There’s some humbling news from the chemical world for anyone who has ever found themselves lost in a garden maze. A simple droplet of organic solvent can find its way through a complicated labyrinth with nothing more to go on than a slight pH difference.
Bartosz Grzybowski’s team at Northwestern University in Evanston, Illinois, used a common polymer to fashion a two-dimensional labyrinth some 2 centimetres on each side. They then flooded the maze with strongly alkaline potassium hydroxide solution, before placing a hydrochloric acid-soaked chunk of gel at the maze exit.
After about 40 seconds they placed a droplet of mineral oil containing hexyldecanoic acid at the maze entrance. The oil, which cannot mix with the potassium hydroxide solution, sits on the surface. But it remains still only for a matter of seconds – it soon begins tearing around the maze at speeds of up to 10 millimetres per second, sniffing out the shortest path to the acid-soaked gel, and solving the maze in the process.
“In the movie files you can see the droplet makes decisions,” says Grzybowski. “It goes left along the wrong path, decides there’s something fishy with that and so it reverses. It looks almost alive.”
But while Grzybowski says the droplet displays behaviour that might be called “primitive intelligence”, there’s a simple chemical mechanism at work.
The droplet leaches its acid into the surrounding solution, losing hydrogen ions in a process known as deprotonation – a process that affects the surface tension of the droplet itself.
“But to the front and rear of the droplet [the surrounding solution] has a different pH,” he says, because of diffusion from the acid-soaked gel at the maze exit. Those tiny pH differences affect the amount of deprotonation that happens at the front and rear of the droplet, and this asymmetry sets up a surface tension gradient that forces the droplet into motion. “I would say the droplet is self-propellant,” he says.
Grzybowski’s team thinks that the behaviour could have implications for cancer treatments. They would like to develop micelles – aggregates of molecules such as balls of lipids – that can navigate pH gradients in the body. “A good reason for that is cancer is more acidic than the rest of the body,” Grzybowski says.
While a person can learn a route through a maze and then negotiate the maze by memory, a person would appear equally smart to an outsider if they simply followed signposts in the maze to reach the exit. “A smart person, like the droplets, is often smart due to canny combinations of internal and external structure,” says Clark.
It’s a powerful idea that is filtering into theories about artificial intelligence. Rolf Pfeifer at the University of Zurich in Switzerland is exploring how to “outsource” some of the cognitive load of artificially intelligent systems. He points out evidence that the way our knees absorb the energy of a jump is controlled by the material properties of the leg itself: the reactions happen too quickly to be controlled by the brain or even a reflex. Through careful choice of materials, Pfeifer is now applying that idea in his robot creations by designing body parts that are capable, to some degree, of autonomously reacting to their environment.
Karl Friston, a neuroscientist at University College London, goes further. He says the human brain and the oil droplet do share some fundamental attributes, in particular in the way they both respond to their environment.
This ties in with Bayesian brain theory, which pictures our brains as attempting to understand the world by observing the environment and making, then improving, predictions about what will happen next. Friston is working on a unified theory of the brain (New Scientist, 31 May 2008, p 30) that mathematically describes how the brain continually improves its predictions by observing its environment and minimising errors.
He sees “deep similarities” between his theory and the droplet’s movement. As the droplet moves towards the exit it is moving towards a state of chemical equilibrium, where it has minimised its free energy.
Work on artificial neural networks has shown that the same principles apply to these networks: by minimising the difference between the predictions a network makes and what it actually senses happening, the network is also driven towards equilibrium. Friston is now showing how the equations that govern neural networks and thermodynamic systems apply to real brains.
A second drop seems to solve the maze even faster. Picking up clues from the first drop?
Two drops solving a maze at once: