Should Humans be Users or Slaves of AI? An Experiment on the Future of Work

  • Date: May 6, 2019
  • Time: 04:00 PM (Local Time Germany)
  • Speaker: Wolfgang Ketter
  • University of Cologne
  • Location: MPI
  • Room: Basement

A defining question of our age is how a broadly applicable artificial intelligence (AI) will influence the workplace of the future and, thereby, the human condition. Apparently, the dominant perspective is that AI poses a threat to human workers, and that this competition will be won by either humans or machines. We try to reduce the fog of hype and argue that the workplace of the future may not belong exclusively to humans or machines. Instead, it can be optimal to use AI in concert with human workers, and to combine their unique characteristics and abilities. In our experiments, we let humans and a state-of-the-art AI classify images alone and together. As expected, the AI outperforms humans in terms of correctly classifying images. Humans could improve by delegating to the AI, but this combined effort still did worse than the AI itself. Researchers have used similar results in the past to conclude that humans are doomed as future workers. We have studied the effect of a novel inversion condition in which the AI classified images, and it delegated images to humans when it was uncertain. Interestingly, inversion outperformed all other settings by a large margin. Because it requires human participation, inversion could mitigate potential economic friction from AI usage, but the result is bittersweet: humans would become subordinates to machines. To be clear: we do not sound the bugle on human workers being slaves to AI masters. In fact, we point out that more research is needed that identifies directional complementarity between humans and machines, and that studies characteristics that moderate the relative advantage of each party. In our experiment, the AI was better at realizing potential from the collaboration than the humans. It is unclear at this point if humans cannot delegate well, or if they do not want to delegate because they distrust the AI. We discuss future research on issues such as human reluctance to delegate, trust between humans and machines, and the positive behavioral impact of letting humans work on complex tasks that cannot be turned over to machines easily.


Go to Editor View