Responsibility lies on:

  • the institution that designs the system: for allowing any of the harms to take place
  • ethics board: for passing an unethical product
  • developers: many harms best anticipated by developers, can only be tested / minimised by the developer, responsible for algorithm transparency, efficiency and power consumption

We can minimise harms of AI through:

  • awareness of ethical issues
  • participatory design
  • team diversity
  • balanced datasets
  • testing with diverse groups
  • documenting datasets, models, algorithms, design choices
  • logging, explainable methods
  • ethics boards
  • whistleblowing