Conceptual similarities between the work of Esther Thelen and Michael Levin

It seems like motor development and morphogenesis should be very similar fields, as they’re both about the coordinated motion of parts to achieve a goal state in three-dimensional space. Esther Thelen’s work on motor development and Michael Levin’s work on morphogenesis have exactly the kinds of conceptual similarities that you’d expect from two fields that are about conceptually similar phenomena.

For a summary of Esther Thelen’s work, see here. For an introduction to Michael Levin’s work, see here.

Similarity 1: Processes that are regular and reliable, but not preprogrammed

Suppose that every year, the local public school begins teaching typing to children in the fourth grade. An alien scientist studying the development of typing skills in children might make the following observations:

—After years of exhibiting no typing activity whatsoever, children between the ages of 9 and 10 suddenly begin exhibiting lots of typing activity.

—At first, children type in an awkward, uncoordinated, and unskilled way, with lots of typos and errors. But over time, children improve their typing skills on a highly regular schedule, such that you can do a good job of predicting how well a child will be able to type, say, six months after beginning to type.

—Relatedly, children develop typing skills along clear milestones on a consistent timeline. E.g., children may initially rest their fingers randomly along the keyboard when not typing, but children reliably rest their index fingers on the F and J keys at around, say, two months after beginning to type.

—Although typing behavior is inconsistent and erratic and first, children converge on a regular set of typing behaviors: the same figure positioning, the same choice of which finger to strike a given key with, similar speed, similar accuracy, etc. There are irregularities, of course, but on average it’s highly consistent.

An alien scientist, making these observations and following the trends of early 20th century developmental psychologists, might conclude that the development of typing skills in children occurs on a genetically determined preprogrammed schedule. It might be further posited that—as the development of typing skills is clearly just a matter of uploading the typing program from the genes to the brain, which in turn exerts complete control over the body—typing skills emerge entirely as a consequence of the development of the brain. The body’s properties may be considered trivial or negligible in terms of explaining the development of typing skills. (E.g., the alien scientist might observe that humans ostensibly need fingers to type, but conclude that the only contribution of the fingers is to be conduits of the will of the brain.)

Obviously, we know that there is no preprogrammed schedule on which children develop typing skills. But this kind of reasoning led developmental psychologists to believe that there was such a preprogrammed schedule for motor development in children. In this view, motor development occurs basically when and how the genes say so, manifested as a collection of dedicated programs and circuits in the brain for implementing various motor behaviors. These mental instructions were thought to induce the body to behave as compelled, with the body’s own properties having little relevance beyond trivialities like “dense enough not to be carried away by the air”.

Similar observations, interpreted according to various background models and conceptual constraints, led biologists to believe that morphogenesis is genetically preprogrammed, and the necessary motions of the cells composing the organisms are determined by the instructions received from the genes, with the properties of the cells themselves being trivial or negligible.

Esther Thelen challenged the then-prevailing consensus on motor development. She showed that motor development doesn’t occur on a predetermined schedule but instead as an interplay between the child and the environment. The child’s wants and abilities interact with their environment to produce motor behavior. Motor development occurs as the body grows and strengthens, the brain’s desires expand, and the brain’s regulatory, coordinative, and conceptual capabilities increase. The transition from crawling to walking isn’t set in stone, but rather is the transition from an efficient mode of locomotion under one set of circumstances to an efficient mode of locomotion under another set of circumstances—a transition the child manages as an optimizing agent figuring out how best to move their body in their environment, not as a consequence of preset instructions. Today, thanks to passive dynamic walkers, we know that human-like walking does not require a brain, but instead can be produced as an interaction between a body and an environment.

Similarly, Michael Levin has shown that morphogenesis is also not reducible to genetic instructions for how to construct an organism. In fact, no such genetic instructions exist. Instead, a model in which each cell optimizes with respect to its environment—an environment often consistent of other cells—leads to morphogenesis as a natural result. This is similar to Thelen’s argument about how motor development is the natural result of optimization by the infant within their environment.

Similarity 2: Unexpected competencies discovered by probing

Esther Thelen’s path to creating a new theory of motor development in infants began with a scientific anomaly: the infant stepping reflex. This is a reflex where a baby, when held upright, kicks their legs back and forth in a manner that resembles stepping. This motor behavior disappears around two months of age, then comes back shortly before the infant is ready to begin walking.

The infant stepping reflex may not sound important, but it was a puzzle for the then-dominant theory of motor development, in which all motor development was presumed to be entirely a function of brain development. Moreover, this brain development was presumed to occur linearly, adding new competencies to existing ones. The brain should strictly increase in ability over time, and so the infant should exhibit increasingly more and complex motor behaviors over time. A motor behavior like the infant stepping reflex that disappears and then comes back does not fit this pattern.

Thelen investigated this behavior and found something that others had missed. When the infant is laid on their back, they exhibit a kicking behavior very similar to the infant stepping reflex even after that reflex was supposed to have disappeared. Thelen hypothesized that the change was not about the brain but the body. Specifically, it’s easier to kick your legs back and forth when laid on your back! (Try it and see.)

Thelen’s hypothesis was simple. Babies gain weight faster than they gain strength. The thick rolls of fat that babies quickly develop around their legs make it difficult for them to kick their feet when suspended in the air. When placed on their backs, it becomes easier, and so they kick more.

Thelen confirmed this hypothesis by submerging infants in water. Upright, but with water lightening the load on their legs, the infant stepping reflex emerged once more.

Thelen then experimented on infants (safely) secured to treadmills. These infants were supposedly too young to walk. But by providing some extra stability and impetus, she exhibited their unexpected ability to walk. Just because infants had never previously exhibited the ability to walk at that age doesn’t mean they couldn’t. You had to do the test to find out.

Michael Levin has found a number of unexpected competencies through testing. Trachea cells can heal scratched neurons, sorting algorithms can solve problems they weren’t coded for, and skin cells can self-organize in ways that evolution didn’t prepare them to do.

Bodies, cells, and even sorting algorithms all have capabilities beyond what are usually displayed. Empirical testing, not theoretical presuming, is necessary to determine what competencies a physical system can display.

Similarity 3: Discovering—or creating—diverse forms of cognition

The A-not-B error is a classic observation in child psychology. The experiment goes like this. You give an infant between the ages of 8 and 12 months old a toy to play with. Once the child is interested in the toy, you take the toy away and hide it under box A, which is next to box B, both within reach of the child. The boxes are opaque, but the infant can see where you put the toy. Naturally, they search under box A and find it. Then you take the toy again and this time hide it under box B. There are no tricks here; the infant can see exactly what you’re doing. Strangely, the infant may search under box A again for the toy, even though they saw you place it under box B. Because the infant searches at A, not B, this is the A-not-B error.

The A-not-B error has been replicated many times. The question is, why do infants make this error? Typical explanations have been focused entirely on the brain and its mental qualities. Jean Piaget, the originator of the experiment, hypothesized that the error is due to an egocentric bias resulting in a lack of object permanence. Other hypotheses centered around some kind of memory issue. Both explanations were falsified by experiments showing that the error is still made when the boxes are transparent. Even when the infant can see exactly where the toy is, they reach to the other box anyway.

Esther Thelen developed a different explanation of the A-not-B error based on dynamic systems theory. She didn’t focus her investigation on questions about the beliefs of the infant. Instead, she framed the A-not-B error as a problem of reaching. That is, the empirical issue to focus on is the question of why the infant reaches toward one object and not toward another object. The A-not-B error is an error in which the infant reaches toward a prior goal location after the goal location has changed.

Instead of presuming that the task of motor behavior is as simple as the brain selecting some kind of given plan from a predetermined list, Thelen observed that a successful reach to the goal location is the confluence of many continuous and complex processes: seeing, memory, setting parameters, initiating a trajectory, monitoring and correcting errors, etc. These processes do not play out in a linear sequence but coevolve and influence each other.

Thelen’s explanation of the A-not-B error, put very simply, is that motor behavior is complicated, and unskilled infants will err when the task is complicated. Notably, infants at five months old reach correctly, but err at seven or eight months, around the same time that they begin taking on much more complex motor behavior and ambitious locomotion efforts. This suggests that the error could be replicated in older infants with sufficient manipulation of their environment, which Thelen and her coauthors later confirmed.

In my view, Thelen’s findings on the A-not-B error show us what’s happening “under the hood” of human cognition. Instead of a homunculus inside of us controlling everything, there’s a complex, dynamic system more akin to an economy. Human cognition is less about semantically identifiable beliefs like “the toy I want is over there” and more about coordinating and regulating a number of complex, continuous, coevolving processes that are the product of the interactions of a tremendous number of subunits, such as cells, tissues, and organs.

Instead of finding the alien mind inside of human beings, Michael Levin created some aliens of his own. Xenobots and anthrobots are artificially created biological robots that demonstrate unplanned and unwritten forms of cognition.

Cognition does not look like how humans think it it looks—even in humans!