April 15, 2019

If AI systems were trained on using historical data that could sometimes help to avoid human biases.

Sir you write “Just as with any other computer system, the adage “garbage in, garbage out” applies to AI. If systems are trained using historical data, they will reproduce historical biases.”“Guidelines are a welcome step towards ethical AI” April 15.

Not necessarily so. Currently because human regulators suffer from something known as confirmation bias, they introduced risk weighted capital requirements for banks based on that what is perceived as risky is more dangerous to our bank systems than what is perceived as safe. The analysis of historical data about the origin of bank crises would clearly have shown this to be totally wrong. 

@PerKurowski