3 ms·
But to do this you would need unbiased people. Since those don't exist, this correction would just be matching the bias to that of the bias adjuster.
by fallingknife 5y ago
But to do this you would need unbiased people. Since those don't exist, this correction would just be matching the bias to that of the bias adjuster.
- zdragnar 5y agoNah, you just need enough people to review the fixes. If at the end everyone is unhappy with the results, you're good to go.
- dbt00 5y agoThere's an infinite number of wrong answers that also anger everyone, so that's not a sufficient metric for correctness.
- samstave 5y agoThis was exactly my point -- at what inflection point is the AI's decision sound, vs when it may be based on bias from its base creation code (whatever the substrate code that 'births' an AI? WTF do we call that (as obviously an AI is meant to iteratively evolve - at what point is an AI required to 'check in changes' such that if a rollback is required across that AI's reach may be accomplished... We need a "product recall" method that doesnt involve Blade Runners and campy one-liners... SERIOUSLY