AI and the Value-Loading Problem
In his book on the future of AI (Artificial Intelligence), Life 3.0, Max Tegmark talks of the importance of ensuring its goals are aligned with ours. This isn’t as easy as it sounds, warns Tegmark. Just think of King Midas: he got literally what he asked for, except that wasn’t what he wanted. Or all those stories where the genie grants three wishes, which almost inevitably end with the third wish being:
“Please undo the
first two wishes, because that’s not what I really wanted.”
Be
careful what you wish for: it might come true.
Put
differently, this is the danger with an AI that takes our instructions
literally:
“The real risk with AGI isn’t malice but
competence.”
So
where does that leave us?
“All these
examples show that to figure out what people really want, you can’t merely go
by what they say. You also need a detailed model of the world, including the
many shared preferences that we tend to leave unstated because we consider them
obvious, such as that we don’t like vomiting or eating gold.”
After
all:
“Children of
hypocrites usually learn more from what their parents do than from what they
hear them say.”
The
so-called (and self-explanatory) “value-loading problem” for AI, is, as Tegmark
wryly says, is “even harder than the moral education of children”.
No wonder so many worry about unleashing AI on the world, without really thinking it through. And no, those who worry aren’t all technophobes; the list includes Bill Gates and Elon Musk.
Comments
Post a Comment