People learn from watching others, and inverse reinforcement learning (IRL) has given machines a similar capacity. But the complementary problem—how a teacher should choose demonstrations to convey a task—has received far less computational attention, according to a new arXiv paper.
The paper, titled "Showing versus Doing: Teaching by Demonstration," makes this asymmetry explicit. Its abstract states that IRL techniques have realized learning from demonstrations in machines, while teaching by demonstration has been less well studied computationally. The title itself highlights the contrast between showing and doing.
The abstract does not describe the paper's methods or findings, so the contribution is not yet clear from the announcement. What is clear is the motivation: understanding demonstration as an intentional act of teaching, rather than merely as data for learning.