“When it makes mistakes, when it’s misinterpreted and gives a wrong answer, then you correct it and it learns for the next time. There are errors. I think what’s important is you put safeguards, that there is still an override, that those don’t become decisions. In the more critical the mission, the more you need to have checks and balances just as we do on humans, right? Nothing’s infallible.
“In our research labs, we have this thing that’s called the debater. We have Watson take the pro and the con and argue with itself so there are two Watsons because then there are better decisions. It’s really important we continue to work on the process and improve the way we think.”