Cognitive psychologist L. Elizabeth Crawford wondered if she could teach rats to drive. It was a question that arose out of a couple of ingredients. The first was her professional interest in the way our bodily locomotion through the environment affects such capacities as spatial memory. The second was more fortuitous: she had a sabbatical year during which she got into tinkering with electronics, combining sensors and actuators to do weird things just for the hell of it. (A keeper of chickens in a predator-rich neighborhood, she rigged up her chicken coop to detect when a hen had laid an egg, and to then send out a tweet consisting, appropriately enough, of the single word “tweet!”) Around this time her husband was starting work on a book about driving, and the two of them had conversations that ranged from “embodied cognition” to animal intelligence; from human-machine interface and the peculiar pleasures of driving a car to the sometimes mentally enervating effects of automation. The connections between bodily skills and other forms of intelligence suggested fertile questions about culture and technology.
So. Could rats be taught to drive?
Crawford didn’t know much about rats but Kelly Lambert, her colleague in the psychology department at the University of Richmond, did, so the two of them teamed up. According to Lambert, we have no idea what rats are capable of because rat research has always been conducted in a highly controlled lab environment, the point of which is to study the effect of one variable and eliminate confounding variables as much as possible. But it is precisely the richness of an animal’s environment that calls forth its evolved capacities, many of which depend on a process of development (during the rat’s own lifetime) in a natural setting no less than the selection pressures that have shaped the species over generations. She told me researchers have never investigated the question of skills with their furry subjects, as opposed to their ability to do simple, discrete tasks such as nosing a button to release a treat in response to some stimulus. Yet they are known to be very intelligent animals. Crawford wondered: could they be taught to find their way through the world by making use of a truly alien mode of locomotion? This would require learning a whole new set of “motor skills” and integrating these skills with their map of the world in a new way, successfully enough to get where they want to go. To get a Fruit Loop, as in previous experiments in Lambert’s lab. But to do it in style, with a nice set of wheels.
Let us digress for just a moment to note a classic experiment that explored the role of movement in perception, and specifically the role of self-motion, as opposed to being transported around passively. Ten pairs of kittens were reared in the dark, except for three hours per day that each pair spent in a carousel apparatus that allowed one of them to move freely, while the other was carried passively by the movements of the first. The active kitten could move up, down, away from, or toward the center of the carousel, as well as rotate in epicycles at the periphery of the carousel’s radius. The kittens could not see each other, and the surrounding environment was contrived so that both kittens received identical visual stimulation as they moved around; the only difference was that one moved itself, the other was passively carried. The active kittens developed normally; the passive kittens failed to develop visually guided paw placement, avoidance of a visual cliff, a blink response to quickly approaching objects, or visual pursuit of a moving object.
These finding were the tip of an iceberg that became “embodied cognition,” now a prominent research program in psychology. As Alva Noë puts it, “When we perceive, we perceive in an idiom of possibilities for movement.” Further, our perception of these possibilities depends on what kind of tools we use to get around, and a corresponding skill set.1
Crawford’s first order of business was to build a rat car, in what historians will surely regard as the pioneering attempt at “rat-machine interface” design (how this differs from getting a rat to simply press a button, we will get to shortly). She went to Radio Shack and got a cheap radio-controlled car, cannibalized the chassis and motors, then made an enclosure consisting of a transparent cashew container (the really giant ones you get from Costco) with some windows cut out of it to allow the rat olfactory connection to the environment. She made a control mechanism using the Arduino platform favored by tinkerers. I am proud to say I had the honor of welding up a rat-friendly joystick housing for an early prototype. It turns out rats don’t like joysticks. (Or at least, Mario and Luigi didn’t like my joystick.) They prefer separate left, right, and straight-ahead controls. In the current iteration of the rat rod (as I insist on calling it), Crawford is using conductive bars that the rats grab with their little mitts, closing a circuit.2
It took about a year of prototype development for Crawford and Lambert to get the basic rat ergonomics worked out, and to fine-tune some other elements of the experimental design. For example, because rats use smell to find their way as much as sight, and Fruit Loops don’t have much aroma, various smelly substances were put adjacent to the Fruit Loop. As it turns out, a wet tea bag is just the ticket.
More significant, this research required a leap of faith to embark on a long-term program of rat driver’s ed. Nothing remotely like this had been tried. There is a hundred years of research based on teaching rats to do various things by inducing a “conditioned response.” Essentially, the rat is treated as a stimulus-response machine that can be trained to display a certain range of behaviors.
Crawford, alive with an infectious geeky charisma, stood in her kitchen one morning in April 2019 and explained to me how this project differs. The initial task that the rats needed to learn was simply to press on a bar to make the car drive straight ahead to where a Fruit Loop awaited them. Their success at this, which came quickly, essentially replicated the kind of training that lab rats have long been subject to. This early training served to familiarize the rats with the car, get them comfortable with the setting, and acquaint them with particular people—generally undergrads in white lab coats.
The next task the rats needed to learn was to press on a bar located to their right to make the car simply pivot to the right, where a Fruit Loop was to be had. This task too was easily learned, and in line with the kind of tasks traditionally demanded of lab rats. (The rats were never taught to turn left, and the significance of this will become clear shortly.) These basic competencies became the scaffold on which the next stage of training would be built.
The car was placed at the far end of the driving arena from the Fruit Loop, and pointed in the wrong direction. After a protracted period of trial and error lasting months, in the course of which the researchers permitted the rats to run into walls, get stuck and generally frustrated, something remarkable happened. Having taught themselves to turn left, the rats began to find their way, navigating to the Fruit Loop in a zigzagging trajectory, overshooting with their steering inputs and forward movements and then correcting course. Crawford explained to me how this task is different in kind from the previous ones. First, it is not simply more difficult, in the sense of requiring some new dexterity of an unaccustomed kind. What the final version of the driving task does is expand the problem space that the rat needs to solve. In fact, there is literally an infinite number of trajectories that could be taken to get from the starting point and initial orientation of the car to the location and orientation that allows the rat to get the Fruit Loop. After each zig, the rat needs to zag in response to the new situation. This is just what we do on a much finer temporal and spatial scale on foot (the subtle course correction that all animals engage in subconsciously). This open problem space resembles that of an animal’s natural environment, as opposed to a laboratory setting contrived to elicit a particular behavior. But unlike the rat’s usual reliance on its own bodily equipment for solving the problem, here its intention has to pass through a strange machine.
To watch the researchers’ videos of rats unquestionably driving, with increasing fluidity, is astonishing.3 Crawford believes it is the first instance of genuine tool use by rats, if by that we mean the flexible use of an instrument in response to some evolving situation, in a feedback loop of perception and action that crafts the ongoing development of the situation toward some goal of the agent. That is, skill. The car becomes a kind of prosthetic, an extension of the rat’s body, in the same way that the limbs of a toddler, initially alien and awkward, may be understood as prosthetics of the brain that gradually get integrated with the brain in the course of development. Thus does an embodied being become competent, in ways specific to its environmental niche. Our limbs and hands and, later, various tools that we become skilled in using no longer feel like prosthetics. They fade into the background and become transparent, that is, unobtrusive conduits for both action and perception.
I believe some clues for the significance of the Crawford-Lambert rat driving project for human culture, including driving, may be found in earlier work by Lambert. In experiments on both rats and people, she explored what she calls “effort-driven rewards.”4 Lambert found that “movement—and especially hand movements that lead to desired outcomes—plays a key role in both preventing the onset of and building resilience against depression and other emotional disorders. Furthermore, we are predisposed to preferring hand movements that our ancestors needed for survival—those necessary for nurturing, cleaning, cooking, grooming, building shelter and farming.” Lambert theorizes that the enormous increase in rates of anxiety and depression over the last few decades may be due in part to our disengagement from the basic tasks of securing our own bodily needs, and “all the complexity of movement and thought processes” such tasks require of us.
“The decreased brain activation associated with increasingly effortless-driven rewards may, over time, diminish your perception of control over your environment and increase your vulnerability to mental illnesses such as depression . . . . Anything that lets us see a clear connection between effort and consequence—and that helps us feel in control of a challenging situation—is a kind of mental vitamin that helps build resilience and provides a buffer against depression.” Closely connected to her work on effort-driven rewards, Lambert found that rats who enjoyed an “enriched environment” more closely resembling the natural world, with its problems to solve, were more persistent in solving problems and less prone to get stressed out, compared to rats kept in standard lab enclosures. In the rat driving study, Crawford and Lambert found that rats raised in an enriched environment learned to drive more readily, and that rats who drove themselves had a stress hormone response different from that of the rats who were passively driven. It is a difference that is associated with lower anxiety in humans.
As I see it, this work has a clear upshot for human beings. As we grapple with the challenges of automation, we may want to arrange our own environment like that of the happy rats, rather than the overdetermined world of their anxious counterparts. Of course, we do not simply live in the natural environment. But the built environment of technology and cultural practices can likewise be rich enough that it demands the use of our full repertoire of intelligence. Flourishing—that of rats and humans alike—seems to require an environment with “open problem spaces” that elicit the kinds of bodily and mental engagement bequeathed us by evolution and cultural development. These exquisitely honed human capacities include the glorious, century-long development of the automobile, that astonishing tool, and the social intelligence that we have brought to bear on the problem of sharing the road together. If instead we put ourselves in a Plexiglas enclosure in which all our most basic needs are met, we will have nobody but ourselves to blame if we begin to feel like standard lab rats in a massive laboratory of social engineering. We would be safer that way, no doubt. But remember, all rats die. Not every rat lives.