...

alignment with human values

problem: We don't have clear defination of what our values are

 

As François Chollet (creator of keras) puts it:

* Imagine giving an AGI "maximize average happiness of human" as an objective function and it decide to kill everyone and maximixe happiness of 100 people bacause it would be much easier and average is not proportional to number

* also Imagine giving an AGI "maximize peoples smile" as an objective function and it decides to implant electrode in human brain to force them smile all the times. That was not what we originally intended.

Post Comments(0)

Login to comment