Corey: All right. Okay. Welcome to our podcast. My name’s Corey Washington, and this is my co-host Steve Hsu. And we’re going to tell you little bit about how we hope our next episode should be going and some of the plans we’ve got for the show, but we’d like to lay out our general philosophy to start.
Steve: Now, one of the things we really want to do in this show is always present the opposing point of view, but in a respectful and informed way. We always feel that a stronger sense … A stronger understanding of a particular subject emerges when you hear both sides of it. And so …
Corey: Or many sides, as the case may be.
Steve: Or many sides, yeah, if there may be more than just two. And so there may be times when we have a guest on who’s controversial and neither Corey nor I agree with what that person’s thesis is, but we’ll listen respectfully, and then we’ll try to raise what we think are the most salient objections. Similarly, we might have a guest who actually we don’t have any objection to what they’re saying, we’re just fascinated by them and want to hear what they have to say. But nevertheless we will try, as a duty to the audience, to still raise some points against their perspective, so that everybody emerges with a broader sense of what’s going on.
Corey: I think that’s going to be true for a lot of the science shows that come up. First of all, we’ll be talking to people who … We’re not experts that area, and so they’ll have strong positions about their view. And also, we just may disagree with their science, but science works best when you have … Lay a critical eye at it and begin to look at its assumptions, and some of the claims are being made in a critical way. So we really want to have that, a good natured, but …
Corey: Sharp, yeah.
Corey: And adversarial discussion sometimes.
Steve: And we won’t just be talking about science. We plan to have people on who are philosophers, maybe who want to talk about politics, who want to talk about UFOs.
Corey: UFOs, that’s right.
Steve: Whatever it is, but …
Steve: God. Yes. So we’re very excited about our first guest. This will be the first interview that we’ve done.
Steve: Our guest today is Bobby Narayanan, a neuroscientist from Argonne National Labs and the University of Chicago. I’ve known Bobby for several years, and we’re friendly, at least up until this interview. And so I’m going to be the main interlocutor. Corey is welcome to interject, hurl invective at us, disagree with anything that we say, but I’ll try to lead the interview with some specific topics that we want to cover.
Steve: And let’s start with your biography and your career. I’m going to do my best at just saying a few things about it, and would love if you would elaborate on some of the things that I’m about to say about you.
Steve: So Bobby, you were born in the United States, is that right?
Bobby: No, in India. [Coonoor 00:03:11], India.
Steve: Oh, and when did you come to the U.S.?
Bobby: When I was six years old, or five years old, in 1980.
Steve: And where did you grow up?
Bobby: I grew up in a small … Most of my time, I grew up in a small village called Coonoor, India, in the southeast of India. And then spent a lot of my time in Madras, India, which is called Chennai, India now, which is where my grandparents lived.
Steve: But how about in the U.S.?
Bobby: In the U.S., I grew up in the great state of New Jersey.
Steve: New Jersey.
Bobby: I am more from New Jersey than I am from India.
Steve: Okay. So you grew up in New Jersey, and somehow found your way, probably not far from where you grew up, to Princeton University.
Steve: Where you did your undergraduate degree.
Steve: And what was your major?
Bobby: I had an odd major. This will come up as we talk about my career, because I’m on technically my fifth or sixth career currently. When I was in college, I was really interested in the space of scientific public policy. So I majored in molecular biology, and had a second major in something called the Woodrow Wilson School for Public and International Policy, which is at Princeton.
Steve: Very good. And so subsequent to Princeton, you were a, I believe, a Road Scholar, and you did your PhD at Oxford. Am I right about that?
Bobby: Correct. Correct.
Steve: And was it at that point that you became a neuroscientist?
Bobby: So there were a few jobs, potential jobs in between. In between public policy and where I am now. I really thought I would make a lot of money in healthcare. So I started medical school to try to just get my understanding of medicine so I could make a lot of money in healthcare. It was pretty obvious after a year that that’s not for me. So I decided oh, well maybe basic research is the right way to go. So I had already done a year of medicine, and just randomly applied for this scholarship, that then I got. And then because I got it, I decided to become a scientist.
Steve: Great. And I think your MD is from Washington University in St. Louis, is that right?
Bobby: Correct. Correct.
Steve: Great. And so then subsequent to your PhD, you followed the typical academic research track. You were a postdoctoral researcher at … Let me see if I have this right. Boston University at Harvard?
Bobby: I was at Harvard doing my postdoc, and then I was briefly a professor at BU.
Steve: Got it. Okay. And …
Corey: In what department, Bobby?
Bobby: At BU … Sorry, Harvard, almost all the departments are variants of anatomy and neuroscience, neurobiology, et cetera. I think at Harvard, it’s called Neurobiology, and at BU it’s called Anatomy and Neurobiology.
Steve: Great. And it was there that you really got into brain mapping. Is that right?
Bobby: Mm-hmm (affirmative)
Steve: And I think you have a patent on some kind of device that does fine slicing of brains, and then floats the slice on water so that it can be, I guess, interrogated by … Is it X-Rays?
Steve: Electrons. Electron microscopy.
Corey: It’s like a tiny conveyor belt, is that right?
Bobby: It is literally a tiny conveyor belt. It is this as Rube Goldberg-esque as you can imagine an invention. And we could discuss this, but making a tiny conveyor belt and adding it to the normal workflow of electron microscopy is the reason why I have an academic job right now [crosstalk 00:06:32].
Corey: So how big, Bobby, is this conveyor belt?
Bobby: My conveyor belt is about … In fact, we modeled it on the very old audio tapes, so it’s about eight to 10 millimeter in length or so. And the reason we did it was that one of the things … And Steve, do you mind if I push past or should I wait?
Steve: No, no. Well, I think this is sort of along the lines of where we’re heading anyway. Just quickly … Maybe just quickly say, so it’s eight millimeters, and then the thickness of the slice is roughly …
Bobby: Something on the order of … Each brain slice is on the order of 40 to 50 nanometers in thickness.
Steve: So pretty thin.
Bobby: Or another way to say that it 100 to 500 atoms, if you imagine that an angstrom is about the size of an atom, and 10 of those is a nanometer, you get to something like that. And the reason we have to cut these things so thin is that in a human brain, there’s something like 100 billion neurons. And each of them make 10 thousand connections with each other. So that means the number of connections between neurons in a human brain is on order quadrillion, or hundreds of trillions, if you multiply those two numbers together. The only way you can fit that number of anything into the volume of a cranium, if you will, into the volume of a human skull, is to make each of those elements extremely small. Smaller actually than the wavelength of physical light. So a neural connection is actually smaller than a wavelength of visible light, so you can’t use standard optical microscopes to map a brain. You have to use electron microscopy.
Bobby: The downside of electron microscopy is that the electrons don’t penetrate very deep into your sample. They scatter really fast. So the only way you could map a brain is to cut it into a whole bunch of thin slices, each 50 nanometers or so. And you have to be able to collect something like 50 thousand slices, each one 40 nanometers in a row. And that seems physically impossible until a collaborator and I at Harvard were like, “Well, maybe a tiny conveyor belt solves the problem.”
Steve: Great. And so the … In terms of the actual devices, is it true Zeiss sells a device based on your work around the world?
Bobby: Yes. Correct.
Steve: And so how many of these … How many labs have these around the world?
Bobby: So, I would say there are about 15 or 16 labs in the U.S. that have this device. And the reason I think more people don’t have this device is that the device only solves half the problem, which is that once you collect a [inaudible 00:09:20], even something a millimeter cubed … So just to give you an idea, that’s smaller than a grain of sand, a grain of dust, if you will. If I did a millimeter cubed of a mouse brain at EM resolution, it would be something like a million gigabytes of data.
Bobby: And so the conveyor belt gives you the ability to capture that data, to collect that data, but we’re still struggling with what to do with the data, how to analyze it.
Steve: Right. The situation you’re describing is pretty common across multiple sciences, where the instruments that are relatively inexpensive now can be used to collect far more data than we really actually know what to do with.
Bobby: Correct. And in a lot of fields, like Steve [inaudible 00:10:02], apparently [CERN 00:10:04] was a pretty good instrument. [crosstalk 00:10:07]
Steve: But an expensive one.
Bobby: I’m sorry?
Steve: But an expensive one.
Bobby: Yes, but hopefully worth it. And I had heard that at CERN, they collect similar numbers. They collect a thousand terabytes a day, is not unheard of for CERN.
Steve: Yeah. It’s even …
Bobby: But what they wind up doing is throwing away a lot of that data as they’re collecting it, because if …
Steve: Yeah. It’s even worse than that, because at the hardware level, they design [VITOs 00:10:32] that choose not to record big chunks … Most of the data, actually. It’s only piping … The number you gave was the data that’s piped out, and that’s after passing through several hardware level filters.
Steve: So it’s generating far more data than that, actually. But yeah. So …
Bobby: Yeah, I know. And it’s because physicists are fundamentally, I don’t know, smarter. We can … Remind me to tell you …
Steve: [crosstalk 00:10:53] say that.
Bobby: … A story about physics and neuroscience. They know exactly what to look for, to understand in their particular system. Something that’s interesting about neuroscience is we’re not really sure what the right metric we want to measure to understand a map of a brain. So we’re forced to collect all the data now, and see which parts of it provide understanding. And maybe the future brains that we collect, we’ll be able to adopt a CERN philosophy. We’ll be able to throw away a bunch of this data.
Steve: So in my notes for this section of our discussion, I want … What I had written was describe the research for a general audience, but then also for an expert. And I feel like we’ve done that a little bit, but let me turn to my ombudsman, Corey, and say have we described what he’s doing in a general enough way and also then what would an expert want to know more about that he does?
Corey: So I don’t want to rain the parade a little bit, but I think it’s actually not clear to the audience what Bobby’s conveyor belt does or how it fits into his …
Steve: So get him to elaborate on that.
Corey: So Bobby, we don’t have facials for this talk, right, but I think … Can you explain basically how a slicer’s going to work and how they’re going to cut thin sections? And then what happens to them?
Steve: It’s not like ham?
Corey: It’s a lot like ham actually, exactly, right?
Bobby: It is. It is.
Corey: But then you …
Steve: [crosstalk 00:12:14]
Corey: So give people a visual image of the size of the brains you’re cutting, the size of slices that are coming off, and the way your conveyor belt sits in that process.
Bobby: Awesome. So this is … Let me start off with what the constraints are, and then I’ll tell you how the solution matches those constraints. So the very first thing I said was that we have to use electron microscopy, because the density of all the things, the size of all those things in the brain requires it. What that means is we have to cut the brain into a whole bunch of little thin sections in a row, like cutting slices of a salami or a sausage or et cetera. And one thing to say is that it’s very hard to cut tissue in an exacting way, because it’s wet, it’s weird, it squirts around, et cetera. So scientists, particularly neuroscientists have developed ways to prepare a brain, usually a dead brain, a fixed brain. We stain it with a bunch of metals that the electron microscope can see. And then we embed it in plastic. We literally turn the whole brain into a piece of plastic.
Bobby: And if you could imagine that piece of plastic attached to a lever arm. Can you guys imagine that in your mind? The lever arm is going up and down, and it’s advancing in X nanometer steps, right? Right. And that’s what we call the sample. Right next to the sample, Steve, if you could draw … Make a little fist for the sample. Yeah. And right next to it is a knife. And that knife is made of diamond. Literally, it’s a piece of diamond sharpened down. As that block advances, sections get cut. Those sections then float on water. And then at the end of that water trough is that conveyor belt.
Corey: So …
Bobby: So this section …
Corey: Bobby, let’s stop, because this is the problem you’ve solved. And many people who’ve done slices in their lifetime have had to deal with this. Previously, you’d basically have a little slice come off and it falls into some sort of solution. You pick it up by hand.
Bobby: By hand.
Corey: With a tweezer.
Corey: Put it on a slide, right, and you do this repeatedly. Half the time, you mangle the slice because you twist it a little bit.
Corey: And so you wanted to solve this problem of trying to get thousands of slices all done perfectly.
Corey: Without massive carpal tunnel syndrome.
Bobby: Exactly. And in fact, the human record for the most number of slices ever collected is about 1000 or so. And even worse, the people who do it are artisans. The people who have the patience and the motor skills to pick up section after section after section. That’s actually a dying form. It’s a lost art form. People aren’t trained in it anymore. So the idea that we want … We wanted to produce something that was robust, that we could distribute to labs all over the world, and that could scale. So it turned out the same way John Henry loses to the steam machine. I suspect these artisans are winding up losing to a conveyor belt. And just like John Henry, they’re super upset.
Corey: And so I think that gives people a picture of …
Steve: Very good.
Corey: Your invention.
Steve: Yeah, I should say that we’ll probably put in the show notes links to a technical seminar that you’ve given. I think you gave one [inaudible 00:15:48] at MSU a year ago which has lots of slides and visuals. And then even … I think I’ve seen on the web little videos of your machine, or an animation of your machine in action. And that will make it clearer to people who still are a little bit confused about what it does.
Corey: How much money have you made off your machine?
Bobby: Currently, because Harvard essentially owns the patent, I have made zero dollars.
Corey: Oh, that’s sad.
Steve: Okay, but Zeiss has probably sold, I don’t know, tens of millions of dollars worth of these machines, right? Or no?
Bobby: Someday I hope to make dozens of dollars, but currently … This is neither here nor there, the first set of instruments they sold were prototype … Et cetera, et cetera, et cetera. So they felt like they shouldn’t share the money for those first X set of instruments, because they needed to use it to keep revamping the system.
Steve: So one of the topics further down my list is scientists are underpaid, and society under-invests in scientific research. We won’t branch off into that just yet.
Steve: So let’s stay on the research end. So Corey, you’re an expert in neuroscience. What’s a thing about what Bobby does that would be … That an expert might like to hear more about?
Corey: So I guess I’d be very interested to find out how the techniques that you’re using right now, Bobby, are advances beyond what went before. I remember you talking about how physicists throw away a lot of data.
Corey: What’s pretty interesting is that early neuroscientists did this almost by accident, but it was very, very beneficial. So as you know the early stains, right, done by Cajal, using the Golgi method, only stained about one percent of the cells, maybe less. We actually have no idea precisely. But as a result, you could actually see individual neurons, and Cajal was able to see, effectively, synapses, right? So he … Because the stain was so inefficient, it was sparse enough to get a nice picture of it, which I thought was really fascinating, because that’s really why it worked.
Corey: But since then, there’s been a whole train of research of ways of mapping the brain. And you’re doing work right now which is using really cutting edge techniques. So how would you describe your work as an advance of what went before? What was there before you came in, and what are you doing that’s new?
Bobby: Yes. Yeah, so I think that’s a great point, Corey. I think the advantage and the thing wrong with Cajal … By the way, the guy that Corey mentions is a guy named Santiago Ramón y Cajal, who won the Nobel Prize probably in the early 20th century, for really changing one fundamental view of the brain, which is that it’s made of cells. The brain is made of cells called neurons. Neurons have specialized parts to them, dendrites and axons. And that they connect with each other over empty spaced called synapses, or connections.
Bobby: What’s amazing is he didn’t actually see a single connection, because he had this sparse labeling, and it … Most of the times, you don’t know who that cell is connecting to, because 99.9 percent of the cells are not labeled, if you will, in the data set. And what most people do is then they go from animal to animal to animal. So if only one of 100 cells was labeled, you might imagine, well, if I do it over 100 animals, or some 500 animals, you should be able to recapitulate. And the labeling is random, et cetera. You should be able to synthesize potentially all of that back together, maybe, into one brain and not go through all the hassle of the cutting and the millions of terabytes of data that I’m claiming that we have to do.
Bobby: I think the main issue with that is that fundamentally, brains, especially mammalian brains, are not identical from one brain to the next brain, to the next brain. If I had the same identical brain … Sometimes people think invertebrate animals, like flies, have the exact same brain again, and again, and again, and again. And therefore, it’s possible to sparsely sample the same identical network multiple times and make an inference about how that network happens. I think in more complicated brains like mammalian brains, who you connect with is actually dependent on who else they connect with. The system itself, it’s not identical. That system is designed to … For the history of that brain. And that history of that brain makes sense when you look at the network, not when you look at individual neurons over many brains that have different histories, essentially different connection matrices, and put it all back together.
Steve: Yeah. So …
Bobby: Another way I would say it, if you don’t mind …
Steve: Go ahead.
Bobby: Is that implicit in the word circuit, a neural circuit, which is the collection of neurons connected together to make a behavior, is this idea of a circle. It’s hard to say the word circuit without imagining the word circle. So at some point, you would imagine that … And it should be true, that you could go from a neuron to another neuron, to another neuron, to another neuron, and back to yourself. And in reality, no one has ever visualized a single circuit in neuroscience that is responsible for a behavior. We just do proxies of it, we claim its circuitry, but if you ask somebody … The one time perhaps it’s ever been done is a very small animal called [C elegans 00:21:18], and have 302 neurons. And someone mapped all of its connections by hand, manually, like you were saying. These artisans who pull C elegans section after C elegans sections.
Bobby: And when you look at that wiring diagram, how that circuitry connects with each other, it’s way more complicated and has motifs in it, circuit motifs, repeating things that would be very hard to infer just by looking from single neurons from animal to animal to animal. So, the last way to say this, and Steve knows I think this way … I’m always surprised by why we haven’t achieved more in neuroscience. The annual Society for Neuroscience is, I don’t know, 55 thousand people a year. It’s held once a year. I think … Or 60 thousand. It’s the single largest convention for science. I think the cardiologists used to have more than us, but we beat them recently. The NIH budget for neuroscience is something like a couple of billion dollars a year. And only about half the people go to neuroscience. So if you have 100 thousand smart, dedicated human beings spending billions of the government’s money a year, and you multiply that over 15, 20 years, you might ask, well, what the hell have you guys discovered?
Steve: Well, but a very hard problem, right? And I guess it’s cliché …
Bobby: Possibly. Possibly, it’s a very hard problem. Or possibly …
Steve: But the brain is the most complicated thing we know of in the universe, right?
Bobby: Yeah, I mean, I wonder whether neuroscientists say that as job security, just in my opinion. But let me push this a tiny bit more, Steve, if you don’t mind. Another version is … Of course, it’s a complicated system, but you know the immune system is pretty complicated. The heart is pretty complicated. The kidneys are pretty complicated. We have gone so far that we can make artificial versions of almost all of those things that I’m describing to you. We’re not even close to that, to a brain. One version, my version is that because neuroscience is theory rich and data poor. Most neuroscientists, whether they work in lab or they work with math, are actually theoretical neuroscientists. They’re working on very sophisticated hypotheses, very intimate theories of the brain. How individual molecules can individual memory and behavior is the standard routine for neuroscience.
Bobby: Given that we don’t even know what 99 percent of the cells do based on this sparse staining, it just seems like the cart before the horse sort of thing. I don’t know how to do this analogy, but you guys are better at it than I am.
Steve: So I think you’ve let precisely into the next point that I wanted to get to, which is a statement that I wanted you to react to. So let me read the statement, and then we can discuss a little bit more about how this statement’s related to the point you were just making. The statement is tool-builders, perhaps secretly, are the real drivers of scientific progress. Broad ideas by themselves can be overrated. So actually, I’d love to hear both of you react to that, but why don’t you go first, Bobby?
Bobby: Okay. So I’ll react. It’s … I think this is pretty telling, and I do think you did this on purpose, Steve. I can’t agree with a statement more than the statement that you just made. When I give my talks, I do this thing which we can put on the podcast if you want, which is I make a list of the Nobel Prizes over the last 100 years that have been received for a specific idea. And the vast majority of them, I’d say 100 percent of them, but that’s a little … Probably too much, are because that person had access to a tool or a technology that nobody else did. And in fact, to go back to my hero, Cajal, who we were discussing earlier, that guy had access to two things that the vast majority of neuroscientists didn’t have access to. He had access to the stain that Corey was telling us about, that only labels a very small percentage of all the neurons. He didn’t invent that. His frenemy, Camillo Golgi invented that. And he was also one of the first people to use a compound microscope with glass diffraction optics, to look at brain slices.
Bobby: Now, there are things that Cajal discovered using those tools, but I don’t think I could … I’m not that smart, right? But I would say 70 or 80 percent of the things that he discovered, I could amassed if I had access to technology that nobody else had.
Corey: I guess my reaction to the statement is I think it’s really a false dichotomy between technology on the one hand and ideas in the other. I think, is my reading of neuroscience, history of neuroscience, is that whenever you have these kind of significant advances, because you have an idea what [inaudible 00:26:05] technology is. Someone had a theory, and they had a way of testing the theory, right, which may have been novel. But sometimes they had a theory, and not super novel technology, but I don’t think if you have them separated … If you have ideas without technology, you’re pure philosophy. If you got technology with no particular ideas, that doesn’t lead anywhere either.
Corey: So I think this is actually one of my problems with a lot of neuroscience. I think it’s a lot of technology driven research without a particular theory behind it. When I was in grad school, the two photon microscope was becoming very popular, and now they’re fairly common. But our lab was one of the few that had them. And people were doing a lot of experiments with a two photon, but it wasn’t clear what these experiments were supposed to show. You could get some pretty pictures, and you could see some synapses … I’m not sure they’d been seen before, but you saw [inaudible 00:27:00] in a more brilliant way than you had before. But there was no outcome to that as far as I could tell. So that struck me as a case where technology was not whetted to ideas.
Corey: And I think the key thing to be successful … I think we agree about this, Bobby, you have to have both ideas. I think many great scientists are actually philosophers at the same time. They’ve got a way of making these ideas, their theories practical and testable.
Bobby: Yes. Yes.
Corey: I mean, Cajal I think is quite interesting. I remember looking back at Cajal and I was actually in a reading group in early part of grad school. And one person pointed out about Cajal is not only do you have the Golgi stain, he was actually probably the best practitioner of the Golgi stain. He was better than Golgi at the Golgi Stain.
Corey: And so his stained racks … So good, you could almost see the synapses. You couldn’t see them in Golgi’s, right? And they’re a little bit hard to see, but you could see them in Cajal’s and not in Golgi’s when you looked. You probably see why he got them and Golgi didn’t. Golgi’s were just a bit too fuzzy around the connection, but Cajal’s were picture perfect. And so I think it really requires a lot of technical expertise to do what he did, and that’s what I think led to a lot of his discoveries. So …
Bobby: [inaudible 00:28:09] Steve, do you mind if I push back one more time?
Steve: Yeah, you jump … You jump. Go ahead. Go ahead, Bobby.
Bobby: On this? Because I’d like to tie the philosophical question back into the technological question.
Steve: Go ahead, Bobby.
Bobby: The reason I think technology is the answer is because of philosophy of science. So let me push back, or not push back. Let me try. Let’s say there’s two main ways to imagine how to do science. The first is this way … Actually both ways were published quite close to each other in the 1950s. The first way is if you think of Popper, and how Popper … Karl Popper, one of the great philosophers of science, said well, the way science advances is by falsification. I have a theory, I try to falsify that theory with data. And of course, technology can help with that falsification, but I really need that theory first before I can invent the technology to help falsify it.
Bobby: I think the alternative view of the philosophy of science is this kind of Thomas Kuhn view of science. And Thomas Kuhn, who I think wrote shortly after Popper did, wrote this book called the Structure of Scientific Revolutions. And in this book, Kuhn makes this argument. He’s actually the first person, I think, to make these words paradigm shift. He comes up with this idea of a paradigm shift. And his view of science, how science progresses is a lot different than how Popper views it. In the Kuhn-ian view, the existing scientific world, the sociology of it, does everything it can to prevent the destruction of its worldview of science. There’s an established field.
Bobby: The two examples, the interesting example is this Copernican versus Ptolemy view of how the planets move around the world. So they move around the Earth or they move around the sun. So there was a worldview for a long time that the planets move around the Earth. And lots of people collected data consistent with that. Of course, it was rudimentary data at the time. And then what happens is somebody collects a piece of data that is not consistent with this … Or throws some dirt into this worldview, which is it turns out that for all the planets to move around the Earth, some of the planets have to go backwards, or do weird loops.
Corey: The cycles.
Bobby: Et cetera, right? And that little dirt in the oyster is what becomes a pearl of a new scientific theory. Another version of that is black-body radiation. I imagine Steve knows this better than anybody. There’s these weird little facts that don’t make sense with the worldview, that are surprises. And then you knock down that worldview and build up another. That’s not a gradual view of science.
Corey: [inaudible 00:30:54] speed of light.
Bobby: [crosstalk 00:30:55] If you’re interested in that second worldview, I would argue that the way to do it at is to invent technology that reveals a surprise, and that the real advances are these surprises in science that knock down views. So even though the two photon microscope invented by Winfried Denk, who then went on to invent a lot of these [inaudible 00:31:16] tools that I was telling you about. So that’s interesting. The two photon microscope, I think hasn’t been proven yet because it yet hasn’t discovered the surprise. It’s really good at validating or invalidating hypotheses, but what we want are data sets where the surprise lives.
Bobby: And my argument, maybe to go back to earlier, Corey … The reason I want to collect all this data in a way that nobody has seen it before is somewhere in there is the surprise. Now, that’s not good for granting agencies and others. I have to come up with reasons for it, but the reality is I think that’s where technology helps. It is this thing that produces surprises that often winds up being the truth.
Bobby: That’s my last piece. Sorry.
Corey: Yeah, I don’t want to get too deeply into …
Bobby: Yeah. We can do [crosstalk 00:31:59] on this.
Corey: Philosopher science. Yeah. Yeah, I think the Popper-ian view is … Some parts of science work that way. Neuroscience not so much.
Corey: I think Popper’s … I think, mostly thinking about physics, actually, and chemistry to some extent. Kuhn’s right, there’s certain anomalous phenomena do arise that lead you to get out, change things. Kuhn has a very strange kind of … Almost anti-realistic view where he thought these paradigms were almost self-sealed, right? And you couldn’t really talk about falsification at all, you just move from one paradigm to another. It’s not that one was more true than the other. It’s just a different idea system. It’s almost … Kuhn’s almost post-modern, I mean, in that sense.
Steve: Or maybe the people have adapted him to post-modernism.
Corey: It’s possible. Even in him, right, he says things that are pretty … Like that.
Steve: A little bit, yes. Yes.
Corey: But I think both have something to say. But neuroscience, I think if you look at some of the developments, and pretty interesting, right … If you look at … For instance, when people discovered the currents. When Hodgkin and Huxley did their voltage clamp experiments, they didn’t have a new technique, but they also had a certain theory about you want to be able to control for voltage level concurrents. And out of that, began … Out of that, they made a mathematical model that although did not posit channels as physical entities, effectively had sodium and potassium channels in it.
Corey: So they pretty rapidly developed this theory based upon certain assumptions, which drove their technology use that then led to a huge range of other experiments, right? So the idea … A general idea first, a technology, much more specific idea after that, and then …
Corey: Since then modern neuroscience took off. But yeah. I don’t want to hijack Steve’s narrative here.
Steve: No, that’s okay. I don’t have much to object to in anything that either of you said. I would just … And I think most scientists would agree that technologies are important and ideas are important. I mean, the latter maybe just because the space of possible hypotheses is so big that you couldn’t just randomly generate hypotheses and test them. You would never finish. So you need some sense of taste or some way of narrowing the set of ideas that are worth testing, that are worth spending all this time developing technologies to test. So I don’t think there’s too much super controversial in that sense.
Bobby: Steve, can I try to make it controversial?
Steve: Okay. Well, if you insist. Go ahead.
Bobby: Okay. I always insist. I’m assuming that’s why invited me to a podcast called [crosstalk 00:34:20].
Steve: Yes. Go ahead. Go ahead.
Bobby: Here’s the last piece. There is no person, very few people in neuroscience practice either Popper-ian or Kuhn-ian science. By that, I mean I have never met a neuroscientist, especially the more famous they get, who is interested in disproving their theory of the brain. Most … And Corey, I can see you. You can’t see me. So, you don’t see me smiling. We’ve gone through theory after theory, long-term potentiation is the basis of memory. Spike-timing-dependent plasticity is the basis of memory. Retrograde nitric oxide, so crazy to think that was the molecule of the year, is the basis of memory.
Bobby: All these scientists who come up with a theory don’t want that theory to die. They want their theory to actually be correct. And this is …
Corey: And isn’t this true … Isn’t this true, scientists and human beings in general, Bobby? Nobody wants a theory of anything falsified.
Steve: Yes. I think it’s a matter of degree, though.
Corey: Yeah. [inaudible 00:35:17]
Steve: I think the best, say, physicists wouldn’t defend … Irrationally defend their pet idea.
Steve: But I think the incentives are what Corey says. I mean, almost everybody does actually defend their idea a little bit too long.
Steve: Maybe all the way to the grave.
Steve: That’s the worst case scenario, but certainly a little bit more than an uninterested observer who hadn’t written the key papers for that particular hypothesis would defend the work. So I agree with that. I don’t disagree with that.
Bobby: Well, then maybe I’ll try something controversial later.
Steve: Okay. Let me go on to another statement, which I want you both to react to, which maybe it is controversial, maybe it’s not. But it’s my view of what’s happening right now a little bit. So let me start with AI. So in AI, we’ve made a lot of progress in using things called deep neural nets. And deep neural nets really are, in a way, a crude cartoon version of how the human brain might operate in the sense that there are layers, and there are connections between the layers, and there are things you could even call neurons or synapses. But in any case, individual nodes of the neural net affect other nodes, and there’s a certain strength by which they affect other nodes. So there’s a connectivity, and then there’s a strength of connections within that network of connectivity.
Steve: And we have examples now of really successful deep learning, really successful neural nets. So for example, recently a company called DeepMind trained a neural net called AlphaGo, which outplays the best human players in Go. And Go is a game which has been played seriously by humans for thousands of years. And most computer scientists thought it was going to be a long time before computers could beat humans at Go, but now they’ve just … In a very short time, totally eclipsed humans at the game of Go.
Steve: Now, unlike your problem, the trouble … The huge effort you have to go to to image the inside of, say, a mouse brain or something like that, these scientists or engineers at DeepMind can just open up the neural net, which plays Go. They can open up AlphaGo, and they know everything about it. They know its connectivity. They know how the connections work. They even wrote the way in which information is processed by that network. So there’s nothing unknown to them about AlphaGo.
Steve: However …
Corey: Well, let me … Then let me push back on that, Steve, because isn’t it … I mean, this was … Maybe its connection were done by simply cyclical learning of previous games of Go. So the many, many connections in this neural network, and it’s not clear they actually know how it functions. They know that it gets the right answer, but in fact, these scientists actually do ablation experiments on neural networks to see how they function.
Steve: [inaudible 00:38:03] So this is the point I’m trying to make, actually.
Steve: So from a descriptive viewpoint, there is nothing further to be learned. I mean, it is represented fully in silico.
Corey: So that the anatomy …
Steve: That the actual neural … The anatomy.
Corey: The anatomy. Okay
Steve: The anatomy is fully known of this neural network.
Steve: So it’s as if Bobby’s program had been carried through to completion.
Steve: So you had a complete one angstrom level resolution map of the mouse brain.
Steve: Okay. And even more than that, you had really accurate dynamical models of how a charge buildup here would do something here. So they know everything. Given the state of AlphaGo at Time T, they can tell you perfectly what the state will be at Time T plus one. Okay? No question. Now, where in that huge body of information, its millions of connections, strengths, millions of numbers, right, that specify the neural net … Where in that is an understanding of the game Go?
Steve: So you can have an example where … And so this is related to the statement that our brain might be too complex to be understood by our brain. So what I mean by that is even if you gave me a perfect anatomical description and even dynamical description of how our brain processes information or shifts electrons around, we might never actually locate in the brain the exact place where it knows that if it rains at night, the lawn is wet. Right? That’s a thing that we think most humans can figure out, that little syllogism. But we may not figure out mechanically how it’s represented by all the complexity of information we get from the anatomical description of the brain.
Steve: And I think that’s clear now, from what’s happening in deep learning and deep neural nets, that there is no way we can really open up AlphaGo and figure out the game of Go, at least not easily. Now, Corey actually already leaped ahead to ablation experiments, where you say … Well, let’s suppose I blow away or randomize some sub-chunk of connections. What does it do to gameplay? Does the thing start making bad moves, but only bad moves in the opening? Or only bad moves in this subset of situations? So you can probe it, but that in and of itself is a maybe, perhaps 100 years project to figure out how does Alpha … How is the game of Go represented within AlphaGo?
Steve: So maybe just react to that.
Bobby: So I would agree, Steve. I think the assumption is why would anyone imagine we would understand the brain anyway? Meaning … I would say it another way. What do we actually understand about biological systems that isn’t essentially reverse engineering or an engineering principle couched as understanding? Now, this is going to dive us possibly deep into what the word understanding means, whether I understand what understanding means. We get into these weird loops, et cetera. But this doesn’t surprise me at all. It would be similar to saying do you expect a dog to understand regular physics? No. Of course, brains have limits and capacities, and I would argue we barely understand anything about the brain.
Steve: So, but to say something a little bit controversial. If I were to go to the annual meeting of the neuroscience whatever it’s called. The association … And poll people, have them react to that statement to … How many of them do you think would have a relatively deep understanding of what that sentence might mean? Our brain is too complex to be understood by our brain.
Corey: Well, can I jump in here? I think partly it’s just the question’s a little bit grossly formed, right? The idea of understanding the brain as a whole level I think doesn’t make a lot of sense. I think … And this is true of biological phenomena in general.
Steve: And that’s broadly understood by neuroscientists.
Corey: I don’t know. I haven’t polled people. I think so. I think so.
Steve: Well, you live in a community …
Corey: No, I think so. People work on tiny, tiny phenomenon. This is true of biological phenomena in general. Here’s an analogy. Suppose someone says, “How does the natural world work?” Right? How does the earth work? I see these animals, and they get some food, and then they go mate, and they have some babies. Then things die. It’s all so complicated. I mean …
Steve: Well, you just gave a nice compressed description.
Corey: So by narrowing … But I narrowed down to a small, tiny phenomenon. I didn’t try to theory the whole thing, right?
Steve: Yes, exactly. Exactly. Exactly.
Corey: That’s [inaudible 00:42:11] how biology works, neuroscience works. You have tiny theories. Generalizations that don’t go very far, apply to a fairly small region. Maybe there’s a similar phenomenon in another animal or similar phenomenon in another part of the brain, but you basically have micro-theories. And that’s how it works. Now, to have a global theory of the brain, I think is fantasy. Right?
Bobby: I agree.
Corey: Maybe some super-intelligence could have it, but …
Steve: Very good.
Bobby: I’m pro reverse engineering [inaudible 00:42:36]. At some level, I can make the case that I understand brains to a pretty good approximation already. If I didn’t understand brains, how could I live my normal life? I make predictions, which is one version of understanding, about brains around me all the time. Not as a neuroscientist, but as a human being living and working in the world, I can make predictions now, about if I started talking politics with you guys, or sports, or whatever. As a human being, I’m really good at making predictions about brains.
Bobby: The question is whether from a bottom up, I can make predictions. Whether taking these small facts exactly like Corey was saying, and turning them into broad general theories. I’m not sure that’s ever worked.
Steve: So Bobby, let me … In reference to this particular topic that we’re on, let me reference a little short story that I wrote on my blog. [inaudible 00:43:33] as a character in this story. So you’re working in your lab with your slicer and dicer and electron microscope. You’re working very hard, and you’ve got 1000 grad students slaved in harnesses doing whatever they need to do. Now, I come in from down the hall carrying a little box, and I say, “Hey, Bobby. Look what my former post [inaudible 00:43:54]. He works at Deep Brain now. And this is the very first artificially intelligent device. It actually has passed the Turing test, it passed the advanced Turing test, it passed Turing Test Five. And look, it’s a little neural network that they trained in a virtual world. So, they actually have the neural net, and they trained it. But then they instantiated it in this little blob-y silico thing.”
Bobby: We’re going to stop you. You need to explain what’s the Turing test? What’s the advanced Turing test? Turing Test Five? Just briefly. I know we don’t have a lot of time today.
Steve: Turing test is a functional way to determine whether something has quote real intelligence or that an artificial intelligence has really past some milestone. And the way the test is conducted is by just having people probe the artificial brain and ask it questions, see if it can learn things. And if it can fool them into thinking that it’s a human, then in some sense, it’s got at least sort of human level intelligence.
Steve: Anyway, in the little story, I run into your lab and I’m carrying this box. And the box is a physical [expansiation 00:44:59] of a neural net that was trained maybe inside a computer, but it now can do all kinds of things that the human brain can do. Maybe it’s even smarter than the typical human. And then in the story, to poke fun at you, [inaudible 00:45:12] that is Bobby in the story, rushing over, grabbing the box out my hands because you want to immediately start imaging it to see all the neural connections that are going on inside the box.
Steve: And I weakly try to explain to you, “No, Bobby. We’ve got that all. I could print it out [inaudible 00:45:29].” But you were just really obsessed that you wanted to measure all of the connections between all the little sub-pieces, because after all, this is a super-intelligent alien being that we’ve just discovered.
Steve: What do you think of my little story?
Corey: Is that in fact what you would do, Bobby?
Bobby: Wait sorry, which part?
Corey: The … If you had the brain, would you be inclined to try to image it if there was a data set showing all the connections were anyway?
Bobby: Fair point. I mean, as Steve said, here’s this box and here’s what we call the wiring diagram, which is how every neuron connects to every other neuron. I’m generally not a trustworthy person. Sorry, I’m not a trusting person. I think that was a Freudian slip. I would think about whether I trusted Steve or not, and if I didn’t, I would redo it. But if I trusted him, I don’t see any reason to go to redo the wiring diagram.
Steve: That part of the story was an exaggeration, because if you really understood that we had actually produced the thing through artificial means and therefore had the wiring diagram stored somewhere else, he wouldn’t grab the physical object and … But it was meant to poke fun at this … Perhaps what could be interpreted, and I think maybe some other neuroscientists interpret, as an obsession with measuring …
Steve: From … The gain from which is uncertain, that …
Steve: [inaudible 00:46:49].
Bobby: In fact, Corey, another thing that Steve has called me in the past is the ultimate stamp collector, that I’m literally interested in collecting every stamp of every connection of every neuron in a brain. And there were times where that was useful. Darwin cataloging all the things he cataloged with the Galapagos is stamp collecting. I think it’s a very fair question to ask, is there a return on this? Do I have to collect every stamp? What if I just collect 30 percent of the stamps? Would I be okay? I don’t have good answers to these questions, because it hasn’t even been partially done yet.
Steve: I think my initial reaction, when I learned about the really … The big advances in brain mapping and how far we had come was wow, that’s amazing. We should just do it. In terms of the total neuroscience budget or scientific budget, it’s not such a big fraction. And so why not just get a decent X level resolution map of the brain? Let’s do it. Corey has argued that it’s clear that we don’t really need that right now. We’re not going to get all that much out of it. Maybe they’ll be surprised that people didn’t anticipate, but people can’t point to, right now, a high probability gain that we’ll get from such a detailed map. Is that fair?
Corey: Yeah. I just … I’d say let’s look at history, right? We’ve now got some pretty good small maps of different brains. And there’s no doubt we’ve learned some things from anatomy, but again, I think anatomy works in conjunction with physiology, right. You first learn where connections are and you probe those connections by measuring electrically from them. And then you may begin to get an understanding of how a certain set of cells reacts to stimuli. But that’s where I think, personally, I see the value in anatomy. Instead of going whole hog and mapping everything, you basically make a part of a broader theory development program, where you know where things connect, you learn how those connections behave, they try to understand a small region of the brain rather than going for …
Corey: So that’s very different from your approach, Bobby. I’d like to hear your reaction to that.
Corey: I’m defending conventional small scale neuroscience.
Bobby: [crosstalk 00:48:57]. Sorry, Corey. I didn’t hear the last part. I apologize.
Corey: I guess, I’m really giving you a … Use the Kuhn-ian statement. I’m giving you a statement of the current mainstream neuroscience paradigm.
Bobby: Yes. Yes. So, I would … And then take this with a grain of salt, because I’m super biased. My argument is that this kind of functional mapping of the brain is what’s wrong with why we don’t understand the brain. And I’ll argue, if you don’t mind, specifically for neuroscience but also by analogy. The kind of analogy I’d like to make is to another big mapping project, the human genome project, which also, when it was introduced, people said, “Geez. This sounds a little unnecessary. We already know where all the genes are. Why do we have to map all the regions that aren’t genes? Why do we care about that?”
Bobby: And it turned out, of course, that the comprehensive map revealed that the non-coding regions … What I learned in college was called junk DNA, is hardly anything but. That was the Kuhn-ian moment, perhaps for genomes. Once we did the whole genome, we realized things that we didn’t before. That’s the first.
Bobby: The second is if you think about how the genome worked, nobody ever mapped a functional genome. That’s because the function of something and the way neuroscientists test it is just an ad hoc made up explanation for what a human brain thinks the function of a neuron is. We’ve really gotten really far with this. There are visual systems. There are para-visual systems. All of these things of human beings from ablation studies, from functional recordings, have set up these maps.
Bobby: If you made the functional genome and you said, “Well, let’s map the function of every gene,” it would be a complete … I can’t curse, but it would be a complete S show.
Steve: You can curse [inaudible 00:50:50].
Bobby: The reason is that every scientist who works on a particular gene thinks it has some weird function related to whatever it is that their personal history of working with that gene is. Number one. And number two for neuroscience, the really crazy part is when we test stimuli on animals, we’re testing, I don’t know, less than .1 percent of the universe of stimuli. We never look at stimuli in combination. Very rarely audio, visual, olfactory, et cetera. And the response properties are super simple compared to the wiring diagram, or the 10 thousand connections that neurons have. There’s something wrong about this. There’s something wrong with sampling a few neurons with a paucity of a stimulus space across many different conditions, who knows whether their eyelids are dilated, what their personal history was just before these functional maps were taken. And then trying to collate it, accumulate all of that and make sense is as crazy, I think, in my mind as making the functional genome.
Bobby: And the anatomical genome, the anatomical map, what I like is there’s an easy start and there’s an easy finish. In the genome, I started with the first base pair, and I got to end with the last base pair, and say, “I’m done. Here’s the data set.” I can do the same thing in an anatomical map of the brain. I can start with the first connection, end with the last connection, and say I’m done.
Bobby: In these functional worlds, I feel like it never ends. At least I haven’t seen it converge in my time in neuroscience. Okay, was that controversial enough? I’m really trying hard.
Corey: That’s pretty controversial.
Corey: I think there are many ways to respond to that. What I say is look … And I’d like to begin with one argument, which is that again, if you’re looking at a global theory, right, I think your argument may hold water. We’ve learned an enormous amount from neuroscience, right, by focusing on very small of the molecular circuit phenomenon. I mean, all the lists that people who won Nobel Prizes, often they worked out some particular system, at least some of the mechanisms of some particular system.
Corey: So take the lab that I was in, Richard Axel, right? There, they work on smell. And part of the idea has been … First of all, they first cloned the genes in mouse and then they went to flies, which is actually an unusual direction, going to more complex to simpler. But they now begin to understand how the olfactory system works. People can record from hundreds of cells right now. It’s a fairly small brain, and we do have an understanding now of how smells are processed in the olfactory system of these animals. So again, we’re not aiming at a global theory, but nevertheless, if you view neuroscientists simply another version of biology where small theories are what dominate, then the approach seems to have been bearing fruit. Whereas this whole anatomy mapping hasn’t really born fruit so far as I can tell, right, at a global scale. It seems like we’re trying to solve a problem, which … I thought we agreed may not be solvable, which is the brain at a macro level.
Corey: So that’s one of my first responses. Now, I think it sums up the analogy, I think, between the functional genome and say, functional neurons, is actually … Has led … Is different. We have learned recently that there’s cells, large number of cells in the brain, right? Glial cells, which we thought essentially had no function whatsoever, and now it’s clear that they do. So that’s been a revolutionary development. Maybe not … Certainly not on par with realizing non-coding DNA, is in fact regulatory and controls many functions of the brain.
Corey: But we now know …
Steve: To say something controversial, I would say anybody who was smart knew that that could not all be junk DNA, that protein coding could not be the only thing that’s happening inside your genome. But anyway.
Corey: And maybe to say the same thing of people who thought the glia, right, served no function of the brain, but here’s a case where at least there was a consensus over a large part of the field about whether [inaudible 00:54:52] is a function. And now we realize that they actually do have functions. Here, we found the case of the brain, some region of the brain, which was thought to be non-functional now, and opens up a whole range of investigations we can do. So I said that … So I guess I made a couple claims, Bobby, which is again, I still think that the standard paradigm has worked extremely well, in the understanding we’re going to get small, local theories. The best thing to do is probably start with simple brains, because those are ones that you want to have an understanding on the global scale. You have a better shot than the human brain.
Corey: And if there is a case where we actually have come to see this something, which was punitively non-functional is functional.
Bobby: Okay. Just, on that last piece, only because I like arguing, so I apologize.
Steve: Two minutes.
Steve: Take two minutes and give your rely, because I’m going to switch the topic.
Bobby: Okay. [crosstalk 00:55:37] that I just want to make a case for simple and complicated brains, because I’ve seen this rub, or I’ve seen this shtick a few times. And I want to tell you my version of that shtick.
Bobby: My version of that shtick is the average lifespan of drosophila is a couple of days, maybe. A week, two weeks. I don’t know the exact number, but you tell me.
Corey: I think they can … We don’t know the natural lifespan, but they can live in the lab up to 60 days.
Bobby: Okay. Two months. No problem. The number of progeny born per reproductive cycle is over a dozen, maybe hundreds, et cetera. If you look over evolutionary time, they’ve had way more cycles of evolution than we have. The average lifespan of a human, okay, absent whatever, is decades. The average number of progeny born per reproductive cycle is way less. So evolutionary pressure has been working dramatically on the fly. And in fact, it’s a much more evolved organism than we are.
Steve: [crosstalk 00:56:39] true of insects in general.
Bobby: [crosstalk 00:56:40] These numbers alone. And a lot of neuroscientists try to pull this bait and switch where they say, “Well, I’m going to work on a simpler system like the fly and figure out mechanisms and et cetera,” but it’s actually a bait and switch. These animals, these invertebrate animals, are way more complicated than we are.
Steve: [crosstalk 00:56:56] They may have undergone more natural selection than we have …
Bobby: Yes. [crosstalk 00:57:02]
Steve: But in terms of the structural complexity of their brains, clearly less, right? Number of connections.
Bobby: I’ll give a specific example of this.
Bobby: In the fly and in the worm, which is another model system, it’s not that individual neurons compute. At least in the worm, different parts of the neuron compute. So a dendrite, a piece of a neuron, might compute something completely different than the axon of that neuron. And those two might never talk to each other. And that immediately makes the system way more sophisticated at some level, or complicated, than these integrate and fire … The way we think about the entire neuron firing in the mammalians.
Steve: Okay, so just …
Steve: Let me reformulate that a little. So you’re saying that from a naïve analysis, it seems like the drosophila or whatever, is actually less complicated than say, our brain. However, because it’s undergone so much evolution, it may be processing information in a much, in a radically different and actually more complex way, even though it has fewer atoms in it.
Bobby: Yeah. In fact, one version of this, but you guys please push back, is that a lot of these simpler organisms are analog computers. They don’t actually work in the digital … A large chunk. Large fractions of their brains are analog computing. They don’t fire what we call an electrical spike when they reach some threshold. And I don’t know the math, you guys will know it better than I do, but analog computing in certain ways is more sophisticated than digital computing, or complicated, or et cetera. So I’m not sure if the inference, the understanding is going to flow from invertebrates to us.
Steve: I think you’re right to say that even if computer one has N1 components and computer two has N2, and N1 is much larger than N2, it doesn’t necessarily mean that the functioning of computer one is more complex than number two, because of the reason you gave, that the actual mechanism by which it’s computing could be quite different.
Bobby: And we have redundancy, apparently. Humans have redundancy in ways that the fly doesn’t. You could remove a substantial part of a human’s brain, either by injury or et cetera, and a lot of the things that they do don’t get affected. I don’t know if the same is true in this world of ablation, in a fly brain, where things are super specialized and perhaps less redundant.
Steve: I’m going to cut it there unless Corey has something really urgent he wants to say.
Bobby: How could you not, Corey? I’m trying my hardest to [inaudible 00:59:33].
Steve: Well, I’m not saying that you were right, but there is a formulation of what you said, which isn’t obviously wrong to me.
Corey: Yeah, look. I don’t want to delve too far into this, right.
Corey: There’s no doubt that invertebrate neurons have many different functions. They’re often not uni-directional in the same way that vertebrate neurons are. And they do … They often do perform different functions the way that vertebrate neurons are actually much more specialized. And seems to be the luxury of having so many neurons. We can have neurons which are relatively specialized, although perhaps not as specialized as neuroscientists have hypothesized in thinking that this neuron performs this particular function.
Steve: Okay. I’m going to cut you, because we’re running out of time.
Steve: Back to my … One of the topic sentences that I mentioned earlier, scientists are underpaid, and society under-invests in scientific research. And I want you to react to that, Bobby, first. And I’ll just mention that you and I have a common friend, maybe a guy you went to Princeton with, you were Road Scholars together, and this guy is running a monster hedge fund and killing it, and every time you talk to him on the phone, you feel like killing yourself. Am I exaggerating?
Steve: Okay. So tell us. React to my sentence. Scientists are underpaid, and society under-invests in scientific research.
Bobby: Okay. I’m going to disagree. I’m going to be very specific.
Bobby: I think I’m underpaid and under valued.
Steve: Okay. Good. Good.
Bobby: And I happen to be a scientist. I don’t know if that fits exactly into your motif. But … A net total amount of payment, not payment per dollar, not payment per scientist. It’s actually the opposite, I suspect. We have way too many scientists doing neuroscience right now. So overall, I think …
Corey: Way too [inaudible 01:01:16] relative to what, Bobby? Sorry. Relative to what scale?
Steve: Return on investment? Or …
Bobby: [crosstalk 01:01:20] What we’re paying and what we’re discovering.
Corey: How would you know what’s appropriate to … I mean, how can we run this experiment right? Do we have another control set where things are being run properly and we’re discovering lots and lots of things? With the same number of scientists? Right? I mean, how would you defend that claim? We have one experiment, which is people doing science right now in the current world. How do you know how much you should be discovering, given we don’t know what half the problems are?
Bobby: Yeah, so one … That’s a really good point, Corey. Thank you for pushing me back on this. I would argue that one way to analyze this is the more money and people you throw at a problem is the answer converging or is the answer diverging? The more effort I throw at something, am I arriving at F equals MA? Am I converging on this idea that I’m going to arrive at F equals MA? Or the more people that I throw at it, the more equations actually come out, that the results are actually diverging. We’re not getting towards a unified theory of the brain. We’re actually getting to more theories.
Corey: But remember, our previous discussion. What I’m trying to suggest, there is not unified theory of the brain, right? There are many, many tiny theories of small parts of the brain and in small phenomena. So see, in the brain, there is no … There’s no special relativity. There’s no general relativity. There’s no fundamental laws of particle physics for the brain.
Steve: It’s sad, Corey. What … [crosstalk 01:02:40]
Bobby: Why shouldn’t there be, Corey? Why is …
Steve: [inaudible 01:02:42].
Corey: The most complicated, you know what I mean?
Steve: I’m kidding.
Bobby: No wait, let me push back, if you don’t mind. It’s just physics and chemistry. It’s not magical, and we have taken complicated systems like hurricanes or … And sort of understood them. I see Steve saying no already. So I’ll back it up.
Steve: Well also much simpler, too, in many ways.
Corey: Yeah, yeah. [inaudible 01:03:08] simpler.
Steve: Keep talking.
Bobby: But how are they … Yeah, simpler is this word I just … I never understand. Neuroscientists use it all the time, et cetera. But there will be an equation that … A global equation. I don’t know how many parts and how nerdy I have to be to understand it, but there has to be, right? Now, one version … It doesn’t have to be. Let me just push back on myself. But it will totally screw all of us [inaudible 01:03:32], which is that the reason I can’t analogize from physics or chemistry to biology is that biology has evolution thrown into it. [crosstalk 01:03:45].
Corey: That’s the key. That’s it, Bobby. That’s everything. That’s everything in this thing, right? It’s a haphazard system thrown together to keep organisms alive and reproducing over time.
Bobby: Yes. Agreed.
Corey: No reason to think there’d be systematic laws governing everything that’s happened there.
Bobby: Yes. Agreed.
Corey: If something gets there that happened to have worked, it stays.
Bobby: Yes. And that could be, actually … And I suspect that is actually what we’ll find, is that there are very few principles when we do these kinds of global brain analyses. And each little part of the brain is going to have its own little rules, how to wire up, how to imagine et cetera. I can buy that, but at some point they’re all connected to each other, meaning … Let me say it another way. Every neuron in a brain has to be connected given a certain number of connections to every other neuron in a brain. It has to be, otherwise you have two brains. Two independent brains. I’m going to make that argument. I suspect it’s true, but I have no idea.
Steve: So on the nature of biology and evolution, I don’t think we’re disagreeing, actually. We’re actually probably almost all in the exactly the same spot.
Bobby: 100 percent, I agree.
Steve: But I want to come back to the poignant plight of mid-career scientists like yourself. You’re not going to …
Bobby: Please, Steve. [crosstalk 01:04:59] …
Steve: Are you going to lie the whole …
Bobby: Early career scientist.
Steve: What’s that?
Bobby: I just got started really late.
Steve: Okay. Scientists at your current point of career.
Bobby: Got it.
Steve: I think the listeners and the viewers want to hear something about what your day to day life is like, the emotional roller coaster, why science is hard, are you happy you went into it? Are you happy where you are in life?
Bobby: Most of me is unhappy about everything. That’s my general view, et cetera, but if you want me to drill down on specifically why I’m unhappy about science, it turns out that at least in biology, at least in my version, once you get your own lab, you never actually do experiments anymore. You never actually touch … Oh, sorry. Never is a rough word. It’s very rare to actually be the person who discovers something, to actually be the person who solves …
Steve: You’re managing others.
Bobby: Yeah. And in fact, my life is the opposite. I’m somewhere between an accountant and a salesman and an HR rep. That is what being a scientist is on a day to day basis. I have to worry constantly about money to keep the lab going. I often have to worry about relationships in the lab that are preventing us from getting data. I have to think about who I should put as reviewers for my paper, and who I should exclude as reviewers from my paper. There’s way more human level thinking and relationships, and I hate it, dude. I’m terrible at that. That’s why I didn’t want to be a doctor.
Steve: I hear you. Do you think what his … Do you think his comments are comprehensible to the modal listener or should we elaborate on the things he just said?
Corey: I think we have to elaborate, because I think there’s a picture of science, which is …
Corey: Much more isolated.
Corey: And much more pure than …
Steve: So I think the average person thinks a scientist goes in the lab and is thinking deep thoughts and maybe tinkers with their device and has a eureka moment. And everyone immediately realizes it’s a big discovery, and they shower him with laurels and et cetera. When in fact, every day is a huge struggle. You’re managing a team. The team is really making discoveries.
Corey: If you’re successful enough to have people working for you … [crosstalk 01:07:15]
Bobby: Yes, exactly.
Steve: Well, I’m assuming someone’s succeeding in the scientific enterprise. You are begging funding agencies and foundations to give you money to continue your projects. You’re worried that you’re going to have to fire a postdoc or a grad student if a particular grant doesn’t come through. When you finally get your beautiful results, your churlish peers fail to understand why it’s important. They misunderstand your clearly written paper. They reject it for the wrong reasons. Everyone is out for themselves. Did I leave any bad aspects out, Bobby?
Steve: We’re sleep deprived.
Bobby: In fact, we just had … Yesterday, we had a meeting with all of the assistant professors in the biology department, meeting with the guy who decides our tenure. And it was an hour long meeting where he said … He wanted to teach us how to get tenure successfully. University of Chicago has about a 70 percent tenure rate.
Steve: Pretty high.
Corey: That’s very high.
Bobby: [inaudible 01:08:09] biology. And I sat down. And he started off with you got to publish papers, you got to get grants, and you got to be a good colleague. Or sorry, you have to teach. Excuse me. Publish papers, get grants, teach. Right? At least that’s what everybody tells you, but not all … That has very little to do with getting tenure. It has a little bit, but not as much as you think. And what really matters for getting tenure is basically the advertising campaign that you engage in.
Steve: Self marketing.
Bobby: Along the way. And he was right. He had a bunch of reasons why he wanted to do this. So as soon as you get a paper out, I don’t care anymore, dude. I already … In fact, I stopped caring before we published the paper, because I already knew the result. And I’m interested in the next thing. But that’s not how it works. You got to publish that paper, and then you got to … What do they call it? Go on a book tour.
Steve: You know …
Bobby: And I don’t want to … That part stresses me out, man.
Steve: It’s … So, it’s funny that you say that, Bobby, because yes. I mean, the way people market themselves, the way they push their results, I think has a huge impact, at least at the elite level in science, whether you’re going to become a superstar. At Michigan State, I … It’s funny. Right before we recorded this, I recorded a video message about the promotion tenure system here, because I’m actually on the committee that does that here. And I was trying to emphasize how things are fair, and you’ll be judged on your publication record, and your grants. And I did nowhere in my presentation, my video that I … My little video that I recorded did I say anything about self promotion.
Steve: But realistically, we know it’s true. And apparently at Chicago, it’s very important.
Steve: We’re out of time. So, but I want to give you a chance to do one more thing. It’s the last item on my list. I heard you once say that professors must profess, and I like to say professors love to profess. So I want to give you a chance to react, or to answer the following questions. Where will neuroscience be in 10 years? And in 50 years? What will drive this progress? What are the most exciting milestones that you anticipate? And how should resources be reallocated, if they should be reallocated, within the field?
Steve: And I would love to hear Corey’s answers to these questions as well, but not today. And we have plenty of time to discuss this in another …
Bobby: Okay. Sorry, Corey. By the way [crosstalk 01:10:33].
Corey: Bobby, go ahead.
Steve: Yeah. So you get the last word.
Bobby: Okay, thanks. That’s like some Adam Sandler movie where he shows up for his final exam and they have … We only have one question, with 18 parts. Or something like that.
Steve: Nah, there’s only six. Whatever.
Steve: You can leave anything out you want to leave out.
Bobby: Yes, please. What I mean but that quote, that professors have to profess things is I’m just sick of professors who don’t have opinions about things. It’s pointless to not have an … To have an opinion. You want to attack opinions, you want to argue. So I’m going to use that thing. I think in 10 years or 50 years, there’s going to be some weird … There’s going to be a very low difference between neuroscience and computer science. I’m not sure which is going to be a subfield of the other. But my prediction is that that neuroscience is going to be a subfield of the computer science world.
Bobby: The reason I think that … I’m probably wrong, but the reason I think that is that right now, you can’t really function a high … At some high capacity neuroscience lab without having a set of your team or your collaborators that are essentially computer scientists. To handle the thousands of neurons that we’re imaging in real time, to handle the billions of connections, to track behavior, and keep track of terabytes of data. Computer scientists are already invaluable for neuroscience. I suspect at some point in the future, I hope neuroscientists are going to become valuable to computer scientists, because there are things about how the brain computes that’s going to be very hard for our hardware to do. The one that I always talk about is that brains are 20 watts, hardware it not.
Bobby: When you’re talking about an algorithm beating Go, an algorithm beating a human at Go, it doesn’t beat that human when you divide it by the energy it took. There’s a denominator. The numerator is how well it does. The denominator is how many watts it took to achieve that result. When you do that achievement over wattage, human brains still [inaudible 01:12:30]. And I think some day, the computer scientists, or the people who build hardware or software, are going to be learning from us. Hopefully. Just like they did with the very first machine learning algorithms, like you were saying, Steve. The very first machine learning algorithms. There’s a very famous Japanese machine learning scientists who based the very first visual machine learning algorithms on the [CAT 01:12:53] visual system. On the network, on the nodes of the CAT visual system.
Bobby: The precedent is there. I suspect the more we learn about brains, the more they’re going to find it useful. And there’s going to be some mind meld between computer science and neuroscience. Of course, I never thought out it, but it seems so obvious now every day, that we’ve both been studying the same thing, which is intelligence. Or not intelligence, but which is how brains get jobs done. And I think that’s what will happen.
Bobby: So that was two of the four questions. I can’t remember …
Steve: Most exciting milestones along the way?
Bobby: I think that really, for me, the milestones will be reverse engineering milestones, will be where the things that brains do, we figure out ways to put into algorithms and robots. And have a list, we’ll start with maybe perception and move upwards. Or it might start with movement or motion. I suspect a wiring diagram of a spinal cord could help somebody make a better robot based on that kind of whatever control theory et cetera. So I think those are the milestones that we’ll see.
Bobby: And in terms of funding issues, you should really fund me and nobody else. [crosstalk 01:14:11]
Steve: How can we argue with that? All right, well unfortunately we’re out of time. Thanks a lot for joining us on the show. Corey, do you have any final …
Corey: This has been great, Bobby. You’re a fabulous guy to talk to.
Bobby: [crosstalk 01:14:23] I hope we get to talk more about neuroscience, Corey.
Corey: I do, absolutely. And look, I think this was a model. If we can get guests who are as fun, engaging as Bobby, I think we’ll do pretty well. Hope you’ll come back, Bobby.
Bobby: Yeah. I do accents. And since you can’t see me, I can come back as somebody else.
Corey: That’d be great. Can you come back as Steve?
Steve: Yeah. Can you pass the Turing test as Steve Hsu now?
Bobby: Yeah. I don’t think I can do Steve, but I have a pretty good Cockney accent if you want to interview a bunch of English scientists.
Steve: All right. Well, we’ll invite you back for the comedy hour.
Bobby: Yeah. Thanks, Steve. Talk to you guys very soon.
Steve: Well, let’s stop there. And if you enjoyed our podcast, let us know. You can send an email to Corey or myself. And we’ll see you next time.