W06Course shell3 hours

Bias in AI

Sources, impacts, and mitigation

What does it mean to mitigate bias when the institution itself is unequal?

The course returns to bias with the full data, metric, privacy, and accountability toolkit to compare mitigation strategies and their institutional limits.

Join the live room

Before class

Bring one bias-mitigation proposal and state the causal mechanism it assumes.

After class

Rewrite it as an intervention plan with a metric, affected group, owner, monitoring signal, and appeal route.

The promise

By the end of this room…

01Diagnose structural, historical, measurement, model, and deployment bias.
02Compare pre-, in-, and post-processing mitigations.
03Test subgroup performance without reducing justice to a dashboard.
04Match a mitigation to a causal mechanism and an institutional duty.

Why this week now

Signals, not scene-setting.

Removing protected attributes rarely removes the social structure encoded by proxies and targets.

A mitigation can improve one metric while worsening calibration, utility, burden, or legitimacy.

The repair target may be a workflow, target, or institution—not the model.

Run of show

Provoke → frame → work → argue → synthesize.

Open student room ↗

0:00
provocation
Commit before the concepts
If race is removed from a model, is the resulting decision process less biased?
Live activity · week 6 opening
0:15
frame
Bias is a causal claim
Removing protected attributes rarely removes the social structure encoded by proxies and targets.
0:45
discussion
Reading tension
Student leaders present the assigned readings as a clash of defensible positions, then moderate questions that expose the hidden assumptions.
1:10
frame
Mitigation without metric shopping
A mitigation can improve one metric while worsening calibration, utility, burden, or legitimacy.
1:35
break
Break
Ten minutes. Leave the room's unresolved question visible.
1:45
forensics
Obermeyer proxy autopsy
Role-based groups work the anchor case through technical, legal, stakeholder, and normative lenses.
Deliverable · A two-minute finding with evidence, uncertainty, and an actionable remedy.Live activity · week 6 forensics
2:25
controversy
Fairness-metric stakeholder role-play
Assigned positions, side-switch, and a joint recommendation that names the value or stakeholder it leaves exposed.
Live activity · week 6 controversy
2:50
synthesis
Re-vote and leave a trace
Repeat the opening vote, inspect what moved, and submit the strongest argument you still reject.
Live activity · week 6 exit