People + AI Research Symposium

Last tended Mar 18, 2023 Edit Source

• Ai
• Bias

Intentional ignorance is a value-laden choice, Margaret Mitchell

• choice of dataset and curation define what is important
• model selection encodes additional bias
• generation of new data from users interacting with the model generates a bias feedback loop
• over and underrepresentation of subpopulations encode value judgements
• foresight is critical:
  • intended use / users
  • foreseeable use / users
  • foreseeable discriminatory uses
  • risks
  • foreseeable harms
  • foreseeable disproportionate harms
• “I’m just an engineer / data scientist” is willful ignorance; inflicts foreseeable harm