Nature Neuroscience

The idea of a disembodied brain controlling an artificial robot is the stuff of science fiction, but a new study brings this scenario one step closer to reality. John Chapin, Miguel Nicolelis and colleagues have demonstrated for the first time that neuronal activity recorded directly from the brain can be used to control a robotic device in real time. Their findings point the way to the development of new prosthetic devices for paralyzed patients.

The authors trained rats to obtain water by using a robotic arm, which the animals could control by pressing a small lever. As the rats performed this task, the authors analyzed the pattern of activity in the brain regions that control movement, using a special type of electrode that could record from many neurons simultaneously. By analyzing their recordings with a computer, the authors were able to identify patterns of activity that were reliably associated with the rats' paw movements. They then reconfigured the apparatus, so that the robot arm was disconnected from the lever and was instead driven directly by the recorded neural activity. In other words, the rat's brain was now controlling the robot arm directly via the electrode and the computer, rather than via the spinal cord and paw muscles.

The rats had no difficulty in maintaining control of the robot arm in the new configuration. Initially, they continued to press the lever, even though this was no longer necessary to cause the robot arm movements. Many animals soon learned, however, that they could obtain water through brain activity alone and stopped pressing the lever. Thus, they had learned through feedback to alter their brain activity to control the robot device.

The implications for developing human prosthetic devices are discussed in an accompanying News and Views by Eberhard Fetz. This is not the first time brain activity has been used to drive a machine, but it represents a significant advance in several respects. Previous attempts have been based on signals recorded from muscles in the stump of an amputated limb, or electrical brain signals at the surface of the scalp. The former approach cannot be used for patients who have lost control of their muscles because of spinal injury or motor neuron disease, while the latter method allows only a very crude level of control. By recording directly from individual neurons, it should in principle be possible to achieve a much higher degree of speed and precision.

The new study represents a 'proof of principle' for such an approach, but several obstacles must be overcome before it could be applied to human patients. Most importantly, clinical success would depend on the ability to obtain stable recordings from the same neurons over long periods of time; although the present experiments involved recording for several weeks, this would be of limited use for clinical applications. Also, the robot arm is a simple device that can only move in one dimension; recording and decoding enough information to control a device in three dimensions would be considerably more difficult. Nevertheless, a number of laboratories have been pursuing this type of approach, and Fetz believes that the obstacles are ultimately surmountable.

Nature Neuroscience

Amnesia—memory loss—can arise from many forms of brain damage. Some amnesic patients not only fail to remember, they also confabulate—that is, they invent simple or fantastic stories about their recent doings. These patients may act according to these stories and insist that their confabulations are true. Why do some amnesics confabulate while others do not?

Armin Schnider and Radek Ptak now show that confabulators are unable to suppress previously acquired memories that are irrelevant to the current memory task. The authors tested both confabulating and non-confabulating amnesics on a visual memory task. Patients were shown sequences of designs and asked to identify designs that had occured earlier in the same sequence. On the first run, both groups of patients performed equally poorly. However, the later runs contained designs that, although appearing for the first time in that run, had appeared in previous runs. Because the previous run was irrelevant to the current task, these designs should have elicited a 'no' response. Nonconfabulators tended to respond correctly, but the confabulators were more likely to identify them incorrectly as targets. The authors conclude that poor memory alone cannot explain confabulation, and they propose that confabulating patients have lost the ability to prevent the intrusion of previously acquired mental associations that are irrelevant to the current task.

See also News & Views by Tim Shallice.

Nature Neuroscience

Is the visual perception of other animals similar to our own? Visual illusions present an opportunity to address this question, because illusions are misinterpretations made by the brain. If animals also see illusions, then their brains presumably make the same 'errors' as ours, implying that the strategies and assumptions of analysis used by the visual system are similar across species.

In this issue, Andreas Nieder and Hermann Wagner show not only that barn owls perceive an illusion in the same way that we do, but also that neurons in a visual area respond to an illusory line as if it were a real line with physical boundaries. The authors trained owls to peck at a key to indicate whether they saw a triangle or a square presented on a background of parallel lines. Once the owls could discriminate the shapes reliably, the outlines were removed, and instead the figures were defined by offsetting the parallel lines, so that the only cue to the location of the boundary was the position of the line breaks. Humans have no difficulty perceiving the shapes of these figures, even though they have no physical outline.

The authors found that owls too could distinguish the shapes even when they were defined only by illusory contours, suggesting that the bird's visual system—like ours—interprets the illusory contours as lines. The authors went on to show that these illusory contours also activate neurons in the visual Wulst, supporting previous suggestions that this brain region is similar to the primary visual cortex of mammals.

Why did birds and mammals evolve the ability to perceive illusory contours? Presumably because it helps us to resolve ambiguities in the visual environment; the brain mechanism that allows us to see these illusions may also allow us to recognize (say) an animal that is partly obscured by vegetation.

See also News & Views by Izumi Ohzawa.