Switch Statement

083: Gödel, Escher, Bach - Bonus - You Can't Whack Circles Out of People

• Matthew Keller

Share episode
Matt:

Hello everyone And welcome to the switch statement podcast It's a podcast for investigations into miscellaneous tech topics

Matthew:

Hey, John. How are you doing?

Jon:

Hey Matt, how are you?

Matthew:

I am doing all right. We have a special episode today. every once in a blue moon. How often, uh, how often are blue moons?

Jon:

Dude, I have no idea. I don't even know what a blue moon is, to be honest. Maybe.

Matthew:

A blue moon is when you have a full moon twice in one month. I,

Jon:

doesn't seem nearly as cool as I was hoping.

Matthew:

dude, you would think it was like some crazy astro logical event, astronomical event. I dunno which of those words.

Jon:

more blue than usual or something.

Matthew:

You would think that'cause there is a blood moon and that is when it's actually read.

Jon:

That looks

Matthew:

Um, but anyway, the event is that we have a special guest today, Dr. Jeremy Bock. Should we refer to you as Dr. Bock?

Jeremy:

Um, just call me Jeremy.

Matthew:

All right, so we have, uh, we have Jeremy, on the line. to talk, talk about genetics and just a grab bag of other things. But, um, but yeah, so I think just to get it, just to kick it off, can you introduce yourself? What do you do? What are you here?

Jeremy:

my name is Jeremy Bock. I'm from the same town Matt is. I won't, I won't dox you, Matt. Um, we went to middle school and high school together. Went to different engineering colleges. I was biomed, undergrad and masters. Then I went to, New Jersey Medical School for med school. Cooper University Hospital in South Jersey for General Surgery Residency. And now I'm finishing my, trauma and critical care surgery fellowship in University of Louisville.

Matthew:

Okay, so what do I need to do to myself to to get you to work on me?

Jeremy:

Um, you need to get shot, stabbed or something. Inside you has to go terribly wrong in which you would require emergency surgery or um, if you have some big surgery, like a liver transplant or big cancer surgery, you can end up in the ICU under my care.

Matthew:

Can I, where do I sign up if I wanted to? To like give my ver to someone?

Jeremy:

You, you could give a piece of your liver to someone. You can't give your whole liver to someone unless you are. That's a whole nother complex conversation that we could get into. But the ethics of liver transplant, uh, death, uh, donation after cardiac death versus donation after brain death and all of that stuff. Uh, but you could give a piece of your liver to someone.

Matthew:

Okay.

Jon:

organs at death, and that's like on your license, did the cops like kill you when they find you at the site of an accident? In.

Jeremy:

No. So that's a, a big, misconception, not cops in general, but doc, like people are afraid that doctors will try less hard if they see that you're an organ donor, but the teams involved with taking care of you and the teams involved with getting the organ are completely. divested from each other and even the, the, OPOs organ procurement organizations that have those conversations with you are not part of the hospital and not part of the system. So we're not even involved in those conversations and, and generally have no idea where the organ goes. Sometimes we know if they were replaced somewhere after someone died, but we don't know, like to specific individuals or where they're going.

Jon:

Nice.

Matthew:

I actually have have a crazy like law related, uh, point here, which is, for a very long time, if you severely injured someone. Yeah, but they didn't die. They could have a good claim against you because like they suffered as a result of you harming them. However, if you killed them. There was nothing for them to do. There was no one alive that you had harmed that could recover from your, uh, your harming them. So these train companies were, were like just operating. It was like the wild west. And if they hurt someone, they would like, they would be like, all right, we gotta finish this guy off because otherwise we could be liable for his injuries.

Jon:

Wait, so there, these are like managers building railroads, just killing their employees.'cause they like hit themselves in the shin with a sledgehammer or something.

Matthew:

I don't think it was employees. I think it was like people who were walking along the, like the right of way of a train.

Jeremy:

how conversations with me and Matt generally go, like I called him after the last episode and talked to him about genetics and we're, the plan was to talk about genetics. But me and Matt, just quickly,

Jon:

You are,

Jeremy:

you know.

Jon:

perfect for this podcast.

Matthew:

We can make this a recurring, recurring segment. So anyway, you are a trauma surgeon.

Jeremy:

I'm finishing my fellowship. I finish in, uh, July, but I have, I'm board certified in trauma critical care and also general surgery. So if, if you wanted me to operate on you for hernia or for, uh, know to get your gallbladder out, I could also do that. You don't have to be critically ill for me to operate on you.

Matthew:

Nice. I have always wanted to like have like an elective surgery just to like, I don't know, like a tune up, take out any of the unnecessary bits.

Jeremy:

Uh, that's another thing. Every single surgery comes with side effects and like, lifelong, like, uh, even like getting a c-section, you have a, a lifetime risk of developing a small bowel obstruction from adhesions in your abdomen. And that's, you know, something that's done. Fairly often. So think about like hernias and other things like that. That's lifetime risk of small bowel obstruction, having to come back to the hospital. Uh, it's not ideal.

Matthew:

Yeah, that's fair. That's fair. But a guy can dream. So the, in the last episode we talked a lot about like the DNA code and, and kind of how, how it represents information. Um, but I think where it might dovetail with your practice is is there any, gene therapy or, the, the care you provide is much more like acute in nature, so I would assume not, but.

Jeremy:

so we're like, in trauma, we're just starting to get like interested in like microbiome and, and, and personalized medicine to see how different people react to trauma and how long it takes them to revert. To a baseline metabolic state, and that could be person to person dependent, but that's just the beginning in surgical oncology. And, um, a lot more, uh, things are being done with gene therapy and um, uh, basically targeting of ca of cancer and things like that. Um, so I, I could just speak specifically to cancer. So the only time I think I've physically ever been involved in gene therapy was, uh, in melanoma, in, in some stages of melanoma. When it's beyond surgery, there an FDA cleared, uh, treatment is called, uh, tve C, which stands for, la. Teve, is a herpes virus

Matthew:

Mm-hmm.

Jeremy:

inject directly into the melanoma. And it, it, uh, the herpes virus codes for, uh, granulocyte, macrophage, colonate colony stimulating factor. So when those cells die. release their own natural proteins. Cancer, like abnormal proteins probably, but also this factor that causes macrophages to come in. The macrophages and, and, and immune system recognize now that the, oh, these proteins are not human, and they can do a better job of attacking, uh, the, the melanoma. Um, this is,

Matthew:

A lot of what you're saying is kind of going, going above my head. I don't know why you drown, but,

Jon:

level.

Matthew:

um, but yeah. So, uh, melanoma, that is skin cancer. Is that right?

Jeremy:

So melanoma, skin cancer in general, and actually University of Louisville has like done a lot of groundbreaking research in the surgical aspect of melanoma. But melanoma, normally, if you catch early. it's a pretty large incision to take out outta melanoma. So if it's, if it's at a adequate depth, you need, uh, two centimeter margins. So a melanoma that is like maybe the head of a pencil, imagine like two inches almost, or not two inches inches on every side, but it's an

Matthew:

Hmm.

Jeremy:

circle and you can't whack out circles of people. So you have to make it an ellipse so you can close it again. Uh, just topographically, so that's even a bigger incision. So you could have a pretty large incision to deal with a very tiny melanoma,

Jon:

saying

Jeremy:

uh, once that met.

Jon:

rub herpes on it.

Jeremy:

So it, uh, no. So it, it's still the standard of care to do that for, uh. Melanomas that are, um, are size appropriate. And again, I'm not a surgical oncologist. The first thing I wanna say, when having these conversations, like anything in, in medicine and life, like, I'm not an expert in tariffs. I'm not gonna talk about tariffs. I'm not an expert in melanoma. I'm an expert in trauma acute care. And critical care surgery. So I will get that outta the way that I'm, I'm not giving advice and I'm not, uh, giving, sorry, I'm not giving medical advice and I'm also not an expert in this area, but I have partaken in residency, we d rotations in, in surgical oncology, colorectal, thoracic surgery, did some cardiac surgery, uh, vascular, pediatric, you know, burn, um, transplant. So we do all of it. So we have a little. Bit of knowledge and everything before we specialize or become general surgeons. Uh, but basically this, this specific instances, a melanoma that's too far gone to get the, a surgery. Uh, so it's another mechanism to deal with it. But surgery would, if you couldn't have surgery, that'd be the ideal thing to do if it was the right stage.

Jon:

can I ask a question about this?'cause, I'm so naive, I'm probably gonna ask a really dumb question, but what I'm hearing is you're, you're essentially like hacking. You're attacking a cancer melanoma with this other virus. I. Uh, and this, this is where I said herpes. You said herpes, right? Or did I just make that up?

Jeremy:

Herpes virus. Yeah.

Jon:

Okay. Sorry. I'm like, am I just hearing herpes left and right? Uh, but you're, you're basically like putting herpes in there so that our body responds to it and the, you're saying the herpes breaks down, drops all these proteins so that our body knows how to attack the cancer better.

Jeremy:

So immunotherapy is like a really complicated topic, you could spend an entire semester talking about your immune system. But basically all of your cells are taking little bits of protein that. destroy and, and putting it on their cell service cell-surface at all times. And immune immune cells are coming by and say, Hey, this is, I recognize this as being human so I'm not gonna kill this cell. Uh, and that's a defense against viruses. Basically. At the same time, there's other cells or the same cells instances where are just like mining the outside of cells saying, Hey, what's over here? When a cell dies, what was in it to see for like, that kind of helps with bacteria recognition more. Um. So your, your body's doing that all at all times. The cancer they've found a great job of convincing cells around it and converting those cells to like acquiescence state almost to say like, we're normal. this is normal stuff. And it secretes hormones basically that, um, the immune cells. Will not attack it. They'll just chill there and say, and then those immune cells will send out signals saying, Hey, everything's normal over here. Everything's normal over here. So what this does is not, the killing of the cancer cell is not that important here, but it's the killing of the cancer cell in of new immune cells that recognize now that these proteins a problem. So then your immune system can attack it

Jon:

Right.

Jeremy:

forward.

Jon:

And this, this actually kind of leads to the actual question I wanted to ask. Like, that was a great sort of foundation for this. discussed this in our last episode, but recently Google DeepMind, as an organization, they, I. Yeah, I, I wanna say they finally solved this great problem in biology of protein folding, where they came up with some sort of series of large language models or some sorts of, of, uh, uh, models, neural networks that are able to fold proteins in all of the various formations. And what I'm wondering is how much does that allow us Uh, unquote hack the human body. Like, basically come up with novel proteins that we can like, put in various parts of our body to like force, know, other parts of our body to kind of respond in certain ways, produce maybe more nutrients that we might need or, you know, I, I mean, I don't know if that question makes sense,

Jeremy:

It does, but, um, it's more important to know the proteins that are there and how they interact the, like, again, that it's not, it's not saying that they're breakthrough is important. It's important for understanding, like, even like prion disease. Like if, uh, do you know what prions are?

Jon:

No.

Jeremy:

Like mad cow disease and, uh, what's it, Fil Jacob Jock Up Disease. That's an inheritable disease. Basically a, a protein misfolds in the brain causes other proteins to misfold as well. So it's like a chain reaction that causes a disease, but it's like, um. It's like the basis of this chapter, almost like the, the most simple form of self replication. This is not an alive thing, it's a pro, it's a protein. One could argue that that's not alive, but yet it is replicating and causing other proteins to misfold. I. Uh, so that would be important in, in that aspect and some other things. Predicting how proteins can interact, things like that. But for treatment, it's more important and we're is getting, um, to understand how the hormones we have work. Like again, we're just finding out how these, uh, was it the GLP one? Uh, uh, medications are starting to work for inhibiting, um, appetite, like that's a, a hormone that exists in our body already. We just had to find out kind of how to make it and what it, what it does,

Jon:

Right.

Jeremy:

and the effects it has on the body and like that. So,

Jon:

need to learn more about this like menagerie of different protein shapes is like step

Jeremy:

Yeah, in med school, we're getting lectured by like biochemists and, oncologists, things like this who like are working on one pathway, like of in the cell and like trying to push it to the next step. And so that's across the body, like everyone's trying to understand a little bit more about that. If you look up like the steroid pathway, it's insane. It just says steroid in cell. Multiple things happen. Like it's not really understood. They just know that steroid turns on a lot of genes like corticosteroids.

Matthew:

Mm-hmm.

Jeremy:

Corticosteroid enter cell, multiple things happen and that's the end of it. Like it's, whereas other, other, uh, processes are more like delineated very well. But that again, is difficult because steroids act on the DNA where other processes may act in, in the cytoplasm or

Matthew:

This actually, I think this kind of dovetails with a, a question that I think, I mean it's probably an enormous topic, but something that has always felt like magic to me, which is like the process by which your body kind of like knows how to like differentiate and like how, like the genes kind of govern the geometry of the body. Um, and I'm not sure.

Jeremy:

Like, that's another whole course. It, it's like embryogenesis and, and growing and from, from a single cell to blastos and, and mesoderm, endoderm, ectoderm, like those three layers germ, uh, then that forms different like skin or muscle like. All those progenitor cells. And they just recently had a, a breakthrough in that, in, in, in de-differentiation. Uh, like I think I looked it up because, uh, it had kind of breached, uh, briefed me on what we're gonna talk about. But,

Matthew:

D. D differentiation.

Jeremy:

Dd. So like in 2006 was the first time from my quick research that they were able to tell, turn a differentiated cell back into a pluripotent cell. But this year was the first time they went from a one cell type. another cell type without going back to the pluripotent, middle stage. So they went, they turned a neuron to a skin cell without going back to its kind of middle stage.

Matthew:

Huh?

Jon:

A brain neuron to a skin cell.

Jeremy:

Yeah. Like, yeah.

Jon:

That sounds amazing, sounds like important research, but did they have like a goal

Jeremy:

That could be potentially important for cancer research. That could be important for like, regeneration and things like that. Um, th this is something that like, basically almost all plants can do by themselves. Like, you know, if you like chop off something of a plant, plant can kind of regrow from what it has. Uh, a root cell can become a whatever cell, or help regenerate that part of it without having any issue.

Matthew:

Do you think that's something that we could give humans? The ability to do? Someone chops off my arm and like a salamander, a new arm grows.

Jeremy:

you know, with, with the future, anything's possible. I can't

Matthew:

I.

Jeremy:

a timeline on that, but, know, with,

Matthew:

You are saying there's nothing physically, uh, to stop, uh, to prevent such a thing, but um,

Jeremy:

I just, it'd be incredibly complex and difficult and require years and years and years of research.

Matthew:

yeah.

Jeremy:

much easier to do that if you lost an arm as like a baby or an infant or a fetus, like, Just healing in general in, in children is completely. amazing compared to adults. Even in like, uh, in like when we do small bowel anastomosis, it's just like connection between small bowel and like adults you have like, either there's a stapler or put lots of stitches in and, and, and like in like infants or kids, you put a couple stitches in and you walk away. And I had a pediatric surgery attending which she would do like four or five sutures and go stem cells and walk away. Like,'cause they have a lot of stem cells available that can help with healing.

Jon:

Amazing. So we discuss, uh, large language models a lot on this podcast because we're kind of obsessed with them. Uh, and I'm

Jeremy:

Mm-hmm.

Jon:

if you have you know, you come across those occasionally or people discuss potentially using them for like to understand these mechanisms in the human body.

Jeremy:

So a lot of surgical publications are coming out now with, uh, like AI model to do this, or a model I model to do that. And, uh, it's like a huge buzzword that gets you published these days, but it's not really as beneficial as one could imagine because the, the benefits of using these large language models when there's so much data that humans can't like the work,

Matthew:

Right.

Jeremy:

Generally, we're not working with databases that have like. All, like any information you can imagine, like in, in, in our, there's a, something called the, the Quip, and it's a database of you, you don't have to be a part of it, but basically it's a national, uh, database where like hospitals input their surgical data and I'm sure there's databases like this for other specialties as well, but for like, this is for general surgery, so I can look up. Again, it's not individualized. I can't see like where it happened or what happened, but I could see like what procedure someone got. What um, if they were a smoker, if they were, had heart, heart failure, and then like the outcomes, 30 day outcomes, basically. Usually like if they died within 30 days, if they were. All that stuff, but that's not that much information. It doesn't tell me how many packs a day this person smoked. It doesn't tell me how many years they smoked for like that alone. So it's not very granular data on, on these big databases to, to actually need some huge model to delineate them. Yes, there is incredible use probably, for these, large language models in again, more basic science research where you're looking at genomes like, genomics between different patients. If you have that data and you have these like millions of base pairs and you're looking for differences and, and you have a lot of results, that'd probably be very helpful. But when, when I have like maybe a database of like. hundred different variables and most of them are not showing up on univariate analysis. And maybe a couple are pinging on multivariate analysis. Do I really need AI to like really churn extra numbers for me, or can I just have a statistician do that for me?

Jon:

interesting. Yeah, I mean, it seems as though I feel like these ais work well in these problems where you have like. Data, that's sort of an input. then you have results, like known results as an output. And you

Jeremy:

Mm-hmm.

Jon:

a model to sort of become this black box, which can convert this input data into this, you know, this magic output results. And you don't even really know how the, how this black box works. as I

Jeremy:

Yeah.

Jon:

Cher Bach, a lot of these like biological. Processes were sort of coming across to me that way, sure this is a a lot in large part because of my own biases, but seemed like there might be just interesting ways to use that technology to help us like, know, understand some of these very mystifying, like hyper complicated interactions.

Jeremy:

No, a hundred percent. And that's why it's, again, I was, it is probably more helpful for these, these basic science and or where you have lots of data in, in, in medicine. You, and again, people, even though that's not the case, generally in surgery, people are trying, attempting to use it for any database they have where before maybe they weren't even sure what to look for and they're just like, Hey, can you putting this data in to see if they get anything out of it?

Matthew:

Are you saying, Jeremy, that, the way that it might work is, uh, a patient comes in and you consult the AI that's been trained on your own personal database as to what it thinks might be the most effective interventions? Uh, is that how you're imagining it might work?

Jeremy:

And they did that even before ai, like just with, with the information they had. But like, even like for trauma for example, like could tell, like, I could put eye color, I could put all these things into a database to tell you. How sick a patient is. But at the end of the day, signs, like your blood pressure is very important. So like a lot of that stuff falls away and nothing will be more important than blood pressure. Like for any model that, like if you get in a car accident, I don't care. If it's a Tuesday and you're this ethnicity and you're an athlete. If your blood pressure is 60 over 40, you're not doing well and you need,

Matthew:

Right.

Jeremy:

like intervention.

Jon:

Yeah. I feel, uh, I feel like I have to use this opportunity to ask you another question because there was a chapter in Godel, Escher, BCH. That maybe for me was like my favorite chapter where he was discussing the brain and he went down all of these tangents. Like he had this section about a grandma neuron. I don't know if they discuss that in med school, but I guess for a while there was this theory that, oh, maybe there's a neuron in your brain that, like whenever you think about your grandma, this neuron is lighting up. And obviously that's not the case. but anyway, he, Posited that when you really break it down, the actual things that are taking place in the brain are fairly simple. You know, you have neurons that are transmitting electricity. and then you have this, I guess glia, I think is what it was called, which is like this structure of the brain, which sort of determines how the neurons can sort of move around the brain. And as you age, this lea sort of erodes and this structure gets more and more, hardened, I guess. And the structure of your brain becomes more and more concrete. and he also discusses this point, which is like a lot of these higher level intellectual capabilities that humans have, having a sense of self or having the ability to distinguish between. I, I think he used the term classes and instances, although I might just be conflating programming terminology, but basically like the concept of a president versus like the instance of a president, you know, Barack Obama or Donald Trump. Like these are higher level mental faculties that say animals don't have, but the underlying mechanisms that are powering these faculties are kind of the same as other animals. It's just that in humans, it's at such a. Order of magnitude, larger scale. Uh, and I just found that to be like so interesting that just from this vastness of scale, that you have this emergence of almost these like new cognitive abilities. And I'm sure I am understanding it in a simplified way and I'm sure also that Douglas Hofstetter was simplifying it. I kind of want to use this opportunity to like. Discuss this with you and you know, you probably have more of a background in this sort of thing, and just get your take on this.

Jeremy:

Yeah, I mean they, they've just started like being able to map, I think like very, I forget which animal. It was very rudimentary like. neuron pathways, it's very co like, so we're not close to this at all for the brain. And that, uh, now that you're saying that, that's a good instance of using large language models and interpreting, um, signals from the brain and, and getting outputs. Like previously there was like these EEG headsets and you could do like some math and try to figure out, oh, this, when the signal went, your arm moved. But that'd be a great instance for that. But again, you need lots of data for that. You would need lots of sensors on the brain to give you data to predict that. But going back to your question, it, it's not as simple as one would think because there's a lot of different neurotransmitters, a lot of different feedback loops. and the complexity is, is significant. Because of that, you don't have input output. You have inputs being supplied to the same output. You have loops and then you have. Like, if you like, you know about addiction and tolerance that happens in individual cells. So if one neuron is getting constant information from, its, its other neuron, it will just down regulate receptors and not respond as much to it basically. So there's constant adaption and there is, you're right, there's constant growth in, in childhood that slowly whittles away to needed. And again, I don't know if that has, was evolutionary due to the fact that we were getting hit in our head earlier in our evolution and required that neuroplasticity again. when we see trauma patients, kids that get significant TBIs have much better outcomes than if you, if you're 90 and, and you bon your head, that could be the end.

Matthew:

Did you say? Uh, TBI?

Jeremy:

Sorry, I doctor speak. Uh, traumatic brain injury.

Matthew:

I think the one last question that I, that I think would be interesting to hear from your perspective, how you see your practice evolving and things in the, on the roadmap for technology that is advancing your practice. Like how is that evolving and when are we gonna get a crazy healing device like the machine from, uh, Elysium.

Jeremy:

Everything is slow in medicine, you need FDA clearance, you need all, all these things to be in place before can implement them. So everything is is slower in medicine research until it's proven to be effective, for that healing device and allium that You need that diagnostic aspect of that alone is gonna be difficult. I think at University of Louisville, there's like one. LLM that's in like approved to be used. That like, I don't remember if it detects pneumonia or pneumothorax, I forget what it detects, but like it gives the radiologist a little ping to say like, Hey, I think this is yes or no to this one specific disease. so to get to there to actually intervening in the body is a whole different thing. Like,

Matthew:

Right.

Jeremy:

right now. Intuitive surgical, the robotic da Vinci, basically the robotic surgery thing has the highest potential for automation in the future. think of it as if they were able to log every surgeon's steps, like they could potentially mimic them, the same time, you need the diagnostic. So we're, we're just at the forefront of it. Again, I, I haven't seen this in person, but I've read about it and seen videos about it, like assisting in the or you put a thing in, it says, I think this is this. I think that is that, to assist with dissection and, and avoiding injury to other structures in

Matthew:

So this is like a heads up display while you're operating and it can kind of like overlay.

Jeremy:

Yeah, and even I, I've actually dealt with, not that heads up display, but there's a medication we give now that when we're using, doing either a robot surgery or even open surgery, if you have like the special lens that you look through, it basically lights up. Where there's good blood flow. So that tell can help us when we make anastomosis to make sure that there's good blood flow on both sides, but also gets excreted into the biliary system, which is in your, like your gallbladder or liver so that you can give it right as you're doing the case. And if there's a lot of scar tissue or difficulty there, you kind of see the biliary system so you don't cut the wrong thing basically. But that again, hasn't been employed and anything yet. I think we're a fair amount away. So, so the only que not question I had for you, but, um, interesting in genetics is that important things are conserved. And that word is, you don't know what that word means,'cause it's like a biological term, but it means there's lots of copies of things that are important. So

Matthew:

Mm-hmm.

Jeremy:

Ribosome, RNA, that gene is many, many places. Is that a thing that's featured in coding or like in, in case of issues? It's, don't worry. It's redundant. Yeah. redundancy, I guess you would say.

Jon:

Yeah, I mean that's commonly, I, I mean that's certainly a concept that's used for,

Jeremy:

Hmm.

Jon:

don't know, things like databases and I don't know how to tie it to

Matthew:

I think,

Jon:

though.

Matthew:

I think one of the things that would be surprising from a coder's perspective to learn that is there's this concept. Which is like drilled into you from a very early age in your coding career, which is, don't repeat yourself. And the idea is like if you have code that does something in a bunch of different places your alarm bell should be going off being like, wait a minute, this is gross. Because like you're doing the same thing like in a million different places, they're probably all slightly different. so that is, that is kind of surprising, but. Generally redundancy kind of takes the form of, like, the code to do this thing is the same, but then we'll run it in a bunch of different places. if I'm drawing a mapping between like a code project and like the human body, it's like, well, the code base is the DNA, and then when I go to run it, that's the human body. And of course, like you're gonna have multiple copies of it when you go to run it. But if you're looking at the source code, you really don't want to have more than one copy of it. Does that, does that kind of answer your question?

Jeremy:

Yeah. and then another thing is, the other thing that programming doesn't have that I'm aware of is, is, um, mutation, random mutation, which even it's more pronounced in viruses, which is a problem with virus and mu viruses mutate quicker than humans, which probably mutate quicker than even fungus. But, um. Viruses basically have a faulty polymerase. They what, what they code for themselves. It's known to be faulty. So when they copy their own genome, they're knowingly doing a bad job of it, but they're creating so many offspring maybe one of them will be better, you know? So that's, I don't think is present in programming as well. Humans obviously are not trying to do that as, as, um, and our, the rate of error is a lot lower than viruses, but still happens and leads to. Evolution.

Matthew:

the thing that that came to my mind was there's actually computer viruses that they have self-modifying code. Um, so, um, because like my understanding is like detecting computer viruses is like, they just have the, this like big list of. Code that like they've seen before. And it's probably not too dissimilar to like the Immune Hu human immune system where it's like, oh, we, we encountered this and now we have a list. Um, but then the computer program will, will kind of like modify its own code and like recompile it and like put itself out there again. So, um, I'm not sure if the mechanism is exactly the same, but, um, but I think. That's the one thing that, that, that came to mind.

Jon:

that. Well, and that's, is that like an example of genetic programming actually, where the A ST itself is kind of like tweaking itself? sort of adding new nodes. Yeah, like A A ST, by the way, abstract Synex Tree. This is sort of how a computer program is represented in memory, where each operation of this program is a node in this vast tree, this concept called genetic programming where. You know, an individual node might sort of expand itself into a set of nodes, or it might delete itself, and in this way you're sort of like modifying a program as it's running and potentially improving an algorithm or coming up with new ways to solve a problem.

Jeremy:

that just reminded me, sorry, of you were talking about long, uh, language learning models. It's also a problem in using'em in medicine is that what are your result, result variables. So like if I wanted to use that x-ray example and say like, okay, we're gonna feed this a bunch of x-rays. We're gonna have this thing tell us if it's pneumothorax or not. What is telling us what it, if it is or not. Is it some, is it the radiologist or is it you doing an invasive procedure to determine if it actually is pneumonia, like a

Matthew:

Right,

Jeremy:

pneumothorax? So sometimes the our, we know is the output is not known. Like

Matthew:

right.

Jeremy:

even like for example, in trauma, if we have an algorithm that tells you. Whether a kidney is injured or not, how would you really know? Unless you went in there and operated and looked at the kidney, you know, you could not be a hundred percent certain, so maybe we're feeding in incorrect into these large language models. This makes it all also difficult.

Jon:

Yeah.

Matthew:

Well, this actually goes back to a question that I had while you were talking, is it seems like in some cases. I don't want to sound like I'm advocating that we like do away with patient privacy, but like if there's like laws about, patient confidentiality that kind of interfere with our ability to share as much information as we would want or in these databases. Yeah.

Jeremy:

European countries that have socialized medicine, they kind of do away with those barriers and you just, you know, your data's kind of out there for researchers to use. not saying those countries are, are the worst in there. You shouldn't do that. I'm just saying that's a different, that that probably would never fly in America and would like, you know

Jon:

should

Jeremy:

where people are.

Jon:

it should be anonymized, but it seems like we could move. Well, I don't know. Again, I'm like naive. I'm not a part of the medical establishment, so I know nothing, but it does seem like we could sort of, I. know, e expand the bubble of, of knowledge by just sharing data and having it available.

Jeremy:

No, a hundred percent. And that has to do also with the, the, just the fragmentation of information in our, uh, health, uh, infrastructure. So like there's, I'll just tell you, there's a million ways to do a hernia, like a repair surgery. There's right now there, and whenever there's a million ways to do something, there's no a hundred percent right way to do something.

Matthew:

Right.

Jeremy:

no one know, like, everyone's like, oh, my way's the best, my way's the best. But say. You do a hernia repair and your patient has a recurrence 10 days after you see them in the clinic and Oh, I did a great job. And then they go on their way. 10 days afterwards, they have recurrence. They go to a different surgeon, say, I have recurrence. That original surgeon will have no idea that person had a recurrence. If they, unless they come to see them again, that's

Matthew:

Right.

Jeremy:

floating in the database to show, oh, that was an issue with, you know, that guy had a recurrence. 40, 50 days out that won't just, that just disappears. Basically, that this person had a occurrence from that surgeon's ability to understand his, his complications.

Matthew:

This is something that's always frustrated me about like my interactions with the medical system is the, like, I don't. Own my data. Like there, like, I wish I could have like a USB drive or, or some chip or something. Maybe there's something that's on me. It's like, okay, whatever notes you put, uh, about me, like put them on this little thing that I have and I can carry to,

Jeremy:

So

Matthew:

or, you know,'cause like another, another ideal system would be like, there's one centralized system, but that feels like, I mean, unrealistic in some ways.

Jeremy:

Yeah. They, again, it could be done if it was a. to people who make money and things like that. or if Congress was pushed by its constituents to do that, but I don't think anyone even knows this is a problem to know, to advocate it for it or care about it. that, that would, be super helpful for. Again, the simple case of hernias where we, we don't know, again, these people disappear. We think, oh, I did a great job. And then someone, you know, they went somewhere else.

Matthew:

Was there anything else, you had Jeremy or I.

Jeremy:

see. Um, oh, just some interesting things that happened since this guy, um, published this. So multiple different types of RNA have been found since he's published it. Sorry, double stranded. RNA Micro RNA, which both have, are feedback mechanisms to basically destroy RNA strands if, uh, so they don't get. Translated into, into protein, mRNA editing. So there's this whole thing called splices zones. I think

Matthew:

Mm-hmm.

Jeremy:

about it, but there, I thought it was interesting. I dunno if he mentioned this, but the promoters can be in regions of, of the DNA that are being read, like in the, there's introns and intros. It could be in an intro of the DNA that that's promoting it to be read. So I thought that was interesting.

Matthew:

Hm.

Jon:

Hmm.

Jeremy:

Just really complex stuff and it's difficult to have analog to, coding. I, I just, it's really difficult, again, in a situation where you're having linear inputs and three dimensional outputs, that it's really more important that the, what the three dimensional output is than kind of the code itself.

Matthew:

I think that the, I mean if, if I'm just thinking like high level takeaways. The thing that's the craziest and maybe the most dissimilar to coding is like. It would, it would be as if you wrote code that was like a compiler that would then go do something else. It's like, it's so self-referential because you're like, you have this sequence of DNA and now this is making some enzyme that's gonna catalyze some other thing that happened. You know, like.

Jeremy:

Yeah.

Matthew:

coding, you ideally want your system to be like very clearly, uh, like all of the layers should be self-contained and they shouldn't be like affecting one another. But in life, it seems like that's happening all over the place. I.

Jeremy:

Yeah, but the, but I wonder I mean, I don't think we could even conceive. Coding with, there's so much self, uh, referring, but it probably would be more efficient to do so if we could magically think that complexly

Matthew:

That's the kind of code LLMs would probably produce. Uh, you know, once they become super intelligent.

Jeremy:

Yeah. A hundred percent a.

Matthew:

thank you so much for, for joining us, Jeremy. Uh, this is, uh, I don't know, this was super cool to just hear about your, your thoughts on, you know, from more like informed perspective about, about DNA and, and the human body.

Jon:

Yeah. Thank you so

Jeremy:

Anytime. Thanks for having me. Thank you.

Matthew:

All right, well have a good rest of your day guys.

Jon:

Bye-bye.

Jeremy:

Have a.