The next logical direction for #AI evolution should be to implement a new reward system that lets them be wrong and enables their #curiosity:

“As the agent learns, its #prediction_model becomes less and less wrong so that the #reward signal decreases, and the agent must explore other, more surprising situations in order to maximize the reward signal.”

https://www.quantamagazine.org/clever-machines-learn-how-to-be-curious-20170919/