Introduction Getting Started Ethics of Artificial Intelligence Artificial Intelligence Systems What is different about AI? The Most Important Question Algorithmic Bias Understanding Fairness Disparate Treatment and Disparate Impact A Fair Fat Pet Predictor The Impossibility Theorem Context-Free Fairness Learning about the Context Understanding Bias I Two Types of Harms Harms of Allocation Harms of Representation Understanding Bias II Bias from Data Bias from Algorithm Design Bias from Deployment Summary Checklist Resources About

Summary Checklist

Here is a checklist of questions and prompts to ask when implementing an AIS. While there is no strictly correct answer, a good rule of thumb is that we should be okay with publishing our answers publicly.

Again, this checklist is best completed as a group exercise and with extensive inputs from users and people who might interact with the proposed AIS.

Section 1 - Understanding the Context

General Context

  1. What is the ultimate aim of the application?
  2. What are the pros and cons of an AIS versus other solutions?
  3. How is the AIS supposed to be used?
  4. What is the current system that the AIS will be replacing?
  5. Who will interact with the AIS?
  6. Create a few user personas - the technophobe, the newbie etc. - and think about how they might react to the AIS across the short-term and long-term.
  7. Think of ways that the AIS can be misused by unknowning or malicious actors.

About Fairness

  1. What do false positives and false negatives mean for different users? Under what circumstances might one be worse than the other?
  2. Try listing out some examples of fair and unfair predictions. Why are they fair/unfair?
  3. What are the relevant protected traits in this problem?
  4. Which fairness metrics should we prioritize?

Section 2 - Preparing the Data

  1. What is our population?
  2. How does our dataset distribution differ from our population distribution?
  3. Are we measuring the features/labels the same way for different groups?
  4. How are our annotated labels different from the ideal labels?

Section 3 - Training the Model

  1. How do our input features relate to our protected traits?
  2. Do we use the same model or different models for different inputs?
  3. If we are importing a pre-trained model or external data, what are possible conflicts between these imports and our current context?

Section 4 - Evaluating the Model

  1. How does our test distribution differ from our population distribution?
  2. What can we say about the fairness of our final model?
  3. When we detect some unfairness with our metrics - is the disparity justified?

Section 5 - Deploying the Solution

  1. How do we detect errors from the AIS after deployment?
  2. What are alternative solutions in case of failure?
  3. How can we allow users to gracefully opt out of the AIS?