Sessions

GOTOpia February 2021

In light of the recent deep learning driven success of AI in both corporate and social life there has been a growing fear of human displacement and a related call to develop IA (intelligence augmentation) rather than pure AI. In reality, most current AI applications have a significant human-in-the-loop (HITL) component and is therefore arguably more IA than AI already. From here, there are currently two trends in the field. In one, increasing machine autonomy is pursued, first by placing the human-on-the-loop (to verifiy the result of the machine computation) and then by hoping to take the human completely out of the loop (as in the pursuit of artificial general intelligence). Two main challenges of this approach are a) the value-alignment problem (how do we ensure that the machine satisfies human preferences when we often cannot even express or agree on these ourselves) and b) the extensive human deskilling that often accompanies algorithmic advances. In my talk, I will discuss how these two challenges may potentially be overcome by the second trend: the pursuit of human-machine hybrid intelligence (HI), in which the two interact synergetically and continually learn from each other. I will describe the outcome of a recent round table series that we have hosted with leading scholars from AI and cognitive science and corporate representatives such as Microsoft and IBM Watson. I will give examples of how we may transition from HITL to HI in fields ranging from image recognition and manipulation, air traffic control and AI-assisted product design to corporate decision making such as corporate take over and optimal team formation.