The quest to find intelligent behavior in animals has been a long one, and it has unraveled some rather spectacular examples. Witness, just to mention a few, crows creating their own tools, sea slugs that learn through conditioning, and magpies who can recognise their own reflection in a mirror.
How complex does an animal have to be in order to show signs of intelligence? Not very, according to Toshiyuki Nagakaki and his co-workers, who claim to have seen intelligent behavior in the slime mold. This fascinating and beautiful life form consists of single cells that often live separately but that are able to join up and form an organism consisting of a single cell with very many nuclei. Thus, the cell nuclei can exist in two radically different levels of social complexity.
In their paper, Nagakaki et al have used the slime mold physarum polycephalum. This mold forms a network with pulsating tubes between food sources. Essentially, what they have been able to show is that the physarum is able to optimise its network structure to form the shortest route between several food sources. Furthermore, it is able to do so in a labyrinth that the scientists had constructed.
Encouraged by this result, Soichiro Tsuda and his colleagues at the University of Southampton have tried a radical approach: they have built a robot controlled by a slime mold brain. In this case, they are using the physarum’s predilection for darkness to let it steer the robot, and they have apparently succeeded in building a system that consistently walks the robot away from light. (Their article gives a brilliant example of the kind of lucid reasoning in which mathematicians and physicists excel.)
These examples challenge the definition of intelligence, but in my opinion they fall short. In neither case does the system show any adaptive behavior. The mold is only able to solve a labyrinth problem by testing all possibilities, i.e. by growing into all parts of the maze and then retracting from those where no food was found. And in the robot case, the mold functions by giving a specific response to a specific stimulus, which any transistor setup could accomplish as easily.
Tsuda and his colleagues note that biological systems display an enviable combination of adaptivity and robustness. Artificial systems have, until now, been unable to replicate this successful trade-off. Tsuda’s proposed solution is to put biological systems in control of the computers, but there exists a different and more appealing way forward. We can learn to understand biological information processing! When we are able to formulate a functional principle-driven model of the cell’s brain, we will be able to replicate it with silicon components.