If a transformer had infinite data and parameters, I'm sure it could simulate human reasoning to a high degree. Humans don't work that way, so we may need to create a more general definition for artificial reasoning