Tutorial 1 - COMPAS
Task
Ahead of the tutorial, familiarise yourself with the COMPAS case study through the readings below.
Summary
Correctional Offender Management Profiling for Alternative Sanctions (COMPAS) is a risk assessment algorithm developed by a company called Northpointe, Inc (now Equivant), designed to assess a defendant’s potential risk of recidivism – the likelihood that someone who has committed a crime will reoffend in the future. The algorithm, now used by several US jurisdictions, combines multiple sources of information, including a questionnaire filled out by defendants, to assign a recidivism risk score from 1 to 10 (however, the company has not disclosed how they modeled the algorithm, nor what kind of data they collected). The algorithm classifies offenders as low-risk (1-4), medium-risk (5-7), and high-risk (8-10). These results are then included in the defendant’s documentation and submitted to the sentencing judge. A study conducted in 2016 by ProPublica has shown that the algorithm is disproportionately biased against black defendants. In particular, the study found that, while COMPAS predicted recidivism in both groups with the same accuracy rate (60%), the algorithm treated black defendants unfairly with regards to incorrect scores. Indeed, black defendants were almost twice as likely to be labeled as high-risk, but did not actually re-offend, compared to their white counterparts (45% vs. 23%). Similarly, white defendants were almost twice as likely to be labeled as low risk when they did in fact re-offend, compared to their black counterparts (48% vs. 28%).
Core Readings
Angwin, J., Larson, J., Mattu, S., & Kirchner, L. (2016, May). Machine bias. ProPublica.
https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing
Corbett-Davies, S., Pierson, E., Feller, A., & Goel, S. (2016, October 17). A computer program used for bail and sentencing decisions was labeled biased against blacks. It’s actually not that clear. The Washington Post.
Other potential readings:
https://www.theatlantic.com/technology/archive/2018/01/equivant-compas-algorithm/550646/
More technical: https://link.springer.com/article/10.1007/s00146-022-01441-y
Submission
There is no submission required for this tutorial, but there will be for the subsequent ones.
Prompt
This tutorial will spend more time talking over the format and case study, but the following example prompt is provided as practice for how to think about prompts in later tutorials.
"How might this case study be an example of developers acting as though they have “absolute moral authority”, or an example of the limitations of computer science’s focus on quantitative methods?"
Advice:
- If you don’t know how to approach this, try to think about the following:
- Do we need to define any terms?
- Are there other related questions that this brings up?
- Is there missing information we might need? (you are encouraged to search for more sources)
- Don't feel like you should make a "new" point all the time. Respond to other people’s points by building on them or (respectfully) challenging them.
- Remember you're not looking for a definitive answer here. The prompt is a starting point for discussion to branch off of (though try not to get off topic entirely).