Rethinking human-AI interaction

This essay by AI specialist Jessy Lin explores some of the possibilities to rethink how humans and “intelligent” machines interact today. The focus is on practical application — how can technology that exists today be leveraged creatively to make applications that are vastly more powerful and useful?

Humans-as-backup as the dominant paradigm of human-machine collaboration is limiting our imagination. It suggests that the only way we can give feedback (if at all) to machines is to label data points as examples of what we want. It embodies the idea that the only way for machines to help humans is to do the easy tasks for them, leaving the hard cases up to the human. But there are actually other ways we can imagine, too: what if humans + machines together could seamlessly handle all the work? Or, what if machines could help us do hard things better, even if they aren’t able to do them on their own?

These questions suggest new directions for human-AI interaction to consider:

  1. Change the loophow can humans direct where / when machine aid is helpful?
  2. Change the inputshow can we make it more natural for humans to specify what they want?
  3. Change the outputshow can we help humans understand and solve their own problems?