Human AI Validation
How much time do humans need to devote to validating AI? If any?
You’ve used GitHub Copilot right? It’s great isn’t it.
You type a few characters and the autocomplete on steroids just writes the next few lines for you. Amazing.
How long do you spend checking what it wrote? Anywhere from under a second for when it autocompletes a function call I guess? To maybe closer to a minute if it wrote several lines of code? Shall we say 20 seconds on average?
Seems fair enough. Besides, if it’s wrong there are multiple places to catch its mistakes, right? Syntax highlighting in the IDE, the compiler, the unit tests… Hardly a big deal (though it seems code churn may have gone up as a result of precisely this).
What if hitting tab didn’t just complete your line of code though? What if it meant someone gets killed. Or 10 or 100 people get killed.
How much time is reasonable to spend then before you accept the suggestion?
20 seconds still sound ok? What if there are no other checks in place either? 20 seconds still enough?
