I'm not sure to what application of simulated annealing you are reffering to, but the algorithm is largerly about choice and the initial tempeature is one (it might be some 'good' choices of initial temperature though). Like, If you did some boids based learning for your AI, at the first temperatures you would move particles in random directions, and when the temperature is cold you follow the heuristics (like movement = temp*random + (1-temp)*heuristics etc.). A good choice of initial temperatur might be 1. Not sure if this helps
Edit: I think the correct name of the boids thingie, might be Particle Swarm Optimization. Maybe not really a learning thingie, but it has something to do with AI .
Edit II: I don't mean that the initial temperatur =1, it is a choice..