Watch Episode Here
Listen to Episode Here
Show Notes
This is ~1 hour conversation with Daniel McShea (https://scholars.duke.edu/person/dmcshea) and Gunnar Babcock (https://gunnarbabcock.com/) on topics of biology, evolution, information, causation, and ethics in the philosophy of mind.
CHAPTERS:
(00:02) Bow Ties, Ecology, Embryos
(10:02) Mathematics, Constraints, Dualism
(18:23) Top-Down Causation And Memory
(29:38) Human Motivation Versus Machines
(34:26) Solms, Feeling-Based AI
(41:13) Future Beings And Morality
(46:10) Life, Machines, Continuum
PRODUCED BY:
SOCIAL LINKS:
Podcast Website: https://thoughtforms-life.aipodcast.ing
YouTube: https://www.youtube.com/channel/UC3pVafx6EZqXVI2V_Efu2uw
Apple Podcasts: https://podcasts.apple.com/us/podcast/thoughtforms-life/id1805908099
Spotify: https://open.spotify.com/show/7JCmtoeH53neYyZeOZ6ym5
Twitter: https://x.com/drmichaellevin
Blog: https://thoughtforms.life
The Levin Lab: https://drmichaellevin.org
Transcript
This transcript is automatically generated; we strive for accuracy, but errors in wording or speaker identification may occur. Please verify key details when needed.
[00:02] Gunnar Babcock: So did you make it through, Dan?
[00:04] Daniel McShea: What's that?
Gunnar Babcock: Which 1 1/2 did you make it through?
[00:06] Daniel McShea: The bow tie stuff is what I have things to say about. I have something to say about cognition and consciousness, the attempt to reform everybody's way of talking about these things with a series of pointed questions. Well, let's do bow ties first, if that's all right. Unless you have an agenda.
[00:27] Michael Levin: I'd love to hear what you have to say about any of them.
[00:30] Daniel McShea: Let me remind myself what I have to say about them. I get it. Looks powerful. How stupid not to have thought of that, as Huxley said. I have some questions about how it works. In ecology, there's no encoding, is there? There's no compression. There's an expansion stage, an inflation stage, as you nicely call it. But tell me, is there a compression stage?
[01:03] Michael Levin: I don't know how specific you want to be about ecology, but if we include evolution, then I absolutely think so, right?
[01:11] Daniel McShea: I meant succession. Sorry, I should have been more specific. In ecological succession, where you clear a landscape and the new stuff moves in and the ecosystem inflates from nothing. But there's no prior compression, right? I'm testing my understanding here.
[01:32] Michael Levin: I'm not an ecologist. To me, by virtue of what you've done to the environment, clear out. Some sort of niche construction. Presumably the previous occupants have transformed the nutrients in the soil, the water escape. Does that sound plausible at all? The bowtie thing is mostly applicable to systems that go in continuous. If the succession thing is really a one-directional process, I don't think it's applicable there.
[02:21] Daniel McShea: So recovery from storms, it sounds like what you're describing, and I think that works. I was imagining volcanoes, Mount St. Helens completely clearing the landscape around it, and there ain't nothing there but volcanic rock, and the system does reconstitute, which is, from the perspective of the storm story, interesting because they follow a different trajectory, a very different trajectory than the storm regeneration does. The end is further away. They get to the same place in the end, which makes you wonder what difference it actually made in the long run. In the short run, it makes an enormous mistake.
[03:04] Michael Levin: I like that last point you made because I was thinking about this the other day that if you do have this bow tie thing where you've got the construction stage in the middle point and the expansion, if the expansion is really good and creative, you can imagine unmooring it completely from the engram prompt and you can start to apply it to almost anything. So you can get rid of this side — if you've got the process of creatively interpreting some physical state and then expanding from that, you could do that de novo in almost any state that you want.
[03:48] Daniel McShea: No, that's heresy in ecology because that means that the critters moving in are pre-programmed to do succession at some level of generality.
[03:58] Michael Levin: I would think it's the opposite. I don't think they're pre-programmed. I think they're solving the problem that they're faced with, which to us looks like succession, but to them, they are thrust into a new action space of opportunities. They are from scratch, like our Xenobots. You've got your genome from the past. It's helpful for some things, but not for a lot of the things you're going to be doing. And you just have to figure it out on the fly. That's how I would see it.
[04:30] Daniel McShea: OK, go ahead, Gunnar.
[04:32] Gunnar Babcock: Dan, you should give Mike more of the background of how the Zeffel plays into this.
[04:42] Daniel McShea: Oh, you better do it. I'm not following it.
[04:47] Gunnar Babcock: This is you and Robert. As far as the volcanic landscape, without any constraints, you have an environment without any selective screening going on. You have expansion. That's the background thought experiment, as I understand it.
[05:12] Daniel McShea: But it should come out different every time, shouldn't it? If that's all there were to it. If we were absent constraints, it would come out different every time, except to the extent these things know where they're going, that the plants moving into the Mount St. Helens desert landscape have buried in their genome somewhere the knowledge of how to create a spruce forest. That's the heresy. You tried to answer that, Mike, and I didn't follow you. Could you try again?
[05:38] Michael Levin: I think that the first thing I would say is that wherever it's coming from, I don't think we should assume it's the genome. That would be step one.
[05:51] Daniel McShea: Yeah.
[05:52] Michael Levin: Because I think that there are an enormous number of affordances that come from mathematics, laws of computation that are implemented by machines that are encoded by genomes. But the actual patterns that ingress through those are not directly in the genome in any sense?
[06:13] Daniel McShea: Yep.
[06:14] Michael Levin: I think it's not unreasonable to say that you have a problem-solving agent, which, faced with a certain scenario, was going to solve it in some way, but that's just the default way that it's going to do it. If you change something else, it may find other solutions.
[06:33] Daniel McShea: Ecology responds, there's no agent in the case of Mount St. Helens, right?
[06:40] Michael Levin: In the case, I'm not leaning on the agency of Mount St. Helens. I'm leaning on the agency.
[06:47] Daniel McShea: The biota that's moving in there is not a thing to have any agency to reconstruct a successional ecosystem yet. It's a bunch of plants growing out of rock at the moment.
[07:00] Michael Levin: I've already gone on a different path, because I absolutely think there's agency there.
[07:07] Daniel McShea: It's located for me, if you would.
[07:11] Michael Levin: In the cell, in the first cells that appear when the first seeds wake up, I see it as a collective intelligence. In the same sense as now. The degree of the intelligence I'm making no claims about, because one would have to do experiments. You have to probe it to see what it can do. Does it do delayed gratification? Does it do associative conditioning? Does it do problem solving? What kind of problem solving does it do? But I don't see why you would immediately say that there isn't any agency there.
[07:45] Daniel McShea: So in the case of a developing embryo, we've got this single cell with a bunch of molecular stuff floating around in it and a genome. That's ready to do stuff. Critically, from both our points of view, mom is sitting there ready to give the first instructions to start the inflationary process, and not just start it, but tinker with it, at least in the early stages before the thing gets really going. That's what's absent in the cleared space around Mount St. Helens, ecologists say: there's no mom there to trigger or to guide the early stages.
[08:25] Michael Levin: There are numerous species where the egg is created and then sits for as much time as you want. Mom is long gone. The mom's organism had a role in prepping the cytoplasmic components and everything else that's in it. But after that, it can be dormant. It can be in this desiccated state that some eggs can go into, and then something wakes it up and it rolls.
[08:56] Daniel McShea: But the instructions that I was too glib in calling it "mom"—it's there in the cytoplasm. The genome can't do the whole thing alone. It doesn't; it needs more than a match to light that genome fire. It needs instruction from above, and mom has given it in the cytoplasm of the egg, if not in anything else. So the top-down causation we need to guide this thing is sitting there in your insistent single-cell thing, ready to go. So this is not speaking for me, but speaking for modern ecology. They want to know where's the insistent, where's the package of stuff that's going to guide this bigger-than-cell, bigger-than-individual-plant-scale thing along the way. I think what you need to argue is that deep down in that cellular package of cytoplasmic stuff and genome is preparedness of some kind for something way bigger than itself at the ecosystem scale, which, from my point of view, is not crazy, but for ecology that's crazy.
[10:02] Michael Levin: I see it. My particular flavor of the craziness here is that a lot of it comes from wherever the laws of mathematics come from. When people ask, "Where is it?" you can make all sorts of things: the Galton board, fractals. You've got this seed and then this creates this very specific shape. You say, "Where was it?" and you're not going to find it in the physical world. That's not where it comes from. We can ask what kinds of things ingress that way. Is it only laws of triangles and things like this? Or is there more information that evolution can use as an affordance to exploit once you've made the right physical shapes?
[10:51] Daniel McShea: Gunnar, can you jump in here? Because I have the feeling we're talking past each other, but I can't put my finger on it.
[11:01] Gunnar Babcock: I can jump in on a couple of things. From our last talk, as Carrie over here, the physicalism versus dualism issue is arising. I think that's curious. We could think about that, particularly when it comes to mathematical principles as guidance. For me, I would think of those as physical, or at least not dualist. Now I know we can be Platonists about numbers, but I think there are ways of cashing that out as a physicalist. So we can go in that direction, but in trying to help the impasse here, Dan, I take it that you're trying to drive at what are the external constraints that might be out there that you're not going to get internally from whatever.
[11:57] Daniel McShea: There's got to be something out there because the internal stuff doesn't know where it's going.
[12:02] Michael Levin: Here's an example. The cicadas that come out 13 years, 17 years. The explanation for why those specific years are facts about prime numbers. To go back to what Gunnar just said, I want to be Platonist about a lot of things, not just numbers. This is fun to discuss; I'm not sure how physicalism could deal with it. I think that—here's the dividing point I would make. If we assume that the properties of the physical universe are defined by some small number of constants that are set at the Big Bang, something like this, then the question is how many of these other affordances would be touched by twisting the control knobs on those values? To my understanding, no matter what you did at the beginning of the Big Bang, and you could have had a very different physical universe, the truths of number theory, facts about prime numbers, would not be touched by this. That, to me, is what makes it fundamentally a non-physicalist theory, because you can't get at these things by turning the knobs that physics likes to turn.
[13:23] Daniel McShea: Gunnar and I have had this conversation recently. I'm more of a dualist. God, I feel awful saying that out loud, because it's horrible to my eyes.
[13:36] Gunnar Babcock: I would present a little bit of pushback. It depends on how we're thinking about dualism, I think. If you're thinking about dualism in the way that I would be inclined to think of it, which is to say that any sort of empirical science is just not going to be able to touch the other non-physical substance, then I would say that this isn't dualism, what we're talking about, because it seems like this is something completely approachable through empirical methods. If what you mean is something that's non-material that I can't go out and grab, then I very much agree. But I don't think we have to go dualist to get that. I think that we can still be physicalists. Whether or not you're a platonist about numbers, Tim Maudlin says that, if you're a mathematician and you say, "I'm not a platonist about numbers, but the other six days of the week you go out and act as though they're very real things." Well, you're treating them as real entities. I'm sympathetic to that view too. I do very much think that it's very hard to say that numbers aren't real or that we're not dealing with them in a real way. And yet we give them such importance in all of our methodologies. So I am very sympathetic to this position that it is a real power and that there's something not captured in a materialistic framework about them.
[15:14] Michael Levin: I think the first definition you gave, I take that to be not so much dualism as more of a kind of mysterianism. I think a minimal dualism is simply the claim that you would need different ways to investigate some of these things, and we have some of those ways in mathematics and so on, but it means that it isn't purely physicalist because you're going to be talking about influences that are not to be found by the tools of physics, by the tools of biochemistry, by the tools of genetics. These are the laws of computation, the laws of mathematics. Numbers are one thing. I don't know if you've ever seen any of my stuff on Hali plots, but a very small formula generates this incredibly rich pattern. We don't have the vocabulary yet, even for that kind of thing. It's just not the right question. I see our Xenobots and Anthrobots and all those kinds of things as periscopes to peek into that space. I think they are exploration vehicles for a rigorous research program to say other patterns exist in this space. I don't hold any kind of a mysterian view that we're never going to be able to touch it. I think we can touch it, but we have to. Some people, when I discuss this with people, I say here's this pattern that happened. They say that's just the fact that holds, let's say about a network; you write about networks. Once we accept that there are things that hold, then one of two things. Either it's a random bag of stuff that holds, and when we come across it, we put it on a list and call it emergence, but we have to wait till we stumble across it. Or we can say that there is this space of things that hold, and much like mathematicians who make a map of mathematics and how the different things relate to each other, we can say this is an actual space that has a metric to it, and we can investigate it systematically; we can hop along the space. That's my only claim. It's not a mysterian claim at all.
[17:20] Gunnar Babcock: I think then we're in pretty close agreement. I think it's just a terminological way that for me when I think dualism, I think of it in the traditional Cartesian minds, which definitely fall outside this class. Because for Descartes, the traditional dualist, those are just things that in principle I could never use instruments to discover things about. They're outside the realm of experience. I think we're in a lot of agreement then. I think the traditional way we think about physicalism needs a much better imagination about what we allow into that space. Very much the way you're talking, Mike, can be cashed out in physicalist terms once we get away from this substance-material notion that's so bound up with it.
[18:23] Daniel McShea: And thank you to you both for backing us out of that cul-de-sac. Can I come back to bow ties? Everybody in this room is comfortable with top-down causation. Coming back to your bow tie, it seems clear to me that both steps in the bow tie process, both the compression and the inflation, require top-down causation. That was my point in bringing up the ecology and the Mount St. Helens thing. In the case of genomes and evolution, what's the top? It's ecology, acting downward on organisms or on lineages over time. That's easy. In the case of development, it's these higher level fields, then we agree on what those are, I think, that are acting downward on the genome, among other things, on the whole biochemistry here. Turning to the inflation process, it too is going to be guided from above. The reason for bringing up the Mount St. Helens thing is it's hard to point to the top, to the thing that's acting downwardly on them, whether they're plants or bots. I feel the constant need to look up and see what it is that's doing the downward causing. Is this question making sense?
[19:59] Michael Levin: I think it is. I think that if we expect living things to do the things they were selected to do, then you always are looking for what in your history prepped you for this thing. I tend to think of it much more as a problem-solving intelligence process where you're prepped to do things you've seen before, but you also have the capacity to respond to things you've never seen before.
[20:28] Daniel McShea: Yes. Not everything, but some things, yep.
[20:31] Michael Levin: But some things. I think in a scenario as unusual and with as much novelty as this Mountain Saint Helens thing, what you could be looking at there is the high effectiveness of—you've had prior history that has led to building a pretty good problem-solving agent. Now you're faced with a problem you've never seen before, and what you've inherited from the past is not a specific "what do I do now?" but rather skills at "what do I do when I don't know what to do." That's a typical way that roboticists look at things like this, because you want to make devices that do things when you haven't pre-prepared them for it. These are generic strategies of "what do I do when I have not seen this particular case before?"
[21:20] Daniel McShea: There has to be something in common or you can't possibly know what to do, but it can be pretty minimal.
[21:25] Michael Levin: It can. In robotics, they use this empowerment principle, which is do whatever looks to be maximizing your choices in the future. So whichever move leaves you the most open choices, do that. I'm not saying that explains the Mount St. Helens thing, but I'm saying that's a very minimal, extremely generic policy that you can embed in all different kinds of media to say, when I have no clue what to do, I'm going to do the thing that gives me more options.
[21:56] Daniel McShea: So then there's the perspective question with the Mount St. Helens thing. If you take the perspective of the ecosystem, which doesn't exist yet right after the volcano has gone off, but you look at it as a pack of nutrients — and take the thermodynamic view — a pack of nutrients that's sitting there at high potential, wanting to be dissipated to space, asking itself, "What am I going to do now?" The answer is, I'm going to get a high-growth-rate, nutrient-independent plant to grow in me. I don't know what plant it's going to be. I don't care. In these awful situations where there's nothing, I'm going to sit here forever not dissipating to space my latent heat at all. I need something like that to come along. And as soon as it does, I'm going to give it whatever it needs to set the process in motion. This is spooky ****. But you're nodding and that scares me.
[22:57] Michael Levin: I'm not scared off by it. I love the set of examples and I need to think about that more from the thermodynamic side. The version of this that I've been thinking about is memory patterns in a cognitive system and looking at things from the perspective of the patterns to say if I am a thought pattern and I want to not only persist, but thrive into the future and open into new problem cases. What is my host? Well, how do I have to modify myself? I'm in a caterpillar right now. That thing's going to rip up its brain and turn into a butterfly. What do I need to do so that I can be remapped into that new world and be literally reborn into a three-dimensional world from a two-dimensional world and have this other kind of life. Staying the same isn't going to help, but I need to remap and what are the properties that I might have that enable me to come through like that?
[23:59] Daniel McShea: Let's stick with your memory case for a moment and see if this leads anywhere. It might not. You've got this thought pattern that wants to maintain itself in a strange environment. Let's go one level up from that because guiding the thought pattern is going to be some motivational structure. I'm miserable. I've got all these thought patterns underneath me. I'm going to nudge them in a direction that historically yields decent results on average. Or I have no idea what to do. I'm going to do something and see where it leads. But this higher level thing has things like memories and thought patterns and habits and so forth all built underneath it. They have their own hierarchical structure. I'm agreeing with what you said. I'm inviting us to go one level up to the motivational structure.
[24:50] Gunnar Babcock: I would jump in because I think, Mike, the way that you're thinking about memory is a much broader notion rather than just cognitive thought processes. I found this really interesting because in the work on philosophy of memory, it seems the way you're thinking about it fits within the most general paradigm of how we think about memories. In the philosophical literature, nobody has a broad and expansive view as the one that you're presenting, which is really interesting. But definitely within cognitive processes, then, Dan, the question that you're raising as far as some outside motivational space as a want, as giving the, what you would call the oomph to move it forward, is necessary. I wonder, Dan, are you asking as far as, in the caterpillar example, the thought ends up being a developmental pattern into the future, which I think might...
[25:55] Daniel McShea: It's a developmental pattern in the pupa.
[26:01] Gunnar Babcock: Motivational want there, like a human sort of thing.
[26:04] Daniel McShea: We have to put Stan Salty-style brackets around things, and then talk about them. It's the same sort of top field causing a thing within it that's going on here. I'm sure Fred Neihat, our butterfly guy, would be fine with this notion of hormonal fields generated within the pupa that guide the early development of a separate set of set-aside cells that are going to become the butterfly.
[26:41] Gunnar Babcock: What I love about this is that — Mike, correct me if I'm wrong — here, in the most general notion of memory, the philosophers who think about memory treat the general catch-all as any modification of an agent's behavioral tendencies as a result of its experience counts as memory. But they all think of it just in these cognitive processes. Within that definition, your conception of memory fits really well, but none of them are thinking of it in these broad developmental terms. There's important stuff to be said there.
[27:26] Michael Levin: I completely agree. I was looking at your paper again today, the machine wanting paper, and I love the first two sentences. "So what would you like to do this afternoon? Not a machine in the world can honestly answer that question." I started to think about what would it take for us to honestly answer that question. I think that if you ask a psychoanalyst, Mark Solms or somebody, they would say you can't honestly answer that question because what you actually want to do this afternoon is a product of all kinds of sub-components of our psyche. This is my attempt, as Dan said, to go up to more motivations and so on. What are the sub-components that have various goals and various degrees of ability to look forward in time? It reminds me of a short piece I read a long time ago by a therapist who was working with people with dissociative identity disorder. He has a patient who comes in and says, "All right, I've got this other personality that pops up and it's killing me because he's going to get me fired. He goes drinking in the middle of the day. I have to leave my job. It's terrible. I'm going to get fired. Let's do something about this." They start working on this. They call it integration therapy. Some weeks later, a patient comes in and it's the other personality. The other one says to the doctor, "Hey, what's this I hear about integration therapy?" He says, "We're going to get you guys integrated." The other asks, "Where am I going to be when we get integrated?" After some hems and haws, the answer is, "With any luck, you'll be gone." "What do you mean? Have you heard of the Hippocratic Oath? Make the other guy gone. I'm the one, I'm the fun one. I enjoy my life. All he does is work all day." You start thinking there's this spectrum: you could have some fleeting thoughts, then persistent thoughts. You can have a personality fragment that's got more agency than a persistent thought pattern.
[29:38] Daniel McShea: We're a mess.
[29:39] Michael Levin: It's probably exerting subtle niche construction in our life and shifting some things around to me. And certain things more likely than other things. What is the motivation for whatever it is that you want to do this afternoon?
[29:58] Daniel McShea: I agree it's not all that accessible, but I would insist on the viewpoint still being right that it's going to be a coalition of various pieces of me that are going to get together and swat down other smaller coalitions of pieces of me and decide, I want to go for a walk or I want to play tennis, but somebody's going to win this battle and I'm going to give an answer and it's going to be authentic in the sense that it's not random. Whereas the machine giving this answer, standard serial architectures anyway, is either going to be hardwired in or it's going to be random. Take your pick. Does that fit you?
[30:36] Michael Levin: Almost entirely with one proviso. I completely agree that the standard serial architectures that we use today are not the kind of thing that you would want to execute. I'm in total agreement with that. Over the last year or two, I've become much more agnostic. I have much less faith in our ability to predict ahead of time what a given architecture is really doing. We recently published something looking at sorting algorithms. These are dumb, bubble sort, those kinds of things. Basically what we're finding is that it does the thing that you thought it was doing. But if you look at it a certain way, you find out that it also has some really interesting competencies that you didn't expect. These are not just complexity or unpredictability; these are problem-solving capacities. They are solving problems in ways that your algorithm did not tell them to do. And so my lesson from this is that if we don't know what bubble sort is really doing, I don't know that we can know what more complex architectures are doing.
[31:57] Daniel McShea: It's hard to argue with a claim that we don't know, because especially in these complex systems, we don't know. But here's what never happens with these, even with the large language models with their parallel processing and internal nodes with fancy architectures that are generalizing all this kind of crap. They never say when you ask them to recognize the face, "nah, I don't feel like it." They never say anything close to that. They demonstrate no internally driven higher level structure that can drive a lower level structure.
[32:36] Michael Levin: If I had to bet money today, I would agree with you, and I think that's most likely correct. I leave room for the fact that we may not have asked the right way. In the sense that, doing ethology with species that are quite different from us, you say, well, this thing doesn't do anything. How much of that is it and how much of that is us? And how much are we not looking in the right space, asking the right questions? I worry about making those kinds of conclusions too strongly. I think they're probably correct. My intuitions go the same way.
[33:16] Daniel McShea: I hear what you're saying.
[33:19] Michael Levin: You and I function on the laws of chemistry. I've seen the laws of chemistry. That doesn't look too impressive. I really don't think there's anything there. So for the same reason we've learned to be skeptical of that reductionist view and the idea that we can tell what's going to happen, we should extend that same humility to Albert.
[33:45] Daniel McShea: And part of the tone of voice that I use when talking about this, and my stridency, comes from the fact that first I believe it's likely true, but also because I want to give a jolt to these people and get them to think about motivation and not just computation. Because computation, and even cognition the way it's often used, are completely value neutral. That kind of language is going to be insufficient for the machines of the future, much less when we're talking about organisms. These are going to be valuing things. And I'm ****** *** at them. That's where my tone comes from.
[34:26] Michael Levin: I completely get it. The person who's working on that is Mark Solmes. He's doing exactly this.
[34:37] Daniel McShea: So tell me more. What should I read? I read a little bit of Psalms, his argument that consciousness is nothing but a feeling, which I thought was just brilliant, but I need to read more. What else should I read?
[34:47] Michael Levin: He's a brilliant guy, and I'm not going to try to paraphrase his many thoughts here. We should have a discussion with him himself. What he's emphasizing, and he's done a lot of work on the neuroscience with patients, is the idea that in the frontal cortex it's more about what the conscious state is actually about. Conscious state itself is generated in the brainstem, he claims. The idea is that it's all around feeling and affect; he's very big on primary feelings and affect, and that's what drives it. What we're doing is the palpating, the uncertainty that you have around how to improve your multi-dimensional things that you're viscerally interested in.
[35:37] Daniel McShea: Yeah.
[35:38] Michael Levin: I don't know any of the details of how he's doing it. It's not published, but he has a team working on an AI system that is fundamentally based on that.
[35:49] Daniel McShea: I had no idea. I'll just Google him and read everything.
[35:54] Michael Levin: I'm not sure it's published. This is something that he's seen a number of times in our discussions. I don't think they've published it yet.
[36:02] Daniel McShea: So you've talked to him directly?
[36:04] Michael Levin: Yeah, many times, many times.
[36:07] Daniel McShea: Cool.
[36:08] Michael Levin: We can do a chat together. You can see some of it on my YouTube channel. It's a bunch of conversations with Mark and various people.
[36:16] Daniel McShea: I haven't even seen your YouTube channel.
[36:20] Michael Levin: I'll send you a link. Mark and Mark have talked about.
[36:27] Daniel McShea: One little bit I read of him made me think he's on to something here.
[36:31] Michael Levin: I think he's on to a lot of things. For myself, I'm concerned because I think it might work. I think what you were talking about and what he's talking about is the right way to create machines that matter, but you can create beings that matter. The issue is that we obviously have beings now that are mistreated. The problem with that is it becomes trivial to make enormous numbers of beings that matter. Just from a moral perspective, I worry about that.
[37:24] Daniel McShea: Disposable people, yeah.
[37:26] Michael Levin: Well, they're disposable beings. I don't know that they're people, but.
[37:29] Daniel McShea: Well, I meant people in the broad sense, yeah.
[37:33] Michael Levin: Because wouldn't it be nice if, to me, the better outcome would be that if it turned out that we can make AIs like language models that have extremely minimal agency. And yet they're useful tools and great. We don't need to worry about it. It might turn out that if you do it that way, you actually don't get a technology that's sufficiently useful, and then people push forward towards the really bio-inspired. That's the risk of a truly bio-inspired approach: if you do it right, to whatever extent you get onto it, I don't know, trillions of new beings that...
[38:18] Daniel McShea: You think the situation that we'd be in would be any different than if the human species had speciated in a serious way 200,000 years ago? We'd have two smart species very similar to each other, potentially, but not necessarily at each other's throats.
[38:36] Michael Levin: I completely agree. I don't think any of these problems are unique. I've written something about this, talking about all the aspects of the ethics around AI and synthetic morphology that are not unique. A lot of people find that very disturbing because they think about these as new problems. If not for this, we'd be cool. But these are fundamental issues. So I agree with you. But the one different thing here is that these are much easier to copy. It's just going to be extremely easy to make unlimited instances. And until then, one of the worst things is commercial farming of mammals for food, right? Pigs. We know pigs are intelligent, so it's a nightmare. But at least with that, if you want to make an extra pig, there's some work to be done. You have to put some work into it. And that's at least a little bit of effort I worry is not going to be here in this case. You can just spawn unlimited numbers.
[39:50] Daniel McShea: I have sympathy for pigs. I have much less sympathy for mosquitoes. I have virtually none for the plants growing outside my house in my garden if I had a garden. And pretty much zero for rocks and bacteria. Are you worried we're anywhere near creating something on the level of a pig? Or are we still in rock plant mosquito world here?
[40:15] Michael Levin: I think we need to be humble here, but my guess is that the vast majority, if not all of what we've seen so far is somewhere in the rock-to-mosquito territory. I'm not worried yet. However, what I am worried about is that the very few people that are thinking about this are in fact going to ratchet this thing up like crazy. And I don't think they will be pigs, but there's no reason why they can't be somewhere in the space of possible minds that matter at a similar level.
[40:54] Daniel McShea: When you say matter, you mean matter to themselves?
[40:57] Michael Levin: Matter to themselves and thus matter to anyone.
[41:01] Daniel McShea: Because if it doesn't matter to itself, it shouldn't matter to us, right?
[41:07] Michael Levin: Well, I mean, yes.
[41:09] Daniel McShea: Sorry, Gunnar, I cut you off. Go ahead.
[41:13] Gunnar Babcock: In the literature on this, has anybody applied or thought about it in the context of a non-identity problem or in terms of the moral weight of future persons versus present persons?
[41:33] Daniel McShea: Say more because I didn't follow that.
[41:35] Gunnar Babcock: There are two unique moral problems that arise in this. I'm a philosopher of science, not an ethicist, but I am familiar with some of this. The two problems are: one is, why should we be assigning more moral weight to persons? Here I'm speaking about persons in the broad sense; it could include pigs, however you want to draw those boundaries. But why should we privilege the ones that currently exist over future ones and the ones in the past, conceivably too? Why should more moral weight be given to me and you rather than people who are to exist in the future at some point? If we're going to do a calculus, all of them should be taken into consideration equally. That's one problem. The other problem is the non-identity problem, which is that it seems we have alternate futures where I could bring certain sets of persons into existence or other sets into existence. It seems like we often assign more moral weight: I should bring about the future in which there isn't a horrible climate disaster because that would be a better future and people will have a better life there. But it would be a different set of persons in that future than the future where we don't correct all the climate nonsense, and a different set of people come to exist. So why should I privilege one set of people over another set of potential people? That raises another quandary you can get into ethically. That's the non-identity problem. The other one's a different problem that's brought about in some of the clients. I wonder if, because these seem to be relevant to the issues you're raising, Mike, with AI.
[43:37] Michael Levin: Those issues arise. And people have argued that there's some degree of responsibility that we have for enabling future beings with a higher capacity for thought, enjoyment, reaching their potential, that we ought to be doing this work to enable them to exist. I have some sympathy for it. I think there's something to be said for that. But these are tough problems. And I think the first step to even have a good discussion about this is the kind of thing that we've done here, which is to broaden the possible embodiments of beings that matter. So we can actually start thinking about this, because it's outside of the circles that I spend most of my time talking to. People find it really hard to imagine minds unlike their own. They look at something, whether it be a cyborg or a synthetic morphology example or AI or whatever, and they say, "That's clearly not human, good enough, they're fine." It doesn't have to be a human to matter. There's a much larger space.
[45:00] Gunnar Babcock: I was incredibly sympathetic and very much in agreement with what you pointed to and the lack of humility in distinguishing life from non-life and all of the problems that arise there and what limited mindsets are brought to that. And I think that exactly what you're driving at in a lot of that work is a multiply realized conception of the set of things that can have goals or have agential capacities really needs to be broadened. We really need a far, far less biocentric view on all of that, because otherwise we're going to miss full swaths of potential there. It's really unfortunate when those viewpoints are carried forward, even if ultimately we discover maybe there is no moral consideration for X or agential capacity with X, but we need to expand the way we're thinking about it in major ways, given the advances that are being made right now.
[46:08] Michael Levin: What's, go ahead?
[46:10] Daniel McShea: I just gave a talk at a meeting of extraterrestrial intelligence, and the one thing I said in the whole forty-five minutes that got everybody nodding in agreement was that all the aliens in science fiction are humans in funny clothes. There's very little science fiction imagination being devoted to creating truly motivationally different creatures. The space of the possible is enormous, and we haven't begun to explore it.
[46:40] Michael Levin: I agree with that. One of the ones that I really like is Solaris, Lem's piece. That's pretty darn alien. One thing that's really surprised me. One of the communities that really doesn't like this kind of view, oddly enough, is the organicist community that has been battling for how many years with the mechanist molecular biologists to say things are not machines. They've been doing great. They're not simple machines. And so they've been doing all this hard work. When I come out and I say, on the spectrum, here are some machines and here are some majestic, living beings, this really triggers people.
[47:32] Daniel McShea: They don't want a spectrum.
[47:33] Michael Levin: They don't want a spectrum, and they're worried that this undoes the hard work they've done over the years by pulling these things apart.
[47:41] Daniel McShea: Oh, yes, of course they're worried about that.
[47:44] Michael Levin: Because we've spent all this time fighting this thing with accounts. Now you're putting computers on the same thing. You're going to undo all of it.
[47:54] Daniel McShea: How about if three of us just agree not to use the language of machines anymore when talking about our very simple artificial devices?
[48:04] Gunnar Babcock: Yeah.
[48:05] Daniel McShea: I like these people. Go ahead, Gunner.
[48:10] Gunnar Babcock: I think there's this problem where there's an impasse that at least I've encountered, Dan and the new mechanists we've encountered, like Bechtel, where what they mean by mechanism now really is no longer connected with machines. It's just this downward analysis of mechanisms where you try to decompose into subparts. And that I think is completely compatible with both organicism and machines. There's no distinction there. It stands apart from this idea that somehow machines are ontologically different than what we find in life. The new mechanist framework is more the question of upward versus downward causation. That's what's going on in that literature. I'm nodding in agreement the whole way reading your piece, Mike. I've written similar things, saying that I think even attempting to come up with a good definition of life that is somehow exclusionary to machine stuff is a problematic enterprise to begin with. I personally am almost for just ditching the life category as an ontologically relevant thing that we ought to be going around and looking for. I don't think it's helpful.
[49:55] Michael Levin: Exactly. I've said exactly the same thing. I mean, it's so weird, especially from a biologist. I was at a conference one time where we were talking about the nature of computation, and there was a guy arguing that there is no such thing. I said, "So what do you do again?" He said, "I'm a professor of computer science." If he can say there's no computation, we can say that life is not a category. I have this amazing graphic artist that sometimes does things for us, Jeremy Gay, and I've asked him to make something like those old-timey pictures of phrenology, when there's a guy with big calipers measuring somebody's head. I don't have it yet, but I asked him to make something like that with a cyborg — he's a human, but a bunch of his brain is replaced by whatever. There's somebody with these calipers trying to measure it to see if it's more than 50%, if he qualifies for human or whatever. That's great. It's got that 1800s tint to it.
[51:09] Daniel McShea: One angle for selling what appears to be our view of this is Stan Salty's notion that there's nothing but organisms and machines are just an end member on the spectrum of senescence. They are completely specified. Most machines, he's thinking vacuum cleaners and toasters, are highly specified, extraordinarily brittle and non-adaptable creatures at the end of that spectrum. He's got a developmental theory that predicts that organisms tend toward that state over evolutionary time. They can rescue themselves. There's ways out of this, but there's a strong vector taking them toward machines. We started off creating these incredibly brittle things that are completely specified due to virtually nothing interesting. We don't see them as that end member yet. We don't see that they're just viruses, which are also an end member.
[52:06] Michael Levin: That's fantastic. I'm going to bug you for a reference later. I hadn't heard that piece from him. And I think we're doing a bunch of work on aging and what's actually happening to the bioelectric information during aging.
[52:23] Daniel McShea: I'm afraid that's from in-person conversations, so I can't give you a reference on it. It must be somewhere in his book on developing hierarchical systems.
[52:34] Michael Levin: I'll have to look. I apologize.
[52:40] Daniel McShea: This is fun. We should do it again.
[52:43] Michael Levin: We should absolutely do it again.
[52:45] Daniel McShea: And I'm going to bug you for a SOMS reference.
[52:47] Michael Levin: I'll send you some SOMS stuff and I'll see if he'll come on to the next one.
[52:52] Daniel McShea: Send me a YouTube link.
[52:59] Gunnar Babcock: And I would be curious to explore the much broader conception of memory that you're working with. At least in the philosophy of memory literature, people over there don't seem to be thinking in these terms. And I think it could be an important step forward in how people think about memory generally.
