A Brief Talk on Closure

Hello, everyone. Welcome to Foresight Space Group. Today, we're really lucky to have Ulys Sorok with us today. It's been a while trying to coordinate this, so really happy that it's finally happening. But he's the founder and CEO of GRAM. They're an El Segundo-based foundational AI and robotics company. And today we'll be talking about engineering independence. So with that being said, thank you so much, Ulys, for joining.

Thanks a lot, Lydia. I just want to point out what an honor and pleasure it is to be speaking to this group today. I think the Foresight Institute has done incredible work and, you know, the founders really are folks that I look up to immensely. A, a quick note to the audience, a lot of what my talk abstract was initially posited as was more to do with my company itself. I think there's elements of this that I'll be speaking on, but most of today's talk will be focused on kind of a system-level property I call closure, which pertains to self-replication.

And, you know, just a quick brief on what GRAM is before we begin. I founded this company not too long ago after I departed a former lab called The CEntre for SElf-Replication Research. Essentially, I was trying to understand how you can pragmatically engineer your way to making humanity a galactic species. And what we do is at the intersection of civilizational infrastructure and multi-agent coordination, we call collective intelligence, as well as general purpose robots we call insectoids, which are these large-scale meter ant-like machines capable of doing work. Today, we're not gonna speak about that. We'll be speaking about engineering independence as a quality of a system and how that is relevant to the world as we see it today.

So I'd like to first talk on what I think about the current state of play as it pertains to AGI at large. How are we approaching the measurement and optimization of characteristics found within intelligence? There's obviously been a lot of discourse and, you know, R&D put into this specific topic. DeepMind has approached it with their levels of AGI proposition, very similar to this SAE automation level type task. With METR, you see task horizons, and we're on this very solid curve where with every new model release, we're seeing more complex tasks being approached more reliably over even longer horizons. Generally speaking, I think the current ontology is focused roughly around performance, generality, and autonomy.

What I do think that we are missing completely from the scene, because it requires a certain bit of introspection from the part of the major model providers, is the idea of asking a simpler question, maybe along the lines of, what happens if a data center loses power? What happens if a cloud provider goes down for a week? What happens if the team that maintains the serving infrastructure disappears? That system doesn't merely degrade, uh, you know, it vanishes. You know, th-there's no version of that limps along on local resources.

We don't really have fallback modes with our current systems, and this is not necessarily a thing that we scrutinize because it speaks to the survival of these model companies in a fitness sense against their ability to perform in the world of atoms and infrastructure, and not necessarily their core competency of intelligence. The question I really wanna ask are, are, you know, how can we build systems that perform well and remain useful across degraded, sparse, or even alien environments?

So this leads me to my introduction to the folks today on closure. Closure is essentially this idea of how much of a system's persistence burden it can carry itself. And what really I mean by persistence burden is, what does a system need to exist to maintain itself and to grow over time? And, you know, given, g-given the nature of closure, th-there's a lot of things that we could assess through this particular lens.

For example, the primary, uh, metrics one might use to measure closure are simply are material or energy-based closure, and this essentially, as it pertains to a system, speaks to the necessary inputs it needs to produce a meaningful output. There's other ways to kind of think of the different axes of closure, including information, which you could also make an argument for how everything compresses to just information, as well as organization or time.

The other key concept that matters that closure is very closely tied to is this idea of the environments. W-what we need to make clear here is that environment is not simply scenery, but closure is something that's always measured relative to an environment class. For example, systems that can look very strong in a single environment can be useless in another. Think of biology in a very simple way. Certain organisms require very specific environments to self-replicate or maintain their persistence burden.

The distinction is what really matters. You know, tasks, you know, me- the measurement of intelligence can be seen through the measurement of tasks, and that really asks the question, can a system do X? What the environment asks is, under what conditions can the system continue to function at all? You know, I, I like to look at what, what you could call regime properties, which kind of assign to a system its given richness, stability, legibility, or hostility.

So, so in general, there's an idea embedded here which is, as environments grow less rich, less stable, less legible, or more hostile-Closure becomes more binding or necessary or important to look at. So at, at this point, I think it'd be fair to categorize, you know, what different ways we could look at the world through this lens of closure and what specific systems a-apply to it directly. Obviously, there are biological elements to this, and you could look at closure through the le-lens of the unit.

But, you know, what, what I think is more personally interesting is at the macro scale, how can closure apply at the firm level? So, you know, with space, really the concept that matters is the environment becoming alien, and that directly pertains to closure itself, right? So, so coming back to this idea of firm state level of civilization scale, what's most important for examining scrutiny for, you know, space relevant topics is closure at the civilizational scale. Everything that underpins life and industry that as it exists today is through borrowed support of the biosphere, the atmosphere, the gravity well, in-inherited industrial substrates, and so on.

With the question of what makes a civilizational... a, a civilization truly robust in a space-based expansion sense, one can make the argument that the civilizations that have the highest closure are the fittest to conduct expansion, so to speak. So, you know, continuing down the line, say, I hate to bring it up again, but I do think there's a factor of intelligence that becomes very necessary when it comes to assessing closure as it pertains to space-based colonization.

Because what intelligence allows that is distinct from closure itself is the idea of portability, which means that systems that have a high degree of closure, and this could look like anything from a von Neumann probe to, you know, a highly intensive and a high investment-based civilizational bootstrap mechanisms. They could be measured through closure, but what really we should aspire towards is the ability to expand in through as many environments as possible. Intelligence makes this easier as a multiplier effect, but that alone is not enough.

[Creon signals to talk more about space.]

So I think, you know, I, I realize with Creon the point you make is we should jump to more of the space stuff. That-that's not what I initially prepared for, but I think it makes for wonderful digression of the topic of what makes, what makes closure relevant at the species level to colonization to begin with. And then this was the question I sought to answer for most of my adult life, the question of how you actually make humanity galactic. You know, the answer within that for me was you need to build systems that are capable of building more of themselves, and the extent to which that can be reasonably measured as a function of time and volume is this idea of closure.

And, you know, coming back to the assessment of how one might build systems that are more in line with closure, it, it is-- it's, it can be the likely speculated that the systems that are able to conduct large scale colonization of the galaxy will have very high and frequent alignment with closure, uh, closure-based engineering signatures. If we take a system like a, a von Neumann probe, for example, the more redundant it is, the more repairable it is, the more substitutable it is, the more local flexibility it has as to being able to work with a more heterogeneous or diverse pool of resource inputs, materials and so on to produce industrial outputs, the more likely it is to succeed against alien-based closure systems.

And this all comes back to my indictment of our current trap within society. I think we are all too enthused about building systems that are more performance-based, more intelligence-based, while not necessarily considering the systems they sit with-within and the dependencies they rely on. So, so yeah. All to say, I think a, a very great fear that I have is the, the relation of what I call like two very end and evident terminal states of civilization.

The first one being the galactic future, you know, or, and the second one being what I call the introspective future. And these sit in very close relation to, uh, each other in the sense that one of these prioritize the spread and active effort towards climbing up the gradient of closure, while the other highly incentivize this idea of optimizing for intelligence. In the introspective future, there is greater onus to simply make intelligence cheaper, to make compute cheaper, to do things more efficiently with the energy we have available, and to, quote-unquote, "coast along as a civilization." And this is, you know, from a SETI sense could be seen as a optimal state to kind of prolong your existence as a species, and the introspective future can very well be present in a civilization that is Kardashev II or limited to its solar system in nature.

You know, the interpretation I have of the galactic future is much broader and expansive and vast than that. With a closure aligned civilization, you have the opportunity to essentially assert a stronger position over civilizations that perhaps have opted for the introspective future and undue, you know, optimization. And from a perspective of what is most right for human civilization as it stands today, I think we should definitely trend towards the latter oppo- as opposed to the former. My fear is that we are trending quite heavily to the former.

So I think at this point, it's probably useful to open up the floor [chuckles] to questions, and we can talk through these ideas more, especially as how, uh, uh, you know, how they pertain to space colonization and-

Okay, great. I have a bunch queued up, having read some of your stuff. So first of all, this is pretty philosophical compared to the normal sorts of things we see in this group, which are more engineering and s-space in particular. But let us, if you don't mind, or l-if you-

Sure

... don't mind commenting, take a brief step back from galactic, and let's just look at this solar system. Now, our current civilization, including all of our AI, relies on humans, which are, rely on the biosphere, which rely on the geosphere, which rely on the sun. You know, the, it's a whole... And to the extent that there is closure, it goes, it's like turtles all the way down to the Big Bang. You know what I mean? Like, it's only closed if you keep extending. Could you mute your phone, please? It's only closed to the extent that we c-keep including lower level, lower levels that make us, quote-unquote, "closed." So, like, AIs don't live by themselves, s-human civilization doesn't live by itself, the biosphere doesn't even exist by itself.

Okay. So why would a Kardashev civilization that has mastered the entire solar system be any different? Like, it could very well depend, in some degrees, on Earth and all the things we make on Earth, and, you know, of course, stuff will be made in space, but some stuff will be presumably supplied from Earth. I mean, is that closed? Is that not closed? Does it matter?

Yeah. I think the way nature kind of res-resolved this was by drawing a thermodynamic boundary over what it calculates within its closure requirements. You know, for us, it could look... You know, the type of question you're asking is, can be in some ways looked at as an accounting question. Where we really draw the box around human civilization as it exists. Is this around the Earth's biosphere or the solar system or the entire galaxy? I think that what's most robust is propagating, propagating instances of closure that are highly independent from one another.

In the sense of our current tract of expansion, in order to truly fulfill, you know, robust expansion on a timeframe that matters, we have to remove as much latent dependency as possible, even as it pertains to our reliance on Earth-based manufacturing. And I think this is-

Well, wait, wait a minute. That is a claim you're making-- I mean, you're stating this as some kind of, as some kind of an axiom. I mean, that's just a claim. Like, I claim we wanna be dependent on the Earth. A, why not? We've got it. B, if we aren't dependent on the Earth, the whole thing can go rogue. I mean, why do we not wanna be dependent on the Earth? What is the reason for that, especially if we're limited to the solar system or the inner solar system?

Because I, I-- and I think this comes to, I think this comes to the question of how, uh, humans have historically performed over long timescales. It's very hard for us to hold, you know, generational plans. Th-this is, you know, fraught with, fraught with many instances of us being reliant on, on, on things that are definitely out of just the sphere of engineering, you know. Admin, different types of administrations, different types of budgets, different types of civilizational attention spans at the given moment. You know, the big problem once we get to Mars is gonna be convincing everybody on Mars that, you know, the, the next big jump is, is worth undertaking.

And I don't think that's something to necessarily scoff at. W-what I'm merely arguing for is that by virtue of making systems that are increasingly more independent, we can effectively guarantee that expansion will be persistent as opposed to highly reliant on our societal attention span. That's, and that's, you know, it's not really a, a negative-

Okay, I get it. I get it now. That makes sense. I mean, i-if you want independence in term for robustness, like, yeah, eliminate dependencies. Fair enough. Got it. I mean, we may not have that luxury, but if we do, we may want to exercise it. Okay. Hey, look, next question. Next question. Your prior work, as I saw it, had a bunch of stuff about insectoid robots.

Right.

What happened, and what about that? Do you wanna talk about that? That's kind of more in the sort of traditional wheelhouse of what this group deals with. Do you have anything to say, prepared about that? I know it's kind of out of the blue, but [chuckles] still.

Well, yeah, I, yeah, I mean, right now, I've been very intentional about how we speak about this fr-from an information perspective. But yeah, I can take the time to speak about insectoids and what my current company does. The core realization was that if you want to make work possible at a large scale in, in, in a way that is closure-aligned and driven towards really the core metric of growth, you start losing things that may be s-social niceties and then things of this nature.

And in, in asking what the most optimal form factor for work is, something that is highly redundant, highly manufacturable, is capable of navigating kind of gravity-agnostic environments. You know, the three-dimensional, unstructured, and extreme world that exists, which, you know, I'd argue ninety-nine point nine nine percent of the galaxy is. You know, the, there's this whole underlying assumption that current industry makes that all environments are human-based environments, which, you know, kind of strongly refute outside of the realm of, and the service economy per se. Even heavy industry is not a human-built environment.

The, the-- it's highly antagonistic to the bipedal form factor. We have to bend the world in so many ways in order to cater to keeping things alive or running, even to the form of maintaining or repairing or even constructing this. Humans are just the best option we have so far. So yeah, you know, starting from the standpoint that you don't necessarily want something that is built on false pre- false pretenses and something that is focused on pure throughput, you know, there, there are certain form factors to consider.

The most pragmatic of which I think for us ended up being the insectoid, which is a meter scale general purpose robot that we're currently building.

Okay, couple things on that, mind, and then we can open it up.

Sure.

First of all, the most general purpose is arguably the humanoid one, if only because all of our civilization is built for humanoids already, you know? And so if you want to slot something in that can do any job, then a humanoid is not-- a humanoid is gonna do it, and a one meter size crab ain't, okay? The next thing is, and this is something I have to thank Luke for, who turned me on to it-

Have you seen Hail Mary? Just curious.

Uh, no, I haven't, but just a sec, Steve. Sorry. One last thing here, and then we turn it open, open to Steve, is I have to thank Luke for this. The-- it is interesting, by the way, that the, the crab or the insectoid, but particularly the crab form factor, as I understand it, Luke, you can correct me on this, it evolved independently many times. It's, like, so good. You know, this, like, armored thing with a bunch of legs and some claws and some eyes. Like, you know, you got crabs, you've got mites and ticks, you've got... I don't even know. Luke probably knows.

What do you say about that?

Yeah. It's called carcinization, is the term you're looking for.

And it's just a form that has evolved independently, the, uh, convergent evolution because it has utility and in this world, right?

Yes. Yes.

Not necessarily in space, but who knows?

I mean, I'll quickly take Steve's comment as well. Hail Mary?

Well, it was just a joke when you said, "Hey, you know, gosh, the ideal form factor is a humanoid." I actually agree with the premise of your question, which is if you want to pr- do a landing party to prepare assets for humans, the forward compatibility for that could be important. If it's robots all the way down, then you'd have a totally different design space.

Yeah. So, so I think I have two points to make, you know, o-one of which is an aha. The insectoid really, the, the premise it sits on is it's downstream of its core capabilities. So the robot that we've built today is capable of vertical locomotion. So imagine something that can navigate on three-dimensional terrain, much like an insect would. This is actually highly important for heavy industrial work and, you know, th-this is the reason why we're seeing some pretty great early product market fit in heavy industry as it pertains to civilizational infrastructure, specifically the electric grid. You know, there, there are-

Well, wait a minute. I would certainly agree with you, by the way. Although the humanoids can slot into human society pretty easily and work in factories and drive cars and, you know, build building, work in construction sites, and do all the things that humans already do 'cause they're the right shape, the humanoids are the right shape and size and degrees of freedom, we certainly have all kinds of non-humanoid machines. You know, we got trucks and mining equipment and drilling equipment and all these things that are not human at all, and they-

Yeah

... have their role.

Yeah, I mean, and to the extent of heavy vehicles, I think there are certainly specialized machines. But in, in a certain way, those are also built for, let's call it conducive environments, you know? In the case of the insectoid, you know, locomotion is one very key capability. The second one is bimanual manipulation. So, so most of the robots that have-

Both, both of which humanoid robots have, but let's go on.

Yes. Yes. Ima-imagine a-- Well, let's start here. I think one of the core features that we got super correct from an evolutionary bias standpoint is the ability to do work with our hands, manipulation. And I think this is a feature that must be preserved, and in certain ways is preserved even in insect form factors. So, so here I have with me...

[Sorok pulls out a giant African mantis roughly the size of his hand.]

This is a giant African mantis. This is what you call a raptorial insect. Six legs, but when it comes down to it, it can rear up on its four hind legs and use its forelimbs as manipulators, you know, for interacting with the environment, catching prey, and so on.

With the insectoid, this is actually a core part of, of how we've designed and built these machines, where they're excellent at locomoting in, in places where you'd be hard-pressed to put a human. You know, humans would need scaffolding. Humans would need harnesses. We'd need a lot of additional help. The thing can just climb the-

Not clear how much scaffolding you'd need in, in free, in, in free fall, but okay.

But then when-

Point taken. Point taken. And also, in a way, humans are quite odd because we had to become bipedal like all of our ancestors-

Exactly

... are quadrupedal, and we had to become bipedal to use our hands or it all co-evolved, you know. And so before we were bipedal, we weren't really gonna build stuff with our hands.

So, so you know, what I'll also quickly do for the amusement of the room, I'll call upon my co-conspirator. This is Dr. Damion Shelton, who's in the room with us. He actually founded one of the top humanoid companies in the world today, Agility Robotics. And, you know, he can share his thoughts on, you know, th-this insectoid and humanoid discourse we're currently having. Damion, you there?

Yeah. Hey. So sorry, I told Ulys I was mostly gonna keep quiet and, and listen to him talk about closure, but I can't help myself from arguing impulsively about bipeds. Like, look, I like bipeds. There's a reason why I started a biped company. They don't work well in arbitrary environments because you rely on ankle torque for stability, and ankle torque works well on Earth because gravity is pulling you down. As soon as gravity isn't pulling you down or you get into weird force geometry standpoints, you end up needing a tripod stance.

And if you... And like I put in the chat, if you, if you have a tripod stance and bimanual manipulation, and you're trying to move somewhere, you need the extra leg to do that. And so that's sort of a first principles physics argument for hexapods. Look, for a human environment, the bipedal form factor, I think this is the argument that Creon was making, is probably what you want just because we've already engineered our environment this way. Ulys agrees with that.

And so when you see the sort of the big hand-wavy numbers that Elon throws around or Figure throws around about the giant TAM, then you get that by looking at the human labor market. If you're trying to work in unstructured environments and particularly things where gravity is unreliable, then you end up in a situation where six makes more sense. I'm leg agnostic. I also agree that zero is actually great. Like, zero legs is a really good idea. Forklifts have zero legs. They work pretty well.

So I think where Ulys has a point though, and this is like, why am I, why am I collaborating with Ulys on this? Is I think he has a pretty unique vision with this, and it does have some really interesting characteristics for non-human bounded, non-gravitational environments.

All right.

You know, all to say, I think humans are a force of nature. I think w- we have built the world in such a way to, to conform to us at such a level that it seems almost, you know, what's the word for it? It, it seems almost implausible to, to question that to begin with. But, you know-

Yeah, but there may be places where insectoids are better, just like there's places where forklifts are better. Fair enough. Does anyone else have questions they'd like to bring up?

I wanna hear about von Neumann.

You wanna hear about what?

Yeah. Um, so, so you know, I spent a lot of time thinking about von Neumann probes and, you know, fr- from... What I found most interesting about the field was its ability to kind of bear with the idea that their job was to solve everything, wh- which I found most interesting. But for many structural reasons, it's very difficult to actually propagate von Neumann probes, namely due to kind of some of the earlier points I made earlier, you know, earlier points I made, such as to make a real von Neumann probe civilization scale project work out, this has to be done under the pretense of basically a blank check, you know, unlimited research hours and highly volatile geopolitical worlds, you know, world states that the likes of which we can't really begin to appreciate now, even with w- well, ev- everything that's going on.

You don't find it... There's not a very strong bias for a civilization to take that many of its resources to invest into truly self-replicating technology for galactic colonization. And I also, you know, with Dr. Alex Ellery, under which, who I studied, who I think is the foremost expert on, on self-replicating machines, at least the technical app- application of it, and trying to make them happen, a question I always asked him was, "What do you think the most impressive project in self-replication has been to date? You know, what has really proved the premise of this unbelievably large endeavor?" And his answer was the RepRap project.

And, you know, for folks who don't know, this is the 3D printer that prints itself and can print itself, and so on. I think the numbers sit around, you know, self-claimed numbers by Adrian Bowyer and co. are somewhere between 50% to 80% self-replication capacity. And I thought that this was absolutely ridiculous, and this is, you know, a, an incredibly, let's say, heinous way to look at von Neumann probes of self-replication to any degree, because the way they calculate that number is by total mass. You know, how much total mass can a 3D printer print of itself?

And this is ba- based on the sole input material being like, you know, filament and, you know, what about the electronics? What about the stepper motors? What about, you know, if you could extend a boundary further, what about the human that programs the 3D sli- slicer software and the laptop that enables this 3D printing to occur? Can the child 3D printer construct its own child, and so on. And this kind of brings me to this, like, adage of if a 2,000 pou- kilogram African elephant, you know, releases 50 kilograms of dung, does that make it that percent self-replicating?

And I'd, I'd argue very much no. And, you know, what my belief is as it pertains to von Neumann probes and the practical pathway to von Neumann probes and what the most impressive feat of self-replication thus far has been, it's, in, in my eyes, it certainly has been the optimization of supply chains and the capital engine doing its work over time. I think the greatest contributors of which are folks like Elon and in, in large part, the efforts made by certain industrial Chinese companies, and just this very steady goal of optimizing has yielded, you know, very incredible results for the, you know, grand project of self-replication, believe it or not.

But what I think that all of this oversees and overlooks is if you actually want to get to von Neumann probes, you have to figure out closure. You have to figure out closure because suppose you go to Alpha Centauri, your, and your machine is not 100% closed, it is, you know, maybe 98% closed, and it's, the remaining 2% is some critical component that needs to be Earth, Earth manufactured, per- perhaps some type of compute silicon or core part. You are gonna be very hard-pressed to expand your society, at least at the scale that matters.

And I think if, I think within my lifetime, through very concerted effort, and this is what I intend to do, you can build the necessary technologies and necessary systems that are extremely closure-aligned to then release an irreversible chain reaction of life emerging across the cosmos. But although-

Is it not the case that if you are not just sending out like little micro probes to replicate for the sake of replication, if you're going interstellar-

Sure

... and you send like generation ships, they're gonna be big, and they can have a big bunch of the infrastructure that exists on Earth, not all of it obviously, but-

Yeah, no

... a lot of stuff. A lot... You can bring a lot of stuff, and so closure is a soft thing is what I'm saying. Like, yeah, you can draw your Markov blankets or whatever you want around whatever you want for, you know, theoretical purposes. But when it comes to replicating and propagating, you know, you get to reach as deep into the stack as you want to and as you can.

Well, you know, that, that view, you know, affords the privileged view that, you know, we have time to expend. You know, in, in that sense you kind of-

Yeah, well, look, it takes hundreds and hundreds of years to go to Alpha Centauri, so w- in that sense, we have time. Otherwise, I mean, we better have time.

I... Well, th- that makes assumptions that, you know, I don't think we can really reconcile within this conversation, but-

Well, I know Mike McCulloch was on the line, so he makes assumptions that we have no warp drives, but that's kind of it. But-

Well, but, you know, yeah, i- if we could build w- warp drives and solve faster than light travel, f- you know, that's the ideal state, you know. But as it stands, we have to work with the hand that we're dealt. And in some cases this is generation ships, in some cases this is directed panspermia, you're just sending out the gametes, or you're d- doing some form of cryo-freezing. W- what all of this stands upon is the belief that, you know, you are not the... You are the only party with a vested interest in colonizing the galaxy.

And I think that is a premature and, you know, dangerous position to begin with. You know, although I do concede that, you know, if you had something traveling the speed of the early Voyagers, you know, you, you could... And they were capable of 100% closure and so on, you could probably go colonize the galaxy a billion times over by now. But with the current course of technology and just the way humans and society are, I think you'd be hard-pressed to, A, find your crew, B, build the technology for the generation ship, and C, ensure that long-term project actually plays out at the timescale you're speaking of.

So, you know, ultimately again, coming back to this von Neumann probe idea, you want to build systems with high closure. Whatever method you use to achieve this is, is... needs to be considered quite deeply, and you cannot, uh, have your closure-based system exist within the idea that, quite frankly, you're alone or there is nothing else vying for that prize, which is all of the cosmos. So there's a bias towards speed, obviously.

So, so yeah.

Okay, does anyone else wanna chime in here? We have a bunch of, a bunch of stuff in the chat, which is a little much to, to go through, unfortunately. So who wants to say something?

I have a philosophical question. Given enough time, it feels like humans with our tiny little brains are already independent of our environment. While we can't exist in all environments right now, with sufficient transition time we can adapt to just about any environment. It's like give me 10,000 years and I'll figure out a way to live on Pluto or in space or whatever. Presumably any agent in the universe is gonna need non-zero time to adapt to a new environment.

So if you give some super intelligence a new environment, it maybe takes a millisecond, maybe takes 10 milliseconds. Where do you draw the line between this agent is robust and independent versus this agent is not? Like, how do you decide, you know, taking 10,000 years to adapt to a new environment is non-robust, but taking 10 milliseconds is robust?

Well, just based on e- e- empirical evidence that what we have sent out into space and the cosmos are non-biological systems that are mechanical in nature. And even within our current data set that we have available to us, without generous effort by part of human civilization, accommodating for life and biospheres and so on is going to be incredible investment to make. So in favor of iteration speed, in favor of, you know, non, not speculating on what biological end states could look like for the human species, because that is actually a lot to ponder on.

My, my current conclusion has been allow me to build machines that are maximally useful to current civilization as it exists on Earth today, while understanding as much and as deeply as I can what closure is, and then ultimately release these machines out into the broader world.

All right, does anyone else wanna go from the chat? All right, well, if not-

I have a question.

Oh, okay. Go for it.

It's a little bit off topic, but with the idea of self-replicating robots and sending them out into space, it reminds me of Schrödinger's definition or one of his definitions for life, and I'm not quite getting this right. But basically it's about maintaining homeostasis.

Mm-hmm.

So I haven't seen yet in the descriptions that you've discussed how some of these robots would go to extreme conditions like... I'm look- I'm in Death Valley right now, so I'm seeing some pretty extreme conditions. And maintain something like thermo, uh, dynamic temperature while also having some of the functionality, such as being able to move vertically.

And in Death Valley, you've got rocks and lots of different salts and things that happen, so you have a very quickly changing environment over short periods of time.

So, so I think space is the most extreme version of this. Obviously, you have a huge diversity of environments, whether it be, you know, the edge of a black hole or the inside of a neutron star or some moon on some gas giant or some deep ocean. And yeah, the range of environments is immense. Where I opt to begin with is by looking at how I can push the envelope of robotics as it exists on Earth today, while again, providing terrestrial value that's necessary.

Case in point being our current insectoid. Now we're partnered with, as our first, as our first go to market, we're partnered with the world's largest infrastructure company, and the specific job that these insectoids have to do is essentially repair, maintain, and build large-scale civilization-important infrastructure, namely the electric grid, specifically in the form of transmission towers. And the types of temperature ranges that we have to engineer our robots to build with are certainly not the same ones that most robots today are being designed for withstanding, especially humanoids, but, you know, think of all your warehouse robots and such.

These robots have to perform in, you know, the dead winter with, you know, negative temperatures in, in double digits and also the heat of the summer in, in, in some highly irradiated environments. So the best answer I can procure now is there is a trend line, and we will likely be climbing it better than any other company on the planet, and that's... The hope here is that trend line ultimately follows us to the vast reaches of the cosmos in the extreme environment variations that she has available for us.

What are some concrete steps a person could take today to try to build towards this future that you have envisioned?

Yeah. I think for starters, you know, as a company or at a firm level, start thinking more about what metrics you're optimizing around. Is it necessarily revenue per robot or, you know, headcount growth or things of this nature, or the elimination of certain dependency chains and, you know. People think that vertical integration is, is closure. It's an expression of closure, but it's a brute force approximation of it, which is highly dependent on, you know, certain environment considerations.

So case in point, if you're an engineer, think about one metric that we're actively optimizing for is minimizing number of embedded human labor hours per productive output. Now this sounds quite intuitive and something that people should do right off the bat, but this looks like, you know... Consider a productive output being one megawatt of solar in Africa or, you know, even more simply, one bolt torqued on a transmission tower in, in, in a cold Quebec winter. Embedded labor hours in this case looks like everything that's downstream, which includes the construction of the, uh, of the tower, uh, or the bolt or the solar field, the maintenance of it, the operations of it, and so on.

But also upstream, every human hand that has ever touched a component that goes on this given machine, um, we are-- Our first step is achieving what we call operational closure, which heavily biases hardware and software designs that incentivize things like self-repair, self-deployment, self-maintenance, self-charging, and so on. Basically, you know, removing this linear dependence on human operators per marginal robot.

And, you know, you'd be surprised how many robot companies make robots today where, you know, even if you could make a million robots, you need to... There's a plus million human worker dependency attached to it. So to start off, you know, yeah, question what you're optimizing for and towards, and you'll start seeing a path towards some pretty pragmatic engineering decisions.

So someone posted some-- Let's close with this, if you don't mind. But someone posted in the chat an interesting thing where they said, "If we meet aliens, they're gonna be crabs." I'm wondering what you or anyone thinks about, like, if that's so, if crabs are such a, a kind of, uh, archetypal form in evolution and such a great thing for robots and self-replicating probes, like, why aren't we crabs? Like, why did crabs exist for, you know, a hundred million years in various forms, and now it, it took bipedal primates to make humans?

Because, yeah, I mean, one resolution to the Fermi paradox is that civilizations opted to become more efficient and performant in a singular particular environment, whether that be our biosphere or our solar system, as opposed to being expansion driven. You know, in, in an instance where your civilization is expansion driven, obviously the form factors that you embody cater to, to, to different terminal states. When you are more introspective, as we are as a species today, you know, we're concerned with making intelligence cheaper, concerned with making more of every joule that we have on deck. You know, more interested in, you know, maximizing relative experience per capita. You know-

Yeah, no, we're getting off, but this is interesting.

No, no one's gonna-

My question is very simple. I'm not asking-

No one's gonna argue

... I'm not asking about the Fermi paradox.

Yeah.

I'm wondering, crabs were around for a hundred million years. Why-- And if they're so great and can use their hands, like, why aren't we crabs? You know, it's just-

The simple answer is that humans were made to think, not to work, and to expand, you need to do work. If you want to introspect, thinking is great and excellent. And you know, what that speaks to the future of what humans look like particularly is an interesting question, but-

Oh, I see. So I- if I want to paraphrase, would we say, look, the insectoids won because most of the biomass o- of animals in the earth is insectoids already?

I think that's, uh, that's an observation. I think that, for example, on, on the Kardashev scale, the collective of ants that exist today hold a higher position than the Roman Empire at its peak.

That's sort of what I'm talking about.

Yeah. So-

Look, we have to actually end. We're at our ending point. We try to be pretty rigid about this.

Sure.

So thank you very much. Uh, Lydia, do you have any closing housekeeping or remarks of any sort?

No. I'll be thinking about that all day, though. Why aren't we crabs?

What, what a great meeting. Thank you, everyone, and hope you have a great rest of your day. See you next month.