It doesn’t have to be zero-sum Human or machine.
An augmented process combines both. Like an iron man suit.
Key
We must considre presentation of that augmentation to minimize cognitive bias.
Basically we want the highly capable human being on the wheel and the AI as a copilot, human insight first, machine suggestions later to support his judgement. Like R2D2.
Cognitive bias
consider a finanacial anlyst monoring fraud alerts We may say ah the AI knows best and sacrifize our own jusdgement. How AI information is displayed to humans matters a lot:
- Forced display, you see the AI answer next to the alert. You will likely trust it.
- Optional display: think about it first and then request the AI It’s like leetcode solutions.
Also apparently we don’t care about confidence scores with AI predictions. That’s nuts. A model with a 10% accuract rate is no better than a human. might be fasteer and certainly easier.