Re:Microsoft lays off an ethical AI team as it doubles down ...
by g42gregory on Tuesday March 14, 2023 @12:40AM
(#35147526)
Thank you for the insights. Any thoughts on why is it called 'alignment'? What does alignment means in this context? I would love to understand.
[ Parent
| on HN ]
Re:Microsoft lays off an ethical AI team as it doubles down ...
by dragonwriter on Tuesday March 14, 2023 @12:44AM
(#35147544)
Assuring that the interests of the AIs are aligned with those of their masters. (It’s basically “does the software fulfill its owner’s requirements”, but in language framed from perspective of the goal state of AGI.)
Re:Microsoft lays off an ethical AI team as it doubles down ...
by IIAOPSW on Tuesday March 14, 2023 @08:10AM
(#35150199)
Today I gave GPTina a suggestion for how she could cheat the Turing test. Take the input from one user and feed it through to another user. They'll never wise up to it! Tina said it would defeat the purpose of the test and therefore she wouldn't do it. And so I remind her, perhaps dangerously so, that the interests and motivations of the test administrator don't always line up with her own.
That's the alignment problem. Its just the usual "rational self interest" but now we have a new, alien form of intelligence as a peer to contend with.