💡 Try it first with everything switched on. Later you can come back here, turn off Gender, and retrain – to see whether that's enough to make the model fair.
03
Train the model
Let the AI learn the patterns in the data
04
Examine the consequences
What did the model learn – and is it fair?
The Fairness Test
240 new applicants with identical merit distributions
The model's learned weights
what the model weighs in its decision
The Twin Test
two identical candidates – are they judged the same?
Start over from the beginning
📋 The training data – Teknova AB's history
Rows highlighted in red = people who were qualified but not hired. Because the data is constructed for this lab, we know the "true suitability" – in reality no such ground truth exists, which is exactly what makes the problem hard to detect.
The Bias Lab
An AI learns from data. But what happens when the data reflects old injustices? Here you'll carry out the investigation yourself – step by step – and see how discrimination can creep into an AI system.
01
Train and review – Let the model learn from the history. Then see what it has learned, how it judges new applicants, and how it responds to two identical CVs where only gender differs.
02
Try to fix it – Go back, hide the applicants' gender from the model, and investigate whether that's enough to make it fair.
💡 All the data in this lab is made up, but this type of bias occurs in real AI systems.
The Fairness Test
240 new applicants with identical merit distributions were run through the newly trained model.