Key points are not available for this paper at this time.
In this presentation, we demonstrate a way to develop a class of AI systems, the Disruptive Interjector (DI), which observe what a human is doing, then interject with suggestions that aid in idea generation or problem solving in a human-AI (H-AI) team; something that goes beyond current creativity support systems by replacing a human-human (H-H) team with a H-AI one. The proposed DI is distinct from tutors, chatbots, recommenders and other similar systems since they seek to diverge from a solution (rather than converge towards one) by encouraging consideration of other possibilities. We develop a conceptual design of the system, then present examples from deep Convolution Neural Networks1,7 learning models. The first example shows results from a model that was trained on an open-source dataset (publicly available online) of a community technical support chat transcripts, while the second one was trained on a design-focused dataset obtained from transcripts of experts engaged in engineering design problem solving (unavailable publicly). Based on the results from these models, we propose the necessary improvements on models and training datasets that must be resolved in order to achieve usable and reliable collaborative text/voice systems that fall in this class of AI systems.
Joseph Makokha (2022) studied this question.