 Hello and welcome back to Beyond Networks, the evolution of living systems. In last lecture I gave you a mantra, remember that mantra for the rest of the course. Structure does not determine function. Structure does not determine function. For two reasons. The first is that even simple, very simple sub-circuits in a network like the AC-DC circuit that we discussed last time can produce a wide range of qualitatively different behaviors depending on the exact details of how the interactions happen and the nature of the parts and their interactions in the network. Also, context matters if you put this little sub-circuit in a larger network. Its behavior would depend crucially on how it is integrated into that wider context of a complex large network. And so we cannot just dissect a complex network into its motives or sub-circuits, determine their behavior from their structure because they're simple, and then piece the whole network, the behavior of the whole network together again just as playing Lego with these little network motive, puzzle pieces, Lego pieces. So we need a different approach if we want to understand what a complex network is actually doing. What does this do? It's really hard to figure out. We can't just trace the interactions that have been through this. We can't just mentally rehearse or simulate the behavior of this network because it involves way too many factors, non-linear, unexpected behavior, and so on and so forth. So we need to figure out how to explain that and maybe we should also take a step back and think about what do these network models, these network graphs actually explain? Do they explain anything? And maybe we have to take one step even further back and ask ourselves, what on earth is a scientific explanation in the first place? Have you asked yourself this question? You should. I think it's a very non-obvious question. It's not clear what the answer is, and it's been the source of very long-standing philosophical debate. So in this lecture, we'll take a very quick excursion into the philosophy of what explanations are and what kind of explanations biologists especially use. The classic sort of account of scientific explanation was worked out by Carl Hempel in collaboration with Paul Oppenheim. The classic paper for this was published in 1948, and this account of explanation in science is called the Covering Law Model of Scientific Extension. It's what you probably intuitively would answer as scientific explanations. It's also called the Deductive Nomological Model, very complicated. Words will explain them in due time. And the best place to read up on Hempel's work is this little book, but it's sort of mysteriously awful cover drawing here. So this classical account is sort of making the following statement. It says explanations are arguments. That's very important. So an explanation is a logical and also a deductive argument. We'll come back to that in a second. Showing that the event to be explained, the explanandum is the phenomenon that we want to explain, like the rainbow that you see in the background. Was to have been expected on the basis of laws of nature. Okay, that's what nomological means. The deductive argument derived nomologically from a basic law. But also the antecedent and boundary conditions. And those laws of nature and the specific conditions that apply in a situation, they are called the explanance, sort of the tools, the arguments, the means by which you provide the explanation. So we use the explanance to explain the explanandum. What's important here is this fact about deriving explanations from law. So let's look at this beautiful picture of a double rainbow and think, how would Hempel explain a rainbow? What are the general laws he would derive a rainbow from? The laws of optics, the laws of refraction and diffraction. So that's not enough, right? You also need droplets in the air. You need sun rays coming in and you need an observer. Who is standing at exactly the right position to observe the rainbow. These are the specific conditions and antecedents in that previous quote. Together with the laws of optics, you get the sun coming in, being split up by the droplets like a prism, and if you're standing at exactly the right spot, you will see not only one, but two rainbows in the sky. Beautiful. And in this case, no problem. But remember, the pendulum model, Galileo, came up with this before Newton raised his laws. So basically, you can derive the pendulum model nowadays from the general laws of gravity. But Galileo, I mentioned that he didn't do that. He came up with the model without the laws of nature in the background. Keep that in mind. So let's simplify this complicated quote I gave you before. And the covering law model basically amounts to this. You have some general law, some particular facts. That's the exponents. And you get a phenomenon, an explanation for a phenomenon to be explained. The explainer. OK, but there's three reasons why this model doesn't apply in general. But in biology and the social sciences in particular, it's not a good model for those sciences. First of all, there's a problem with symmetry. So think about this sort of situation here. It's not a rainbow, but the sun's shining. And there is a flag on a pole. And that pole casts a shadow, as shown here in the picture. And from measuring the length of the pole and the angle from here, from the end of the pole, you can estimate its height by doing trigonometry. So you're using, again, sort of general laws of optics. Also, you're using geometry here. And you derive sort of the height of the pole. OK, but if you think about explanation and causation. So what you've just done is you've derived the height of the pole from the shadow. But what's really happening in nature is the shadow is caused by the pole. So it is the pole that determines the length of the shadow. But you've determined the length of the pole from the shadow. So there's a symmetry here that is totally OK in terms of Hemphol's account of explanation. But which is definitely not OK if you want to come up with a causal explanation of the length of this shadow. The next problem is the problem of relevance. There are a lot of causal interactions in the universe that are not relevant. I mentioned already earlier when we were talking about near decomposability. I mentioned the pseudoscience of astrology. You can actually come up with a perfectly physically robust account of how the planets and the stars affect your life on Earth. They do. Everything is connected. The gravity of all the bodies in the solar system have some finite influence on what you experience. But obviously to everyone, apart from those people who believe in astrology, those influences are absolutely negligible compared to the immediate sort of influences on your life that come from other people, your surroundings, your immediate surroundings, et cetera, et cetera. But you're much larger than the influence of the stars and the planets. OK, so here you have an account that's perfectly sensible in some ways, but completely irrelevant. And Hemphol's account of explanation cannot explain what is actually relevant for some sort of phenomenon to be explained. And then, of course, we come back to Wimpsat's rainforest ontology. Where are these general laws in biology? Natural selection. Maybe we're not sure. But usually, you don't have any general laws to derive any models from if you're a biologist, even worse in the social sciences. Because we had this discussion already, we live in this part of the world that is very messy, where the levels of organization are no longer clearly separated, where you need multilevel explanations that only apply locally to specific systems. So while physicists can have generalized laws and theories in scare quotes, because they're also just very generalizable models. In biology and life sciences, the neurosciences, and the social sciences, we cannot have that. So we only have specific models and perspectives relative to specific problems. So basically, the covering law model absolutely won't do in our case. It's not a good explanation. And as logical empiricism, of which it was a part, started to decline, people came up with new ways of looking at scientific explanation. One way is to say biologists don't have any explanations. The ultra-reductionist philosopher of science, Alex Rosenberg, he has an article saying that. Nice, right? Biologists don't really explain anything because they're not physicists. Rosenberg is one of these people that takes the wrong premise, follows strictly logical, very clear argument, and then comes to the wrong conclusion. He's interesting to read, but I don't necessarily recommend him. So if we don't accept his radical conclusion that biologists don't really explain anything, how do we go about explaining stuff in biology? Let's take the rainbow again. So what we could do is we could look at the rainbow as a system and see, does it have parts that cause the phenomena? Okay, and of course, we immediately see the droplets that now act. They're in the air and they act as prisms to split the sunlight into the rainbow. There is no general law. We can look at the behavior, the activities of each of those droplets, and we can notice that it sort of splits the light. So we don't need the general laws of optics to come up with a perfectly causal explanation of the rainbow. Okay, there are droplets in the air and they cause the light to be diffracted, split up into its colored components. This is called the mechanistic explanation. Let's try and make this a bit more precise, and I really recommend a wonderful article by Craver and Tabry on the Stanford Encyclopedia of Philosophy that I've referenced below. The quotes are from that article. So Craver and Tabry say explanation in this mechanistic account is a matter of elucidating that causal structures that produce and rely or maintain the phenomenon of interest. That's also what Galileo did with the pendulum, by the way, without any general law of gravity. He just looked at the forces that act on a ball. So that seems to be much more promising as a sort of explanation, a form of explanation in biology. But we have to ask ourselves what is this question again? What is a mechanism? It's been sort of haunting us for the whole lecture already, but I've never actually gone into this specific context and I've never explained what does it actually mean, a mechanism. And the problem is, as many concepts that we use in biology, mechanisms has many different meanings, two meanings in particular that we have to look at. They're very different. So there are at least two very different meanings of the term mechanism. And one that, as you may have noticed already, and slightly allergic to, is this idea of mechanism as a machine. So all the computer metaphors in biology fall into that, but also the idea that you have to explain everything in terms of molecules, okay? If you submit your paper, a research paper in the future, or you have done so already and reviewer two says, I reject this because you don't have a mechanism. And you can go and ask, what is a mechanism? And when they tell you it's about molecules, you can tell them they're wrong. They don't understand what a mechanism is and what role it plays in explanation in biology. Because there is a second meaning of this term. And this is mechanism as in some sort of causal explanation of a phenomenon. I've shown you this cartoon already in the context of statistical network analysis. So if we want to explain the behavior of a system, we need more than the statistical approach. We need a causal explanation. So in this sense, mechanism is sort of a post to these sort of general network analysis. And we'll come back to that point. But let me first explain a bit more about this difference between machine mechanism and what philosophers of science call neo-mechanism. It's a branch of the philosophy of science that has come up, especially since the year 2000, a little bit before already, that takes a new look at how we explain stuff in biology opposed to the covering law model of Carla Hample and Paul Oppenheimer. So machine mechanisms, we've been through that. So they emphasize that organisms are sort of the same as the non-living work. We're not made out of magic stuff. Descartes was in some ways right when he said, organisms are just mechanisms. But then Delamite said an organism is a self-winding clock. I've never seen a self-winding clock that does this in a form of how the organism reproduces itself. So there is a lot that goes into the self-winding term, right? But what this sort of machine mechanism does is it emphasizes that the stuff we're made of, it's antivitalistic. There is no magic stuff in living systems. Biological holes are determined in this view entirely by activities of their molecular parts. So we sequence everything, we do all the proteomics, and we suddenly understand how the organism works. Sounds familiar? In this view, we focus on efficient material causes and not final causes. We'll come back to Aristotle one module ahead from now. I won't go into this, but remember this. And this approach, of course, is firmly committed to reductionism. So we want explanations in terms of the parts of the system. For example, molecular biology. The only valid explanations are at the molecular level. We need to bottom out, as Ramiro 2 tells us to do. Neo-mechanism is a very different view of mechanism. The emphasis here is an alternative to general laws, which are based on the covering law for biological explanation. These mechanisms normally, they span multiple levels of organization. There's no problem with that. They don't have to go down to the molecular either. We'll cover that in future lectures. And mechanisms explain biological functions as part of an organismic whole. So you have to see those mechanisms as partial. They don't explain the whole organism. We'll definitely come back to this problem. But they're very good at explaining parts of mechanisms, such as developmental processes. This type of explanation, mechanistic explanation is not necessarily committed to reductionism. And it's compatible with higher level and systemic explanation. So one of the main differences here is that machine mechanisms are ontological. Remember that word? So they say something about the existence, the reality of living beings, while this neomechanism is more a way, a kind of a different way of explaining what's going on in living systems. So Dan Nicholson, whose paper I've stolen this from, has suggested to call mechanistic only the sort of clockwork machine mechanism, classical type of mechanism. And he invented the word mechanismic for this new type of mechanism, which is about causal explanation. This is all very abstract. So let me give you a few precise definitions of what a mechanism is. And this typical philosophical debate is very confusing. There are a lot of different sort of definitions out there. But out of those definitions, we can maybe crystallize our idea of what it really means. The first and earliest quote I give you is, by Stuart Glennon, who is looking at mechanisms in an ontological sense as sort of causal structures and reality. He says, a mechanism underlying a behavior is a complex system, a real system in his case, which produces that behavior by the interactions of a number of parts according to direct causal laws. So there's the word law in there, which is a little problematic. And also causality will come back to where the cause is. It's also a complicated topic. You can shift the debate to the level of explanation. Two biologists, George Fundasso and Edmund Rowe, do this in this 1999 paper, highly recommended, very underrated paper. They say mechanism per se is an explanatory mode. It's a way of explaining in which we describe what are the parts, how they behave intrinsically, and how those intrinsic behaviors of parts are coupled to each other to produce the behavior of the whole. So basically there are parts, they do something. That causes them to interact and they produce some sort of behavior or phenomenon. This is a classical quote from the paper that started the whole neo-mechanistic movement, McHommer et al. 2000, and they say mechanisms are entities and activities, again, parts that do something organized such that they are productive of regular changes. They produce change from start or set up to finish determination condition. It's almost like an algorithmic idea. Stuart Glennon revises and extends his definition in 2002. A mechanism for behavior is a complex system still that produces the behavior by the interactions of a number of parts, still where the interactions between the parts can be characterized by direct invariant change relating generalizations instead of laws. Okay, so he's sort of tuned in to that discussion about how mechanistic explanations are exactly not like universal laws. And then here, okay, this is something we're going to come back to in the next lecture, Bechdel and Abrahamson's definition. The mechanism is a structure performing a function in virtue of its component parts, component operations in their organization. The orchestrated functioning of the mechanism manifested in patterns of change over time and properties of its parts and operations is responsible for one or more phenomena. Stuart Glennon again, third try. A mechanism for a phenomenon consists of entities or parts whose activities and interactions are organized so as to be responsible for the phenomena. This is called a minimal definition of a mechanism. So let's parse that last most simple quote in all of them. So just listening to me reading out those quotes, maybe you have crystallized a few basic features of mechanisms already. So one is a mechanism produces some sort of phenomena. And when we looked at actual real systems, a few lectures back, we said a system is something that produces a pattern, is a patterned process, okay? So the phenomenon is some sort of pattern or behavior that we can recognize as something we want to study. That's quite obvious. A mechanism has to be made of parts. So to understand the mechanism, we have to focus on those parts and we have to decompose it into parts. So mechanism in this sort of sense is still a reductionist sort of approach. So we're looking at a system in terms of its constitutive components. It has to have some sort of causings. That's a weird word that occur through the activities and interactions of parts. We're identifying a causal structure and that structure is embedded in those activities and interactions that arise from the parts. And then there has to be an overall organization that's responsible for the orchestrated generation of the phenomenon. So all these activities have to work together in some way to produce the behavior of the whole. And here is where a lot of disagreement comes from. How does this work? How does this work? How does this work? How does this work? How does this work? We're discussing this and how complicated this is. And we'll come back to this idea that the system is more than the sum of its parts, which is not sort of explicit in here. It's not clear if neomechanists believe that. So as has become hopefully quite clear by now, mechanisms are a kind of heuristic model. Let's get back to that in a second. So let me give you a slightly alternative, different definition of a mechanism from Dan Nicholson's work, his 2012 paper. He says mechanisms are heuristic models that enable the explanation of causal processes which realize some biological function or phenomenon of interest. Okay. So tools. Mechanisms are explanatory epistemic tools, just like models. And formal models. Mechanisms can be formulated in different ways. Okay. The idea is a little broader than the idea of a model that we have before. So here is the work of Linda Darden, one of the pioneers of this neomechanistic philosophy. And she has a beautiful paper in 2002 that, that looks at how mechanisms are presented in the literature. And she says very often in biology. Mechanisms are presented as sketches. She calls this mechanism sketches. An example is shown here from Jim Watson. It's very nice because it's an early sort of attempt at explaining how proteins are made out of DNA before they had deciphered the genetic code. And it says there's some sort of complement replication. 1954. So this is after the 53 paper. And I know DNA can, can replicate like this. Some sort of chemical transformation. And then you get a protein from the oxyribose to ribose. You get the RNA. That's also replicating maybe. No, that's wrong. Right. And then you get a pro protein based on gamma holes. Those are not very known. Things anymore. These holes are part of the structure of DNA that we're supposed to be involved in the mechanism of producing proteins by being complimentary to the shape of amino acids and gamma even. Derived the fact that there are only 20. Amino acids that DNA encodes for only 20 amino acids from these sort of structural features. Not important. What's important here is that this sketch is incomplete. It has some wrong information in it. That's what a sketch means. But over time, people have elaborated the view of how proteins are synthesized. And you can see here what Darwin calls a mechanism schema. A schema is a drawing or representation of a mechanism. It's not a formal model. It's a drawing. But it is complete. And we know it is complete. It has no gaps. So you develop a mechanism you represent it first as a sketch. And then in biology, very often as a diagram like this, which is a type of a model, but not a formal model. But of course you can also represent mechanisms by formal models. Just like network graphs. Okay. Because they lent themselves very easily to this. Are the nodes in the network and the activities and interactions are represented by the edges of the graph Especially if it's a directed graph like this one very straightforward So we're tying things together again that we've been through already at this point But and here's the mantra again structure does not determine function repeat after me Structure does not determine function. So to understand the function of this network we need to understand what it does and I think I told you a couple of times already that you cannot do this by just looking at The network, but you can learn stuff from looking at the network alone So it's important to point out that you need a mechanistic explanation in this neomechanistic sense that I just introduced If you want to know what the network does But there are other types of explanation in biology as well We'll introduce some in the future, but I want to mention one at the end of this lecture right now And it's called the topological explanation It was first suggested in a paper by French philosopher Philippe Eumann and So what Philippe has done Is he thought about these networks and how common they are and How they actually do explain some features of a system So he defines a topological explanation as Follows he says network analysis rough theory abstracts away from causal interactions Okay, so we do statistics on degree distributions and stuff like that To pick up topological properties for example the scale-free structure of a network and Those properties they explain some features of the system such as a systems anti fragility its resilience its robustness Okay, we've been through that in the first lecture of this model So go back to that if you need a reminder basically these topological properties can be defined on a graph or other abstract mathematical spaces Keep that in mind. It doesn't have to be clear what that means. We'll come back to topological explanations in a little bit So basically here you you don't even try to explain what the system does But you want to explain other features such that it does what it does robustly That it learns from insults from perturbations other such things and That's completely Viable, so I have to take back my provocative statement a few lectures ago that this is not an explanation This is not a mechanism It is an explanation of some kind a topological explanation But how do we get from here to a causal mechanistic explanation of what the system does? We have gotten no further But by rephrasing the problem I think We can now approach it from a slightly different way. So in the next lecture, I'm going to talk about What is it? that makes a mechanism explanatory in Terms of explaining what the system does. I hope you'll tune in again Thanks for listening. Bye now