In a Californian laboratory, a two-year-old culture of living brain cells has been set loose in a virtual world.
Scientists have been coaxing it to take an interest in the world beyond its dish and they believe that the network of thousands of rat cells may be starting to learn.
If its work is confirmed, the team at the California Institute of Technology, led by neuroscientist Steve Potter, will have taken a step towards watching living memories form under the microscope.
The program, outlined in the journal Autonomous Robots , is at an early stage. The goal is to unlock the rudiments of learning by studying how information is physically encoded in small networks of neurons.
The culture neurons owe their longevity to a protective, gas-permeable Teflon membrane. Sixty electrodes connect the cells to a computer-moderated virtual environment that is displayed as a mouse - the first neurally controlled "animat", or simulated animal - wandering around an obstacle-filled room.
Distinct patterns of brain-cell activity picked up by the electrodes and recognised by the computer get converted into movement. Sensory information about the animat's position in its virtual world is relayed back to the neurons within 100ms in the form of patterns of electrical stimuli.
Dr Potter's team have been looking for changes in the way the neurons are organised, prompted by these signals. They argue that this would be indicative of learning.
There have been signs of neural reorganisation that persisted for up to a week, though this latest evidence is still preliminary and unpublished.