Motor behavior is an example of an unconventional intelligence
Unconventional intelligences are intelligences that don’t look the way we typically expect intelligences to look. They’re often intelligences that are not brain-based and possibly not even alive at all, challenging our ordinary conceptions of what intelligence is and where we should expect to find it.
An example of an unconventional form of cognition is motor behavior. For a long time, the science of motor behavior assumed that motor behavior follows more or less automatically from instructions provided by the brain. Basically, the brain says, “Right arm: reach forward,” and the right arm reaches forward. The task of reaching was assumed to be basically trivial or somehow automatically taken care of once the neural instruction is given.
This idea was challenged by the degrees of freedom problem, associated with Soviet neurophysiologist Nikolai Bernstein. The degrees of freedom problem means that motor behavior is made out of many parts: hundreds or thousands of joints, muscles, body segments, etc. The combinatorial possibilities are very large, but only a small subset of the possibilities give you the motor behavior you want. So how does motor behavior happen?
There are way too many possibilities for the brain to simply compute all of them and choose the best one. Instead, motor behavior relies on the problem-solving abilities of the components. This means that the components will seek goal states, and navigate to those goal states despite obstacles and perturbations. The brain is then freed up to regulate the components at a high level by adjusting their energy landscape to incentivize them to produce one behavior rather than another.
An example of components of motor behavior that seek goal states without neural instructions are the legs. The legs naturally swing back and forth because of their own physical properties. This is shown by passive dynamic walkers, which are physical structures built to walk without a brain. Despite the lack of a brain or anything analogous to neural instructions, they still exhibit human-like walking behavior because the legs have certain natural tendencies of their own.
Basically, the legs have two goal states, or attractor states in dynamic systems theory, which involve being stable against the ground without experience too much of a stretch. As one leg steps forward, the stretch in the other leg pulls it away from one goal state, whereupon it seeks the other goal state by swinging forward and finding the ground. The brain helps to regulate and direct this process, but it does not instruct or otherwise micromanage it. The legs themselves do problem-solving, or cognition.
An example of the problem-solving involved in walking is babies walking with diapers. Diapers interfere with the natural swing of the legs, constituting a constant perturbation as infants try to walk while wearing a diaper. Obviously, there is no genetic program for walking with a diaper, nor can the brain figure out a priori how the legs should adjust. The ability of infants to walk despite the presence of a diaper shows that motor behavior can reach goal states despite perturbations, which is intelligence.
Morphogenesis is the classic example of unconventional intelligence, and the associated models take inspiration from Alan Turing’s analysis of pattern formation in chemical reactions. The dynamic systems analysis of motor behavior similarly takes inspiration from the Belousov-Zhabotinsky reaction. This reaction is famous for oscillating between attractor states rather than proceeding linearly toward equilibrium. The oscillating behavior resembles walking, where the body alternates between postures (one leg forward and one leg back) rather than increasingly adopting a single final posture.
The fact that motor behavior navigates toward goal states despite perturbations and without neural instructions shows that motor behavior is an example of an unconventional intelligence.
Yes, preflexes! Not only there are just waaay too many DoF, nerve impulses are also as slow as molasses, the brain is literally too far away for signals to arrive in a reasonable amount of time, even with near instantaneous processing. So evolution handles it by reducing the scope of the problem, offloading the burden to the material substrate of the actuators themselves, visco-elastic properties of the musculoskeletal system offering a zero-delay intrinsic feedback for auto-stabilization. In other words, it picks a better *format* wherein to engage the problem. I don't know what's the current state of actuators in robotics, but I hope they draw inspiration from stuff like this, imagine how terrible it would be to have to try to walk around with stiff, numb and dull limbs.
Information geometry shows us that the replicator equation has characteristics of an inference dynamic, and physical learning is an emergent field demonstrating that rather large classes of physical systems are capable of primitive pattern recognition and learning through local adaptation. I even know of one group in Japan treating neural nets as models of spacetime, so who knows where this will all end up.
This shouldn't be so shocking, human presumption aside. It's not very hard for a language to be Turing-complete, for symmetry to emerge or even for self-reproducing patterns to form in a random soup of interacting codes. Why should intelligence be restricted to living things, much less brains? Personally, I like the definition from Causal Entropic Forces of intelligence as a mathematical property of a physical system, its capacity to maximize its future freedom of action.
Great Substack, by the way! Your paper with Dr. Levin on cognitive glues was extremely interesting.