GameBanshee Forums

Nature 417, 37 - 38 (2002)
Free animals can be 'virtually' trained by microstimulating key areas of their brains.
SANJIV K. TALWAR, SHAOHUA XU, EMERSON S. HAWLEY, SHENNAN A. WEISS, KAREN A. MOXON & JOHN K. CHAPIN*

Procedures used to train laboratory animals often incorporate operant learning paradigms in which the animals are taught to produce particular responses to external cues (such as aural tones) in order to obtain rewards (such as food). Here we show that by removing the physical contraints associated with the delivery of cues and rewards, learning paradigms based on brain microstimulation enable conditioning approaches to be used that help to transcend traditional boundaries in animal learning. We have used this paradigm to develop a behavioural model in which an experimenter can guide distant animals in a way similar to that used to control 'intelligent' robots.

Depending on the site of brain stimulation, an electrical stimulus can act as a cue or a reward. Studies of these phenomena have generally been concerned with functional mechanisms of the nervous sytem5, and little thought has been given to the potential of behavioural paradigms constructed wholly around such focal brain stimulations. We used stimulation of the somatosensory cortical (SI) and medial forebrain bundle (MFB) as 'virtual' cues and rewards, respectively, delivered to freely roaming rats. We imposed behavioural contingencies so that an operator could accurately steer the animal, in real time, over any arbitrarily specified three-dimensional route and over a range of real-world terrains.

We implanted stimulating electrodes into the MFB of five rats; the same animals also received electrodes in the right and left SI whisker representations. We then mounted a backpack containing a microprocessor-based, remote-controlled microstimulator on each animal. This allowed the operator, using a laptop computer, to deliver brief trains of stimulus pulses (80 µA; typically 10 biphasic pulses, each 0.5 ms, 100 Hz, per train) to any of the implanted brain sites from distances of up to 500 m away.

We trained the rats for navigation in 10 sessions, during which they learned to interpret remote brain stimulation as instructions for directing their trajectory of locomotion. In a figure-of-eight-shaped maze, the animals first learned to obtain periodic MFB rewards (0.3–3.0 Hz) by running forwards and turning correctly whenever left- or right-turning cues were issued. These cues were presented as a virtual 'touch' to the left or right whiskers by stimulating their respective cortical representations. We then placed the animals in open environments that lacked the boundaries and fixed choice points of the maze. All rats generalized their responses to their new surroundings, running forwards and turning instantaneously on cue. They moved at speeds averaging 0.3 m s-1 and worked continuously for periods of up to a 1-hour test limit.
<...>
Our results show that 'virtual' learning, involving direct stimulation of the central substrates of cues and rewards, can effectively expand the scope of the operant method. Its chief benefit is its ability to dissociate explicit schedule variables such as cues and rewards from the physical variables that are normally associated with their delivery, freeing learning from the mechanical and parametric constraints that are imposed by particular physical settings. MFB reward stimulation is relatively non-satiating, and animals need not initiate consummatory behaviours to obtain such rewards. As virtual cues and rewards are perceived within a body-centred frame of reference, they may facilitate learning independently of the external environment. It may also be possible to increase the 'bandwidth' of conditionable information by stimulating multiple brain sites, thereby increasing the variety of reactions that can be elicited.
<snip>