KAIST College of Business Seminar & Forum > Notice > News >KAIST COLLEGE OF BUSINESS
본문 바로가기 사이트 메뉴 바로가기 주메뉴 바로가기

Academic SeminarCollaboration and Delegation Between Humans and AI: Experimental Investigation of the Future of Work

  • Date
  • 2019-07-29 ~ 2019-07-29
  • Time
  • 10:00 ~ 11:30
  • Place
  • Supex Building, Chey A Hall
  • Department
  • School of Management Engineering
  • Major
  • IT Management
We would like to invite you to participate in Management Engineering (ME) Seminar.

1. When: July 29th (Monday), 10:00~11:30
2. Where: Supex Building, Chey A Hall
3. Speaker: Prof. Alok Gupta (University of Minnesota, Editor-In-Chief, ISR)
4. Topic: Collaboration and Delegation Between Humans and AI: Experimental Investigation of the Future of Work
5. Research field: IT Management
* Lecture will be delivered in English.
* Seminar materials: Abstract

A defining question of our age is how AI will influence the workplace of the future and, thereby, the human condition. The dominant perspective is that the competition between AI and humans will be won by machines with evolving technologies such as deep learning that can often provide consistency of decision making that humans can’t. However, this perspective has its distractors that argue that while AI has made massive strides, the future of work is about working with machines rather than against them. We share the vision that AI together with humans might produce better outcomes byleveraging complementary skills that machines and humans’ poses. However, very little work has been done to explore whether humans can effectively work with machines and what might be “natural” barriers for effective human-AI collaboration. In this study, through a set of behavioral experiments,we let humans and a state of the art AI classify images alone and together. Expectedly, the AI outperforms humans. Humans could improve by delegating to the AI, but this combined effort still did not outperform AI itself. We also study a novel condition (inversion) in which the AI classified images, andit delegated images to humans when it was uncertain. Interestingly, inversion outperformed all other settings by a large margin. We then explore the reasons why humans are not effective delegators. First, our results indicate that: humans don’t delegate enough. However, we show that human delegation can be improved by teaching the subjects effective delegation strategies. More interestingly, we show that even with higher delegation the performance does not improve significantly because humans are not a good judge of their own capabilities. Our study point to a gap in research and educational efforts that is going to be critical in effective Human-AI collaboration.
Contact : Lee, Jisun ( jisunlee@kaist.ac.kr )