COMPAS

And so...

A team at ProPublica did a study on a recidivism algorithm called COMPAS.

So recidivism algorithm predicts how likely people are to reoffend.

And it found that it was systematically predicting that black folks were more likely to reoffend than white folks were.

I have to say that in the case of AI systems, it also depends on how we train them.

A well-known example of problematic training is the COMPAS system, which is a risk assessment tool used in US courts to predict the likelihood that a defendant will re-offend.

The COMPAS system, C-O-M-P-A-S, for your audience, was found to be biased against because it disproportionately labeled black defendants as having a higher risk of reoffending compared to white defendants with similar profiles.

This bias arose from the data used to train the algorithm, which reflected historical racial disparities in arrests and sentencing.

As a result, the system perpetuated existing inequalities rather than offering an objective assessment.

We have plenty, our entire paper lists out many examples of existing AI systems that are far from AGI.

I think one case we mentioned Compass, which is a five-weight AI.

It's an AI by the OECD definition.

It has five weights.