Lab 6. AI Ethics

For this lab, you will simply run through these excellent materials:

• Intro to AI Ethics | Kaggle

The modules present concepts in the tutorial followed by exercises to run in Kaggle hosted Jupyter notebooks, most of which use scikit-learn, probably the most commonly used Python-based machine learning platform.

You do not need to turn in this lab, but please do go through all the tutorials and exercises.

Learning Objectives

Upon completion of these materials, you should have an understanding of:

• Human-centered design
  to design an AI system to ensure that it serves the needs of the people that it is intended for.

• Bias
  to determine how AI systems can learn to discriminate against certain groups. Suresh and Guttag (2021) identified six types of bias:

  • Historical bias
    occurs when the state of the world in which the data was generated is flawed.

  • Representation bias
    occurs when building datasets for training a model, if those datasets poorly represent the people that the model will serve.

  • Measurement bias
    occurs when the accuracy of the data varies across groups.

  • Aggregation bias
    occurs when groups are inappropriately combined, resulting in a model that does not perform well for any group or only performs well for the majority group.

  • Evaluation bias
    occurs when evaluating a model, if the benchmark data (used to compare the model to other models that perform similar tasks) does not represent the population that the model will serve.

  • Deployment bias
    occurs when the problem the model is intended to solve is different from the way it is actually used.

• Fairness
  to quantify the extent of the bias in AI. Gajane and Pechenizkiy (2018) explore four types of fairness:

  • Demographic parity
    says the model is fair if the composition of people who are selected by the model matches the group membership percentages of the applicants.

  • Equal opportunity
    fairness ensures that the proportion of people who should be selected by the model (“positives”) that are correctly selected by the model is the same for each group.

  • Equal accuracy
    fairness is an alternative metric to equal opportunity. That is, the percentage of correct classifications (people who should be denied and are denied, and people who should be approved who are approved) should be the same for each group.

  • Group unaware / “Fairness through unawareness”
    fairness removes all group membership information from the dataset.

• Model cards
  to improve public accountability for AI models.