In this position paper, author Susan Epstein reasons about how society and academia view artificial intelligence (AI), aiming to challenge and inspire research. Although her ideas are not new, she touches on a point constantly discussed in science fiction. Will robots and intelligent systems be a threat to humans? Will they overcome us or make us lose our jobs? Will we be under constant machine surveillance? The author claims AI is frequently depicted as competing with humans and, consequently, the popular attitude toward AI is disdain, grudging acceptance, or fear.
The paper goes to the beginning of the field, explaining how AI research emerged in the 1950s. The idea was to precisely define aspects of learning or features of intelligence so that a machine could simulate them. Intelligent machines would use language, form abstractions and concepts, solve the kinds of problems reserved for humans, and improve themselves. In this view, no mention of enhancing human capabilities was made.
Epstein coined the term “collaborative intelligence” in order to emphasize a different perspective to AI. In her view, machines will help humans and vice versa. Thus, collaborative intelligence is not intended to substitute a human employee, but to engage in a task with one. The human decides when the machine will be used, and the machine asks for human help when needed. Collaborative intelligence involves a model of shared understanding that allows two very different kinds of problem solvers to collaborate on interesting tasks.
We already see this happening in many situations. What is interesting in Epstein’s view is the urge to model the human view of the world, as cognitive science does. This way, the machine would be aware of human perceptions, particularly when they differ from those of other machines. Epstein presents several examples of collaborative intelligence in practice. In her view, a crucial evaluation metric for any research in the area is the extent to which a person believes that his or her work experience or product has been facilitated or improved by the collaboration. Thus, any research on collaborative intelligence demands empirical work and evaluation methods, which include the ultimate judge of a collaborative intelligence success: the person who the machine is intended to assist.