A Slinky can hardly be considered a sentient being. Its surprisingly ordered walk down a flight of stairs is purely a marvel of mechanics.
When a person descends a staircase, despite solving roughly the same problem under the same pull of gravity, she owes her success to cognitive processing in the brain…or so we assume.
Not so fast, say researchers at SFI who are working on a theory of embodied intelligence — that is, intelligence that arises from the interplay of brain, body, and environment. In the human case, some computation is outsourced to the body.
Biology provides us with many examples of problem solving that doesn’t occur, strictly speaking, in the brain, says Postdoctoral Fellow Keyan Ghazi-Zahedi, who with Professor Nihat Ay is organizing a September working group on “Morphological Computation.”
Take the problem of carefully grasping a delicate object such as a dried flower. For a robot, this is a computationally costly problem. Its processors need to compute the locations of its grippers precisely: too little force and the flower slips away; too much and the bud is reduced to dust.
With a human hand, the brain isn’t required to do as much computational heavy lifting. Soft tissue in the fingertips, feedback in the finger joints, and tendon friction help the brain solve the problem of gripping with care. This is morphological computation, says Ghazi-Zahedi.
“The traditional view [in robotics] is that all you need is a big computer and then an embodied agent can learn anything,” he says. “We’re learning that the body is not a burden to cognition. We’re starting to understand some of the ways the body, as our interface with the environment, actually contributes to cognition.”
Despite morphological computation becoming a widely accepted concept in the short time since its introduction in 2006, it is not at all clear which kinds of morphological processes are computation and which should be understood as pure mechanics.
In recent years, researchers have tried to parse morphological phenomena into categories as a way of determining what is and what isn’t computation.
“We believe this is the wrong approach,” says Ghazi-Zahedi. “We should be looking for a unifying perspective in the many ways the body contributes to cognition.”
This is the goal of the SFI meeting, which brings together a dozen or so experts from robotics, mathematics, philosophy, engineering, biology, and physics.
Ultimately, Ghazi-Zahedi says, such discussions might open the door to a formal treatment of the phenomenon, even quantification of cognitive contributions in biological systems — a goal of his own research. It also might help reduce the burden on programmers and designers trying to create artificially intelligent robots.