Computers are deterministic. Giving the same inputs training would produce the same model. The comparison with brain is incorrect. You could add noise on input data during the training - it would more of less reproduce the real learning. Still, it could produce less useable models as a result.

The court could ask to show the training dataset.