“Warjacks can’t really perform some of the simplest mathematical equations done by a calculator, yet can intuitively grasp the physics of the world by similar instincts as we use when catching a ball or throwing one. They are most certainly not adding machines”
Embodied cognition is one of the most interesting developments in 21st century psychology. And from reading the above words from Doug Seacat, I wonder if he’s read any Gibson (not the one of Neuromancer fame). But if not, it’s just an interesting coincidence that got me interested enough in warjack cognition to run a game that centres around it. The basic ideas underlying REC (Radical Embodied Cognition) are 1) Perception and Action are coupled together forming a single system 2) No internal representations are neccesary for cognition 3) We perceive the world in terms of affordances, in terms of opportunities for action.
Traditional cogntive psychology holds that our brains are constantly running an internally represented simulation of the world that allows us how to predict how it will behave, and thus guide our action towards it. REC rejects this formulation, saying that if our perceptual apparatus and behaviour is coupled to the right kind of information to solve the task, then we don’t need a complex physics simulator in our brains.
There are two excellent examples of this approach that are wonderfully relevant to Warjacks. One is The Outfielder Problem which is sorta alluded to in Doug’s words above – how do we go about catching a ball? The second is the Watt centrifugal governer, which is pleasantly relevant since it was invented in the industrial revolution.
In the outfielder problem, our acting organism is an outfielder in baseball, tasked with catching a baseball flying off the hitter’s bat. The traditional cognitive science explanation for this behaviour is a computational one – that we have complex symbolic representations of the world in our minds which run a physics simulator. This simulator takes information about the initial velocity and initial angle of the baseball, assumes constant ball size and shape, assumes constant gravity and air drag (since we can’t perceive those things), and calculates a prediction for the final landing place of the ball, and directs him to run to that point.
Perception-> Computation -> Solution -> Action.
The REC solution is quite different. It suggests that our perception and action systems are coupled with the appropriate information needed to solve the task without internal, symbolic computation being necessary. The outfielder can solve this task by changing his movement (running) such that the ball appears to be moving at a constant velocity relative to him, and/or such that the ball appears to be moving in a straight line relative to him. This type of movement will bring the outfielder to the right place to catch without ever needing to predict anything. This type of solution is active and dynamic, and has information that allows Outfielder to make adjustments to his movement as he goes – if the ball appears to be moving faster than him, speed up, if it appears to be arcing, change the curvature of your movement – and these adjustments only require perceiving information that is easily perceived by the optical array, as opposed to requiring a complex physics simulation that makes a lot of assumptions. Warjacks, like humans, are not calculating machines with regard to phsyics. They “instinctively” grasp the physics of the world, but not because we have a simulator already installed. It’s because the simplest solution doesn’t need a simulator, but merely a perception-action system (considered as one whole thing) that is coupled to the right information.
The second example is the flywheel problem, which makes it a little clearer why computation and representation aren’t required to solve a problem that look like it would need computation.
A major engineering problem in the 18th century involved figuring out how to power a flywheel with pistons, which would allow engineers to generate rotative motion. In particular, this was a big deal in the textile industry, but it also opens up a wide range of other machines that would require such motion. The problem was keeping the rotation speed of the flywheel constant with the changing steam pressure and engine workload.
You could install a throttle valve that would allow someone (or a computation machine) to control the steam pressure, but that controlling agent would need to make very precise adjustments at just the right time to keep speed constant.
The “computational” solution would be to allow something/someone to measure the state of the system at various points in time, perform the appropriate calculations, and make adjustments. Of course, there would be lag in this system, so the system would always be a little behind. There would be constant error, however small. Also, this system requires an executive operator, either a person or an AI running the valve, making calculations, and making adjustments. This is basically how standard cognitive science sees the human mind (and clearly not how Doug Seacat sees the warjack mind, as they do not have that kind of calculation ability).
The dynamic (REC style) solution is to couple the opening of the valve to something whose physical properties respond to the steam pressure and engine workload in a particular way. This is the solution that was implemented for the flywheel engineering problem in the form of the Watt centrifugal governer. This was achieved by attaching weighted arms to a flywheel driven spindle and linking the valve control to those arms. The valve control is coupled to the flywheel such that as spin speed increases, the valve closes, reducing pressure, and thus reducing speed, which causes the valve to open more, thus increasing speed, and so on and so on.
If you didn’t know how it worked, you might think there was a computer controlling those arms and making adjustments. But the physical nature of the system couples the relevant variables together so that adjustments simply happen as needed to solve the task. Obviously, this analogy is slightly oversimplified for animal cognition, but that doesn’t mean that the same principles cannot apply. (The reason REC is really exciting: It gives us loads of research to do to see how far this kind of solution can take us.)
My thesis here is that warjacks “think” in exactly this kind of way. It’s still very arguable that this explanation won’t go all the way for humans, but I think it definitely can for warjacks (and animals). So, what then, are the implications for warjack behaviour?
I’m going to go into more detail on this in the next article, but for now, some suggestions
1) It means that warjacks perceive the world in terms of what their bodies can interact with (affordances need a whole post I think) and how.
2) Think of warjacks as a body/enviroment interface which assembles itself into task solutions. Or, in simple terms, warjacks are perfectly capable of manipulating their environment to solve tasks, by “coupling” themselves to parts of it.
3) Warjack minds don’t have “representations” in them. They don’t have built in physics simulators, so that’s not something you can screw with, reprogram, or break. But you can mess with their ability to perceive relevant and useful information.
4) There may be certain tasks that seem to complex for warjacks, but even very complicated tasks are solvable without the need for computation.