If the flaws are not proven it becomes hard to hold the person in charge of the software accountable.
And AI enables purposeful flaws to be built in, covertly (like a "bug").
I see real danger there when the programmer can essentially use any datasets they choose to create something flawed/sold that bypasses traditional responsibility.
And that's just my take on it.
Would be nice if people w/held responsible, I just don't see infrastructure/regulation for that today.
Follow
Machine learning security is still mostly a research field.
