A JOINT MESSAGE FROM YANG LIQUN AND OTHERS: WHY CAN'T AI LEARN ON ITS OWN LIKE A HUMAN BEING

Meta, New York University and UC Berkeley, three researchers suggested that AI lacked the capacity to learn on its own like human beings, and that the current model of post-deployment learning, i.e., fixed, data screening and training programmes were all dependent on human engineers “outsourcing”; the paper proposed a two-system framework for integrated observation learning (System A) and operational learning (System B), which was coordinated by the meta-controler System M, and drew on the two-tiered optimization of evolutionary and developmental mechanisms to solve cold start-up problems; the researchers felt that it might take decades to distance from a fully autonomous learning system, while noting that the greater the degree of autonomy, the more difficult it is to guarantee Zi, could also create new ethical challenges, such as a shift in purpose, emotional attachment to misdirection。