Human-Centered AI

Active Learning as an example of HAI: In an iterative process, a machine learning model intelligently queries annotations from a human to learn from. The goal is to maximize a given objective function, e.g., classification accuracy, at minimal annotation effort, e.g., the number of annotation acquisitions.

From text translation over quality management to recommendation systems, Artificial Intelligence (AI) machines help humans with a wide range of tasks.

Recently, the interaction between humans and machines has become the focus of current human-centered AI (HAI) research. Humans and machines should no longer act independently of each other. Instead, both parties combine their skills to accomplish a goal. A prominent example of HAI is Active Learning, where the machine actively queries the human for information or knowledge to accelerate its own learning progress. The queries are formulated and selected intelligently to minimize the workload of the human.

By supporting the collaboration between humans and intelligent machines, HAI does not replace humans but overcomes the limits of previous AI solutions to bridge the gap between humans and machines.

Central Research Questions of HAI:

  • How to design an effective interaction between humans and machines?
  • How to explain a machine’s decisions and visualize its learning progress to humans?
  • How to combine and leverage the intelligence of humans and machines?