Humans and AI often possess distinct yet complementary strengths. While AI systems can process large datasets quickly and identify patterns, humans bring experience, intuition, and contextual knowledge. By combining these strengths, significant efficiency gains can be achieved. A key question is: how do we decide which tasks to handle ourselves and which to delegate to AI?

This brings us to the concept of task delegation. When humans delegate tasks to AI where they are less competent and focus on their own strengths, both can achieve better results together. In practice, however, people often struggle to decide effectively when to pass tasks to AI. A primary reason is that people may lack accurate self-assessment of their own skills, which is referred to as "meta-knowledge" — understanding what they know and what they don’t.

The Role of Explainable AI in Decision-Making

Explainable AI (XAI) refers to AI systems that make their decisions or predictions understandable for humans. Rather than being a "black box," an explainable AI provides information about the factors that led to a particular decision. For example, it may indicate which features were most influential.

Prof. Bauer investigated whether XAI could help people better assess when it makes sense to delegate tasks to AI. The idea is that if people better understand how AI works and which information it uses, they may also gain a better understanding of their own skills in relation to AI.

How Explanations Improve Delegation

In a field study involving 149 participants, Prof. Bauer examined how explanations of AI functionality influence delegation behavior. The task was to assess property values — an activity that can be performed by both humans and AI systems.

The study divided participants into two groups. One group received explanations about how the AI arrived at its price estimates, including information on key factors like square footage, number of rooms, balcony, and other elements affecting price. The other group received no explanation of the AI's valuation process.

Next came task delegation: participants had to decide which properties to evaluate themselves and which to delegate to the AI. They had limited time to make these choices.

Both the participants and the AI evaluated the properties independently. Additionally, participants were asked to estimate their own accuracy as well as that of the AI.

Results and Practical Implications

The results showed that participants who received AI explanations delegated tasks to the AI more frequently and with greater confidence. The delegation rate increased by one-third, and confidence in delegation rose by a quarter. Importantly, these delegations were also more effective — participants were more likely to let the AI handle tasks where it actually performed better.

The main reason was an improvement in participants' meta-knowledge. Through AI explanations, they could better assess in which areas the AI outperformed them and in which it did not. Interestingly, this not only led to a better understanding of the AI but also provided a more realistic view of their own capabilities.

These findings carry important implications for the design of AI systems. By making AI more explainable, we can enhance human-machine collaboration. People can make more informed decisions about when to rely on AI and when not to, leading to better overall outcomes.

Conclusion

Prof. Kevin Bauer's research talk illustrated the crucial role of explanations in human-AI collaboration. Through explainable AI, people can not only better understand the capabilities of AI but also gain insights into their own. This leads to more efficient delegation decisions and improves the overall performance of human-machine teams.

In a world where AI increasingly influences decisions, it is vital to design these systems to be transparent and understandable. Prof. Bauer's research makes an important contribution to understanding how we can maximize the synergies between humans and AI, while also enhancing self-knowledge.