If an AI needs to make a change, it makes it immediately and will not make the same mistake again.
It's not all that easy - it's difficult for an ai to realize when it made a mistake, and changing its behavior isn't done with just a small change in the code.
What if it used key stroke decisions in an office to learn how individuals worked, tracked results and then just become very good at data entry :)
Also, would the resulting technology make more mistakes than a human on average? Would accidents on the road go down even if occasionally a turkey interrupted the system and caused it to malfunction? Even if it can't correct itself (yet) it could potentially then try a new approach and tell all other cars to ignore turkeys on the road. There might be lots of dead turkeys but alive humans :D
I wonder how long until they are able to rewrite themselves effectively. I find it all pretty fascinating to think about where it cold take us and I swing between heaven and hell ;)
I was just thinking. Tesla needs a blood type sensor in the bumper bars and if it comes back human, dial emergency... drive away faster... ;D