Explainable AI

Explainable AI

AI is a major contributor to the spread of misinformation. Increasingly, it is also being use to mitigate its spread. However, to do this ethically, AI's decisions need to be explained in a way that humans can understand. The problem is that the reasoning AI employs is often so complex, that it defies human understanding. Finding a way to explain such decisions in a way that genuinely useful to humans, is incredibly challenging. Below we describe some progress we have made towards this essential goal.

  • Singh, R, Miller T, Sonenberg, L, Velloso, E, Vetere, F, Howe, P, Dourish, P. (2024). An Actionability Assessment Tool for Explainable AI. arXiv preprint. [PDF]
  • Vered, M, Livni, T, Howe, P.D.L., Miller, T., Sonenberg, L. (2023). The effects of explanation on automation bias. Artificial Intelligence, 322, C, 103952. [PDF]
  • Singh, R, Miller, T,  Lyons, H, Sonenberg, L, Velloso, E, Vetere, F, Howe, P, Dourish, P (2023). Directive explanations for the actionable explainability in machine learning applications. ACM Transactions on Interactive Intelligent Systems, 13(4), 23, 1-26. [PDF]
  • Vered, M, Howe, PDL, Miller, T, Sonenberg, L, & Velloso, E (2020). Demand-driven transparency for monitoring intelligent agents. IEEE Transactions on Human-Machine Systems, 50(3), 264-275. [PDF]

Explainable Models

An alternative strategy is to develop models that are inherently explainable. Such models are constructed from the outset to be understandable by humans. Below is an example of where we have done this. Naturally, the model focuses on forecasting democratic collapses, as avoiding such collapses is a key focus of the lab.

  • Baillie, E, Howe, PDL, Perfors, A, Miller, T, Kashima, Y, Berger, A (2021). Explainable models for forecasting the emergence of political instability. PLOS ONE, 16(7), e0254350. [PDF]

Large Language Models (LLMs)

With the advent of ChatGPT, the field is becoming dominated by LLMs such as ChatGPT. Below is some of our earlier work exploring the capabilities of ChatGPT. Here is a fun popsci commentary on our article.

  • Howe, P.D.L., Fay, N., Saletta, M., Hovy, E. (2023). ChatGPT’s advice is perceived as better than that of a professional advice columnists. Frontiers in Psychology, 14. [PDF]