4 Comments
User's avatar
Anthony S's avatar

“The AI cannot be accused of bias and dismissed.” That’s simply not true.

Keith Dear's avatar

Our article clearly has failed to explain well enough what optimising for the truth means. It simply can’t be gamed, and trust will be won by consistency in performance.

Something for us to look at and explore in the next one. Thank you for the feedback and engagement.

Keith Dear's avatar

It is absolutely true if you are optimising only for prediction accuracy and calibration. Which we, and companies like us, are.

If you - human or machine - allow bias to dominate or influence your forecasts you will be less accurate, less well calibrated, and should be, and will be, less trusted. The bias can be hidden in narrative, but not in numbers.

Anthony S's avatar

It may be unfair to lump all AI manufacturers and developers together, but seeing the behaviour of the “tech bros”, I don’t think the public has much trust in AIs and the people behind them. In any case, why should they? Self-regulation, which despite a bit of internationally uncoordinated government tinkering, is a joke. I don’t think anything is going to change for AI until the question, “Quis custodiet ipsos custodes?” can properly be answered.