 Hey everybody, it's 2024 here in the solar system. I've got a lot to talk about. Let's get started. Although there were some funding sources that seemed quite interested in the idea of a global phase war, the scientists and engineers thought there really ought to be a less wasteful way to build BV's brain. Rather than trying to figure out a single global phase for everything, one team decided to just develop lots of small little diamonds, each one that could be associated with a single sensor or a single motor, and they got that working and then they didn't even really care about the phases. They could move over each other and in and out of each other and took care of each other fine. Of course, there was one question about how are they going to accomplish the communication between those little diamonds, which was the purpose of the global phase to begin with. So that stumped everybody for a while, but then one of the senior scientists remembered there used to be a thing called a swap worm that didn't need a phase or a grid, it just moved around in free space, it could grow longer, and maybe they could take the head of a swap worm and stick it inside a sensor diamond, and the tail of a swap worm and stick it inside the corresponding motor diamond, and then it wouldn't matter where the diamonds were actually in space because the worm would conduct the signals from one to the other. So that sounded like a good idea, there was lots of patents filing and a flurry of implementation, and at first these new free worms looked really good. They could connect a pair of local diamonds and if the local diamonds moved, the free worm moved with it. But the problem was when there were multiple free worms connecting pairs of diamonds, then things got nasty. The free worms are very flexible so they would route around each other, creating tangles and knots and bigger tangles, and the engineers just couldn't figure out how to get rid of it. They tried and tried, but nothing worked. And since the monthly report out to the funders was only a couple days away, tensions were running high. But finally one of the interns said, well if we can't do a global phase because of phase war, but little teeny diamonds are too small to communicate within, why don't we do one big diamond? If there was, ended up being two diamonds, well then we would just let them move in and out of each other. We'd expect that there was only one, maybe we could try to get them to separate with each other, but we don't have to solve that problem ahead of time. Why don't we give peace a chance? And everybody looked at each other and said, could we do that? Is that possible? And the summer interns said, well I've already implemented a lot of it and it seems like it could work. And at that point the big cheese with an eye on the calendar said, make it so. So we will have part two of BV's adventure later in this episode. Here's what we were supposed to have done for today, mixed bag as usual. BV's distributed oolong brain running on the T2 matrix, changing its mind using subsumption. Well, number one, it's still not on the T2 matrix, it's running on MFMS, the simulator. And I'll talk a little bit about a couple other asterisks that come up there. So that's definitely a plus or a minus. On the other hand, funnest fundamental, I enjoyed this. Getting to the point of maybe can BV actually change its mind in a way that is non-trivial, in a way that kind of means something, is something that we've been aiming at for, well, I mean for all of this six month run for starters, but even much longer than that in some sense. So there was a lot of struggling, stuff that didn't work, like the free worm Alzheimer tangles or whatever the heck that was. But still a lot of fun. Okay, so the schedule, just to quickly remind us, we are down to the next to the last one. May 7th, the next update is the final one of the T2 matrix brain challenge. None of it has been on the T2 matrix. Fix that. The goal this time was to get the sensory motor homunculus, which we don't even exactly know what that is. Although I think we're getting closer. Even though at this point, the Breitenberg vehicle heritage is clearly still there and will be there all the way through. It's not like we're now going to switch and have a, you know, data the Android robot that looks an awful lot like a person. So we're probably going to push the stretch goal of growth and healing. Although I mean, we have some of that already just because we're doing bottom up, we're doing robust first. But so demo of the sensory motor homunculus will again be coming around next time. So, oh yes, subsumption architecture. So I'll circle back around to say why there's a question mark there now that there wasn't in the last update. But let's find out what happens with BV. So the plan was to lay out BV's brain inside one big diamond. And there was plenty of room to spread out the sensors and the motor terminals that were required. In addition to the left and right yellow sensors that BV had always had, the engineers added left and right red sensors as well to make more complexity possible. But they all were connecting to just two motors, the left and right motor. And the Breitenberg vehicle 2B format had the left yellow and right yellow cross connected to the left and right motors. It should work just the same as it had before. But there was a problem. Now BV had always been exuberant, but with this new brain, it was like BV was on drugs. BV was very sloppy, switching balls, spinning around, missing things. And it didn't take too much analysis to figure out what was going wrong. The signal times between the sensory terminals in BV's brain and the motor terminals where the signals came back was taking too long. And as a result, the motor commands were outdated by the time they reached the motor terminals. It didn't look very good. Displaying the simulated tiles in the MFMS simulator made the problem more apparent. With spreading everything out for the big diamond, some of the signals had to cross three tiles to get from the sensory input to the motor output. It was taking too long. BV's brain needed to be more compact. Through a couple of iterations, eventually the engineers put all of BV's sensors and motors on a single tile. And then it worked like it had before. But if anybody was curious in what sense this was a distributed computation, nobody said anything at the time. So now the plan to demonstrate BV changing its mind was to let it push balls around but not jump into the void itself. And to do that, they used the red sensors which they added and they added a big red warning track around the edge of the world that BV could be sensitive to. But in order to do that, it would need more controls than it had before. And in particular, it needed a new terminal. It was like a sensor to take input like a motor. It could generate output, but also it could retain information about what it was doing. It needed state. The engineers called it a grip. The idea was each grip could determine the routing between sensors and motors. Instead of having that be hardwired, if you change the grip, you can rewire the brain. So the BV2B was just one grip. And the engineers wanted to create another grip that would help BV avoid jumping into the void by looking out for those red signals at the edge of the universe. And it took a while, but they got it working. So that's it. So BV changes its mind from pursuing yellow balls to stopping on red. Seems alright. A couple of things that I just want to add to that that I fluffed over in the BV's adventure, in the BV's demo. So I keep thinking about what I want to call these cartoons, but they're really not cartoons because they're demos. I can't actually make BV do anything particular. I can't have him run off a cliff, but not fall unless that's what the physics does. These are demos of what the code actually does. But subsumption architecture distributed computation. So, you know, let's talk about the subsumption architecture just for a minute here. So the idea of the subsumption architecture, which is also the idea of the Breitenberg vehicles generally, was that, you know, you'd have a bunch of wiring that would cause the robot, the agent, the creature, the living thing, whatever it is, that has a stake in the universe that has a physical thing that it's capable of operating and sensing through. And that it would have a default behavior and then it would have other more complex behaviors that were built on top of the default behavior. And they would come in and subsume it and say, don't do the default behavior, do this instead. And the point of it, the key of the subsumption is that the default behavior hardware is all still there. And then there's other stuff that comes in and subsumes it if it wants to. And otherwise the lower level behavior just keeps popping up automatically. And working the way that I did this demo that we just saw, the second one, here comes Mr. State, is not that. Because the changing of the grip actually rewires the brain. The yellow sensors that used to be going to the motors now go up to the grip to let it know if there's an opportunity to go chase a ball. And the red sensors which had been going to the grip now actually go to the motors and so on. And I went back and forth about whether I really wanted to do it that way. But the point is, is that I don't actually want the wiring to be fixed. I want the wiring to be programmable so that we have all of this artillery, the software, this grid that we're building up. Which we can now route to any of the terminals, basically all of the motors, the things that need input. They all have gradients so that anywhere in the grid you can figure out how to head towards any of the motors or the grip. The grip counts as a motor. It's a motor and a sensor. So it's modifiable, persistent state inside the creature, inside the agent. And as I've been working through this, it really feels like that is kind of the essence of mind. The essence of mind is it's a decision that you can make, number one, and you can change. Number two, persistent, modifiable state is the essence of mind, I think. Or at least that's what I'm thinking this month and I've been enjoying thinking it. It feels like it's got a little piece, a little germ of truth for sure. So that's the one question, that's the point on the subsumption, the point on the distributed architecture. This is pretty embarrassing, right? The entire sensory motor homunculus is on one tile. Why on earth would I even bother porting it to the T2 matrix when I could just have it sit on the desk? And, you know, on the one hand, I feel like, you know, it's pretty embarrassing. But on the other hand, I feel like, no, actually this is right. Because the point is the direct connections between the input and output like the spinal cord, the kicking the knee up when you hit it with the hammer and all that stuff, that wants to be as fast as we can do it. And of course in the T2, nothing is fat. But in the T2, nothing is fast, but the point is it's as fast as we can get. So our reflex arcs are as short as we can get. And, you know, the previous stuff was using one big tile, and that's why it didn't run into the over control delay lag problem that we saw today. But what are we going to do with the rest of the matrix? And I think the design is, is that, well, no, this is kind of right. And it's kind of a symmetrical sort of thing that we have little BB in the physical world, all the world around it. And inside the brain, there's going to be the sensory motor homunculus with all the brain around it, with all of BB's mind about it. So the important that the reflexes, you make them close together. But the stuff that can afford to take a little bit more time, you surround that. And what it does is it comes up with computations about what should the next grip be. So that feels good. And we'll see how it all goes. So next time, the final, the final step of the T2 matrix brain challenge is, you know, get the super fun mind changing demos running on the T2 majors. I don't care if it's only one tile, it would be nicer if it's maybe a little cluster if we can figure out some way to work around it on the outside. And once again, fun is fundamental. I got to keep reminding myself of that. But a lot of it is like, you know, I really enjoyed doing this. I just keep forgetting to talk about it. Okay, that's it. Thanks so much. I hope I see you next time.