Towards a better understanding of artificial intelligence and its interaction with its environment

Developing new tools to both understand the consequences of interactions between artificial intelligence (AI) and the systems they purport to study.

The need and impact

Researchers are increasingly seeking to solve complex problems using computer generated predictions (artificial intelligence, or AI). Traditionally these algorithms have been framed narrowly as neutral, black-box statistical methods, and these predictions are often applied to living things such as crops for productivity, or pests or diseases for their management.

There have been increasing concerns that this narrow focus is leading to perverse outcomes ranging from overly optimistic forecasts to misleading support for discrimination. Similarly, we rarely consider how living things will further adjust in response to the changes caused in their environment by the application of the outputs from the AI (that is, the feedback mechanisms).

The approach

This project will develop new tools to both understand the consequences of interactions between AI and the systems they purport to study, and will develop new types of AI algorithm which may allow for better interpretability of the mechanisms used to make decisions.

Research aims

  • Develop a conceptual framework for the feedback between artificial intelligence and the systems they study.
  • Employ dynamical systems methodologies to identify possible causal links in feedback loops.
  • Using traditional AI algorithms, examine a “weedkiller robot” as a case study to:
    • Estimate the importance of feedback
    • Test if interactions between people and AI will lead to unexpected feedback
    • Identify unintended consequences using traditional AI algorithms.
  • Using a “glass box” AI algorithm on a toy model of the “weedkiller robot” system, use dynamical systems methods to identify whether any of the above negative consequences could have been avoided.
  • If time and resources permit, use methods developed for other applications relevant to Aotearoa New Zealand.


  • Dr William Godsoe (Project Co-Lead)
  • Associate Professor Claire Postlethwaite (Project Co-Lead)
  • Dr Emma Sharp
  • Victoria Agyepong