Well... good article as usual, but... there *is* something exceptional about AI, just looking at *what it does*, as @pluralistic says.
-
RE: https://mamot.fr/@pluralistic/116219642373307943
Well... good article as usual, but... there *is* something exceptional about AI, just looking at *what it does*, as @pluralistic says.
Until AI, all automation was embodied decisions, but AI is *automated decisionmaking*.
That's both new & extremely fraught, because we don't know how humans make decisions & thus have no way of gauging accuracy or transparency of the simulation - even ascribing value to it is difficult.
We do ourselves no favors if we pretend this isn't a problem.
-
RE: https://mamot.fr/@pluralistic/116219642373307943
Well... good article as usual, but... there *is* something exceptional about AI, just looking at *what it does*, as @pluralistic says.
Until AI, all automation was embodied decisions, but AI is *automated decisionmaking*.
That's both new & extremely fraught, because we don't know how humans make decisions & thus have no way of gauging accuracy or transparency of the simulation - even ascribing value to it is difficult.
We do ourselves no favors if we pretend this isn't a problem.
@jwcph That's just not true. We've been building catastrophic automated decision-making systems for decades. This is the origin of the "Computer says no" meme. O'Neill's *Weapons of Math Destruction* - published more than a decade ago, long before transformer models - described the incredible harms of automated decision-making systems. The UK Horizon scandal started in the *previous century*. Australian RoboDebt dates to 2016.
-
@jwcph That's just not true. We've been building catastrophic automated decision-making systems for decades. This is the origin of the "Computer says no" meme. O'Neill's *Weapons of Math Destruction* - published more than a decade ago, long before transformer models - described the incredible harms of automated decision-making systems. The UK Horizon scandal started in the *previous century*. Australian RoboDebt dates to 2016.
@pluralistic Indeed - I am referring to things like that, included under the term "automation" in this context; "intelligent systems" or whatever we call it.
Current AI status is just that + bubble.
The contrast is to e.g. industrial production automation; an embodied set of decisions expressed in a machine that does [thing] automatically. You're not wrong to hark back into history, but I hope we can agree that in this context, 2016, or even 2006, is still "new".
