The speaker discusses incorporating lessons learned from deep neural nets and deep reinforcement systems into the development of openCOG 2.0, without completely dismissing their potential for AGI.