Despite having cared for laboratory mice for years, I’m wowed by the complexity of this task! The trainer’s site: http://mouse-agility.com/ describes the animals living conditions and training method.
The trainer keeps the mice housed in groups of 3-4 females, in 3-dimensional cages such as a remodeled cupboard, or:
a pyramid of different-sized tables, the several floors interestingly arranged and connected by ropes or ladders – this will keep them fit in all respects. Another advantage of this husbandry is the perfect airing, the good chance to watch your pets and the direct contact to them for the lack of cage bars or glass… To supply their need for digging, I use to place a few litter-filled boxes onto the tables, as well as some bigger tubs beneath them, used as their “basement”.
Starting at 4 weeks, the mice are trained with “clicker training,” an operant conditioning method developed by students of B.F. Skinner. In clicker training, positive reinforcement of behavior with food, is accompanied with a stereotyped auditory cue such as a click. Repeated pairing turns the click into a conditioned reinforcer, that can more inform the animal what behavior they are being rewarded with greater temporal precision, allowing faster and more complicated training. Using the process psychologists call shaping, mice are trained to do progressively longer and harder tasks.
Mice can be difficult to train with this method because they are naturally cautious and anxious creatures and can be frightened by the noise of the clicker or presence of a trainer. They are also trained using a random reinforcement paradigm, where rewards are not always given at the end of a trial and rewards vary between types of food and caressing.
One of the problems with using mice in experiments (and a large reason why rats are used, despite the sophistication of genetic manipulation in mice), is that they are hard to train to do complex tasks. I don’t know to what extent clicker training and shaping is used by experimenters, but I doubt it is widespread. I would guess that it’s use can extend beyond simple operant conditioning and be used to inform the mice of the relavent cues required for the task. Perhaps using clicker training or by collaborating with expert animal trainers outside of scientific settings we could finally come up with tests of general intelligence, task-shifting, etc., that could then be manipulated genetically in mice.