You are not a meat computer | Adam Frank — Transcript
Full transcript
- 0:00So I'm going to say something really radical here which is that the most
- 0:03important problem in physics for the 21st century is not quantum gravity is not the
- 0:08black hole you know firewall paradox it is life. It is understanding why a cell
- 0:16is different from a rock. I'm Adam Frank. I'm a professor at the University of Rochester.
- 0:22I'm an astrobiologist and an astrophysicist. [music] I'm also the author of a bunch of books
- 0:26like The Little Book of Aliens and right now the newsletter Every Man's Universe.
- 0:33Thank you for watching Bigthink. If you'd like to support our work, we encourage you to join our
- 0:37members community. As a member, you'll receive our quarterly print magazine, [music] a beautifully
- 0:41designed collection of the ideas and interviews that matter most. Bigthink is for curious people
- 0:46who want to take deep dives into big ideas. To support the media you want to see in the world,
- 0:51go to bigthink.com/join. And now back to the interview, the physics of agency. So we often hear
- 1:02from science or we're told that science says that, you know, life is nothing but a machine. It's just
- 1:09nothing but a bunch of biochemical mechanisms. And one of the great problems with the life equals
- 1:14machine idea is that's the idea that's being sold to us and that's the dominant philosophy in the
- 1:19heads of the billionaire tech bros who are shoving you know these AI technologies in whatever state
- 1:25they're in down our throats. So many of the people have that view would if you ask them what a human
- 1:30being is you say oh yeah it's just a it's a robot made out of meat and you know you are nothing
- 1:36but a computer made out of goo. Uh so you know I like to use that term you know uh meat computers
- 1:42that you know these people would say that we're nothing but meat computers which goes back to the
- 1:46beginning of modern science. You know there's the famous story of Decart who would operate
- 1:51on dogs while they were alive and while they were screaming you know the dog screaming in pain you'd
- 1:56like oh don't worry that's not you know it's just a machine that's just the way it responds and that
- 2:02metaphor is fundamentally wrong. Uh life is not a machine. Life is something fundamentally different
- 2:09that we don't really fundamentally understand. Um and the reason for that is right there in
- 2:15the word organism. It's life's self organization that puts it into an entirely different class of
- 2:23physical thing. Like black holes are awesome. Uh, you know, comets are really cool, but they pale
- 2:30in comparison to what life is because life is the only system that is innovative. Life is the only
- 2:38system that creates and understanding how that's possible takes us into entirely new territory
- 2:45scientifically. So sometimes when people are trying to convince us that uh we are nothing but
- 2:50machines they will introduce the idea the famous brain in the vat experiment or thought experiment
- 2:56right and it's the idea very popular in popular culture like the matrix one of my favorite movies
- 3:01is that you know you could kind of like remove the brain from somebody put it in a vat hook it up to
- 3:07electrodes send stimulus into those electrodes and then the brain would consider itself to be living
- 3:12in a full world it's like a simulation you would be able to simulate the uh with just having that
- 3:17brain sort of sitting there. But the philosopher Evan Thompson did an indepth exposition of this
- 3:23and there were a couple things that came out of it that really showed this idea doesn't
- 3:26really make sense. First thing to understand is on some level there's no such thing as the brain,
- 3:31right? The brain is a scientific idea where you take the organism, right? this whole organism that
- 3:37needs all of its important parts to stay living and you isolate different parts like oh this is
- 3:43the circulatory system and this is the endocrine system and this is the nervous system right and
- 3:47that's where the brain then becomes a separate thing what Evan was able to show is that because
- 3:52the brain is part of the organism an in uh uh escapable part of the organism that the brain
- 4:00needs the rest of the body as much as the body needs the brain so even something as important
- 4:06as movement, right? As you move around and you're looking at things, the neural computations that
- 4:11are going on um in the brain require in many ways that there is subtle movements happening in order
- 4:18to flush material out. So this idea that you could, you know, extract the brain and make
- 4:24it different from the body, what Evan was able to show is actually the only way to make that
- 4:29work is you would actually have to build a body for the brain, right? uh you know, some kind of
- 4:35robot body and it would have to be so complex and so subtle in order to keep the brain alive to do
- 4:42all the things that the rest of the organism is doing to keep the brain functioning that in some
- 4:46sense you would just end up with a biological body. So a lot of times when people talk about
- 4:51life they're going to use the thermostat example. They're going to be like well you know you have a
- 4:54thermostat which is a system that has ability to respond to conditions. It senses the conditions
- 4:58and changes its state based on the conditions. But the thing about a thermostat is somebody designed
- 5:04the thermostat with that goal in mind. There was a goal um that someone else had that the thermostat
- 5:10then embodies that other goal. The programmer in some sense. Whereas a dog who's hot gets up and
- 5:18moves not because somebody programmed the dog to do that. The dog feels hot. The dog has their own
- 5:24um urgencies uh has their own experience of being hot and moves for those reasons.
- 5:30So that is the fundamental difference between a thermostat and a self-organized organism. So
- 5:39if you want to go beyond thinking that life is just a machine, you have to think on the level
- 5:44of life's organization and it's how it functions. And the most important thing about living systems,
- 5:51all of them, is that they are all autonomous agents. They are all systems that choose their own
- 5:58uh goals. they choose and maintain their goals. They update their goals as they're moving through
- 6:04the world is sometimes called the inactive view. It's a view of life where things like
- 6:09consciousness is not a property of you know uh of of a living system. It's something it's
- 6:16performed. It's an action that living systems take because what autonomous agents always have
- 6:22to do is they need to make sense of their world. They're constantly working to make sense of their
- 6:28world for their own benefit for them to stay alive or you know uh it could be that that is
- 6:34a more complex uh thing and maybe their status in a social organ uh you know in a social system. So
- 6:41autonomous agents are constantly needing to make sense of the world to realize what's relevant
- 6:48uh about the world in a way that no machine that we have built so far is able to manifest. So this
- 6:54is a problem in cognitive science that uh has been known about for a long time called relevance
- 6:58realization. So let's say uh you go to a party, right? You go to a party and you come inside and
- 7:05you're actually uh uh you want to get a drink, but then you see uh somebody who you've been
- 7:09meaning to talk to for a while and then you start making your way over to them to go talk to them,
- 7:14but on the way you actually see a book on the host's uh coffee table that you've been looking
- 7:18for for a long time. So what's happening here on the fly? you are changing what's important
- 7:25and you're finding not only changing but you're finding what's relevant to you as you move through
- 7:30the room. Okay, that's what we call relevance realization. Now, machines can't do that. And why
- 7:36can't machines do that? Why can't they make sense of the world in this kind of fluid, constantly
- 7:42changing way? And the problem for machines is machines live in what we call small worlds.
- 7:47Every uh uh AI basically has predefined sets of inputs and predefined sets of outputs as well.
- 7:56Machines are always trained on some set of uh uh data what they call training data and that defines
- 8:02its world. Uh even machine learning where it's doing um uh uh you know learning on the job so
- 8:09to speak somebody a programmer had to decide what the rules were or what the rewards were for that
- 8:16system to work. So the choice of the data, the choice of the the machine learning algorithms,
- 8:21those are done from the outside. But a living system like you moving through that room,
- 8:26nobody programmed you. You are finding what's relevant to you on the fly for your own reasons.
- 8:32You are making sense of the world in an ongoing way. That's because you live in a large world. In
- 8:38a large world, there are unknown unknowns. Right? In a small world, the small world that the that
- 8:43all artificial systems that machines live in, there are no unknown unknowns. So living systems
- 8:49have this constant ability to find relevance and adapt to whatever is happening because they
- 8:55live in what we call life worlds. They live in large worlds. They are constantly making sense
- 9:02of the world. They are constantly enacting that making of sense by how they move through their
- 9:06environments and change their environments by moving through them. And so this idea of
- 9:12autonomous agency is really what's at the frontier of this new view of life. And it is really kind of
- 9:17the fundamental mystery of how does something become an autonomous agent and always act on
- 9:23its own behalf. Autonomous agents they always have their own itches to scratch in some sense. When we
- 9:29think about relevance realization, when we think about autonomous agency, we are recognizing that
- 9:36life is the only system that uses information. You could describe a rock, a crystal in terms of
- 9:42information, but the rock really doesn't care. There's other ways to, you know, to describe a
- 9:47rock thermodynamically or such that that you know that that would not uh where information wouldn't
- 9:52matter. life. However, there's really no way to describe life fundamentally without taking into
- 9:57account its use of information. By use, I mean its ability to store, to process, to copy, to transfer
- 10:04information. So, you could say like, well, then it's a computer. And the answer for that is the
- 10:09answer to this is no because computers are based on what's called Shannon information. Claude
- 10:14Shannon in 1948 wrote this brilliant paper where he introduced our modern concept of information.
- 10:21And so that's why really when we talk about information and computers, we're talking about
- 10:24Shannon information. But Shannon very clearly like right in the first paragraph says I'm not
- 10:31interested in meaning. You know when we talk about information we often associate it with meaning,
- 10:35but I'm not interested in that. I'm interested in this engineering problem of I have somebody
- 10:39who's sending some information and somebody who's receiving information and a noisy channel between
- 10:44them. And so he worked out the mathematics of that using a formulation of information that
- 10:49had cleared meaning out. There was no description of meaning. But really in life all our information
- 10:56has meaning. All of the information has relevance. All of the information works towards this
- 11:02selfcreated goal or goals that living systems have either to stay alive or to you know have raise our
- 11:10level in our social hierarchy whatever it might be. So what we was required then is a different
- 11:16kind of understanding of information and that means an understanding of semantic information.
- 11:21For a long time people struggled to try and have a mathematical theory of semantic information
- 11:26that would be as good as useful as Shannon's mathematical theory of uh uh information. And
- 11:32by the way when we talk about Shannon information what we mean really is syntactic information. you
- 11:38have a string of characters and all you're really interested in is how often in a message does you
- 11:44know E the letter E appear as opposed to the letter Z right that's it's just about the syntax
- 11:49semantics adds this new level of meaning but then in 2018 there was a paper by Artem Kchinsky
- 11:55and David Walpert from the Santa Fe Institute it was kind of a manifesto for a theory of semantic
- 12:02information it was a beautiful paper and along with Artem and some other people we formed a group
- 12:08um to sort of pursue the theory of semantic information. And over the last four years,
- 12:13we were able to take that theory one step further. And it we were able to show that you could have a
- 12:19mathematical formulation that associated the information that was being exchanged between
- 12:24the environment and an autonomous agent and the meaning that that in uh that information carried
- 12:30in particular the uh meaning of the relevance for staying vital for staying alive. So this was just
- 12:36a first step but it's an understanding it's a beginning of recognizing that any discussion of
- 12:42of information with living systems needs to have this other dimension this dimension of meaning. So
- 12:48in this new view of life this view that doesn't consider life to be a machine the fundamental
- 12:55thing to be explained or the the the phenomena to uh that we're after is something that uh is
- 13:02called autopoesis selfcreation. So this is an idea that came from Ombberto Montterana and Francisco
- 13:10Varela who were Chilean theoretical biologists and what they were interested in is the fact that
- 13:16there is a special way in which living systems are organized. They're they're selforganized.
- 13:23They're self-creating and they're self-repairing. Of course they're made out of physical stuff. But
- 13:29the new thing that's happening here, the thing that we don't understand is their organization.
- 13:34how that stuff is is sort of circularly organized. Take something like the the membrane, right? The
- 13:41cell membrane is a part of the cell, right? It it allows uh it's what allows some things to
- 13:47go into the cell and some things to go out of the cell. And so it allows the cell to exist,
- 13:52but at the same time, the cell has to be operating to keep the me the membrane working, right? And so
- 13:59this sort of shows that loop of uh of autopoes. The cell is both a process and a product of
- 14:07the processes and products that make the cell function. So what happens is you get this sort
- 14:11of almost infinite regress. You get the processes and products which make the processes and products
- 14:16which make the processes and products etc. It's that that um closed loop what Robert Rosen the
- 14:22great theoretical biologist calls organizational closure. Closure also closure to efficient cause.
- 14:29Somehow you have a network of relationships in the cell between all the parts of the cell that
- 14:35always refer back to each other. So the mystery of something like a cell and this is this is
- 14:39what's amazing about this question of autopoes. It goes all the way down to the cell. We can ask
- 14:44questions about consciousness etc. But really what the real question we're asking about is sentience.
- 14:50Sentience is the true mystery and it goes all the way down to uh to the cell. Cells have this
- 14:57ability to be both open systems, right? There is matter flow and energy flowing through every cell.
- 15:05Uh and in fact your body the atoms that are in your body will if you wait long enough they will
- 15:11leave and new new atoms will replace those u those atoms. So fundamentally you are an open system
- 15:18uh materially but you're a closed system organizationally. And you see this with a
- 15:23cell that you've got this beautiful membrane and there's goo inside and then if you lice the cell,
- 15:28if you pop the membrane, the stuff spreads out and then it's dead, right? That or you've broken
- 15:35that circular that closed organizational system and now it's, you know, the system doesn't work
- 15:41anymore. And in this sense, this autopoetic closed um uh uh organizationally closed uh uh
- 15:49uh system is really almost the definition of life. Because what it shows you, if you really want to
- 15:53understand the difference between a machine and a living system, is I can take a machine and I
- 15:59can turn it off and then I can turn it back on. With a living system, I can turn it off, but I
- 16:05can never turn it back on. Right? Once you break once you break that that organizational closure,
- 16:10once you break that autopoetic structure, it's gone. And that's why this level of organization,
- 16:16this level of understanding life, even a cell as a complex system, an autopoetic complex system,
- 16:22that's the frontier for physics, for biology, for computer science. So people often ask like,
- 16:28well, why can't we just build a machine that does the same thing? And that's a, you know,
- 16:31a good question to ask. But what you have to understand is that we have no idea even how to
- 16:37build a machine like that. One of the particularly interesting things about Robert Rosen and his idea
- 16:43of causal closure or organizational closure in uh biological systems was that he proposed a way
- 16:51of looking at causal closure um that could be explored mathematically using something called
- 16:57category theory. And in the 90s he produced a proof that looked at the structure of causally
- 17:03closed or organizationally closed systems. And what he said was that these systems will not
- 17:07be touring computable meaning they would not be representable as computations or other words you
- 17:15couldn't simulate living systems. You couldn't simulate experiencing systems autopotic systems
- 17:20which is a profound idea right it means that fundamentally you can't replace us with computers.
- 17:25You can't you can't substitute computational systems for the kinds of organizationally
- 17:33closed systems that we are now. That um proof has remained you know very contentious. Some people
- 17:39say it holds up, some people it doesn't hold up. My um colleagues and I are just beginning some
- 17:43work that you know we hope to also uh uh probe that. But it really does raise the question of
- 17:50uh is it possible that living systems are just really fundamentally different from anything that
- 17:55happens in what we think of as a computer? Because it's really about um what's called miraology or
- 18:01the whole part relations. What happens with life is that the parts are assembled by the whole but
- 18:10the parts are part of the whole and they're necessary for the whole. You can't break that
- 18:14circularity. You know with a machine the parts are built somewhere else or there if the machine
- 18:18is building the parts it doesn't need the part while it's building it in order to stay alive.
- 18:24So living systems are sometimes the word is used a Kian whole because it was the philosopher Kant who
- 18:30first recognized that this strange self-organized selfbuilt property of living systems is what made
- 18:38them unique. And so, you know, I would never say that we could never build a machine that would do
- 18:44that. But I can say pretty clearly that we don't understand, not even on a foundational level,
- 18:49how to do that. And the real reason is, again, going back to this idea of small worlds and large
- 18:54worlds. The kinds of machine systems we have now always live in what we call these small worlds.
- 19:00They they have predefined ontologies. They have predefined worlds that they live in set by out by
- 19:07the outside. Um, so even if you're trying to build a computer that can decide for itself, you're the
- 19:13one who decided how it's going to decide for itself. Living systems do this on their own.
- 19:17So in a lot of ways, this critique of machines ever getting that kind of self-organizational
- 19:23uh structure is a lot like thinking about those choose your own adventure books, right? If you
- 19:29remember those, where the book allowed you a bunch of different alternatives for an adventure and as
- 19:34you went along, each choice led you in a different forking path. But the important thing was somebody
- 19:39else had come up with those forking paths. Right? So even when we're trying to build a machine that
- 19:46can make its own decisions, what's happening is we're still imposing our understanding of what it
- 19:51means to make your own decisions. We are sort of setting up the the forking paths. We're setting
- 19:56up the universe it lives in. And the extraordinary thing about machines is they do this on their own
- 20:02and they create universes of meaning in doing it that way. And that is how meaning becomes central.
- 20:09That's why you have to do semantic information. That is how meaning becomes central to the
- 20:13universe. So the important thing in understanding this difference between life and machines is
- 20:20to understand how it um how it affects what's happening right now with artificial intelligence.
- 20:26So our current version of artificial intelligence what are called large language models are designed
- 20:32in some sense to uh mimic the structures that are in human language. So what that means is
- 20:40it's very easy for us to become fooled by them into thinking that there's someone in there when
- 20:46in fact they're just hollow shells. I mean they are brilliant pieces of technology but there is
- 20:51a way in which we can be easily be fooled by them into thinking that we're engaging with some kind
- 20:58of subject that there's someone in there that we're talking to. But once you understand this
- 21:04idea of relevance realization and machines do not know how to realize relevance for themselves. Once
- 21:10you understand the real meaning of agency for a living system and that kind of self-organized
- 21:17um uh uh uh uh autopoetic structure that they have, you then come to understand that these
- 21:23machines that we've built, they're not intelligent really because intelligence is not just solving a
- 21:29problem. It's knowing which problem to solve. And a um AI would never choose its own problem
- 21:36to solve in one way or the other. It is given instructions and it follows scripts because it
- 21:42has read everything that has ever it's been fed everything that's ever been uh written in some
- 21:47sense. It's just following along paths that the algorithm finds in every story that it's ever ever
- 21:54written. So once you have that understanding, you can come to these machines and recognize
- 22:00they're not artificial intelligences, they're artificial mimics. Um they are mimics of us.
- 22:06They are powerful tools that we use to do things. They don't do anything on their own that we use
- 22:12them to do it. For example, when Deep Mind beat the world's best Go player, Alvin Noey said, "That
- 22:18computer didn't beat the Go Champion. We used it to beat the Go Champion." These machines are not
- 22:25like us. They do not have the richness of, you know, of even a, you know, a fox, right? They are
- 22:32actually very they pale in comparison to what the richness of our conscious experience our our um
- 22:41our being as experiencers. And so what will happen is is that as these technologies get deployed
- 22:49without any kind of oversight in particular what we'll see is a leveling of humanity the attempt
- 22:55to level humanity. So I uh one time sat in a a lecture where a CEO of a company that was doing a
- 23:03um uh emotional computing was arguing that oh you know we can make computers that the camera will
- 23:11notice the you know position of the muscles in your face which correspond to emotions and then it
- 23:17can change what it's giving you based on the what it what it thinks your emotional state is. Um,
- 23:24and so then we could have robots that could take care of old people. And I was horrified by this
- 23:30idea because one, you can ask like, well, what kind of society do we have where we're just going
- 23:34to leave old we're going to warehouse old people and leave them with robots? But even more, uh,
- 23:39was this idea that, you know, my the the N7 muscle or whatever you would call it in my cheekbone,
- 23:46its position tells whether I'm actually happy or sad. And in fact, that research has been shown to
- 23:51be completely false. Um, so what happens, this is a nice example of how we have these systems which
- 23:56are not actually what we're told they are. They're not as good as they're what they're told they are.
- 24:01And they will, we will be forced to step down to their level. It's as if to build self-driving
- 24:06cars, we leveled Paris with all of its crazy street structure and we built something that only
- 24:11had right-hand turns, right? So the real danger is not that, you know, Skynet is suddenly going
- 24:16to wake up and attempt to eliminate humanity. The danger is that we will be forced to you know drop
- 24:23down to these in the level of these incomplete systems which don't have the richness that we
- 24:28have but because of the economic structures around AI they will be deployed and we will be forced to
- 24:34use them and we will lose some of the best parts of being human beings and being alive. And so I'd
- 24:39like to offer three pieces of advice for when you sit down with an artificial mimic. The first is
- 24:46is tell it to stop referring to itself in first person pronoun. Claude will tell me like, "Oh,
- 24:51would you like me to make a PDF of this for you?" And I say, "Stop referring to yourself in first
- 24:56person." Refer to yourself as the system. And it makes a huge difference. You know, now it's
- 25:01the system can prepare a, you know, a PDF of this document for you. The second thing is you have to
- 25:08tell it to stop praising you, right? the companies who want you to keep using these machines have
- 25:14built in um that sort of modality that it's always telling me like, "Oh, Adam, what a
- 25:19great suggestion. What a great idea you had, Adam. What a nice shoes you're wearing today, Adam." And
- 25:25um you can call it nonaffiliative mode um or non-affirmative mode. And I just tell the machine
- 25:32to stop trying to please me. just I'm going to give you a uh I'm going to give the system uh uh
- 25:37a prompt and I just want the information back from the prompt. The last thing is never say please
- 25:43and never say thank you. It doesn't help the computations that are going on in answering your
- 25:48query and you are literally just purposely burning up um energy needed to do those calculations. So
- 25:55those three tricks can help remind you that this is not an autonomous agent. This is not a self.
- 26:02This is a hollow shell, a tool, an incredible tool that you are using for your own purposes.
- 26:18Four times a year, we print a magazine worth putting your phone down for. Big Think's premium
- 26:23print magazine is built with custom artwork and expert insight from the world's biggest thinkers.
- 26:28No AI slap, no rage bait, just big ideas for curious people. Support
- 26:33the media you want to see in the world. Go to bigthink.com/join to become a member today.
About this transcript
This page contains the full transcript of You are not a meat computer | Adam Frank by Big Think Clips, generated from the public captions YouTube serves with the video. The transcript has 4,691 words across 270 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.