What Functions Explain: Functional Explanation and Self-Reproducing Systems. Cambridge University Press. https://doi.org/10.1017/CBO9780511498510
• McLaughlin, P. (2002). On Having a Function and Having a Good. Analyse & Kritik, 24(1), 130–143. https://doi.org/10.1515/auk-2002-0107
• Stich, S. P. (1990). The Fragmentation of Reason: Preface to a Pragmatic Theory of Cognitive Evaluation. MIT Press.