Totally Model-Free Learned Skillful Coping (2004) [pdf]

It's an interesting article. I haven't quite figured the difference to TD-learning, but this part in particular:

in the spirit of temporal difference reinforcement learning, I hypothesize that certain neurons become critic neurons. The activity levels of these neurons can be interpreted as determining, at each moment, the brain’s critic-neurons-based expected quality of overall performance of the task given that the sensed current environment and the brain’s current state, which, in turn, is a function of what has happened in this particular execution of the task up to now. Although one need not be conscious of this expectation during performance of a task, the phenomenology of skillful behavior suggests that this information is indeed encoded within the brain and is, in fact, available to the conscious mind. A baseball out fielder seems to know only moments after a ball is hit whether it is catchable and with what difficulty and likelihood. A skilled chess player, shown an unfamiliar yet realistic chess position, can report almost instantaneously the probable outcome of the game if it were contested by skilled players

How can that be achieved without a model? I don't understand what "model-free" is supposed to mean. Even Q-learning "works by learning an action-value function that ultimately gives the expected utility of taking a given action in a given state and following the optimal policy thereafter." (Wikipedia). The expected utility depends on a set of modeling assumptions, so it can hardly be called "model-free"...