Do Machine Teachers Dream of Algorithms
2020
Machine Teaching is an emerging approach for incrementally creating semantic machine learning (ML) models, by focusing on improving the productivity of the human teacher during an interactive ML process. One of the challenges of this approach is developing the support for the process of eliciting subject-matter knowledge relevant for a machine learner. This process we call Knowledge Decomposition not only encompasses knowing what type of knowledge to articulate but also the language, one articulates it with. In turn, language and articulation influence and are influenced by the teacher’s mental models. We share our findings on people’s mental models about learning systems from two formative studies where people teach to hypothetical learning systems. These findings are important for practitioners creating teaching experiences, as the wrong notion of how the machine learner works and what it needs can negatively affect the overall teaching task. We argue that addressing this mismatch is core to the success of machine teaching and a multidisciplinary opportunity for HCI, and ML practitioners alike.
- Correction
- Source
- Cite
- Save
- Machine Reading By IdeaReader
9
References
0
Citations
NaN
KQI