Slashdot Mirror
DARPA's Latest Chip Is Designed To Be Bad At Arithmetic (technologyreview.com)
Reader holy_calamity writes: Pentagon research agency DARPA has funded the creation of a chip incapable of correct arithmetic, in the hope of making computers better at understanding the real world. A chip that can't guarantee that every calculation is perfect can still get good results on many problems but needs fewer circuits and burns less energy, says Joseph Bates, cofounder and CEO of Singular Computing. The S1 chip can process noisy data like video very efficiently because it doesn't need the extra circuits or operations needed to ensure every mathematical operation is performed perfectly. This summer DARPA will put five prototype computers, each equipped with 16 of the inexact S1 chips, online for researchers to experiment with.
As in, in order to get a real AI, it will need to have this fuzzy logic.
Which by the way will end up making our new Robotic overlords require human slaves to do math for them.
Which we will do incorrectly, causing their entire robotic empire to fall in a matter of hours.
I've thought this for a long time (well, not the humorous bits), that it's entirely possible that general intelligence fundamentally requires fuzziness and imprecision, and that by the time we succeed at creating really smart artificial intelligences, we'll find that they're just as error prone and fallible as people are. Which isn't to say we can't make them smarter than we are or that they won't be incredibly useful.
I also wonder if we will find we need to make systems that are deliberately crippled for many of the tasks we want them to do. I mean, imagine installing Marvin the Paranoid Android's brain in a car and requiring it to spend all of its time driving people around. It would likely soon drive off a bridge just to end the boredom. Yeah, Marvin is a fictional character, but it's believable to me that true artificial intelligences will get bored, distracted, make mistakes, etc., just like people, making them perhaps not much better than people at many of the moderately mindless tasks we'd like them to take on. So we'll have to limit them to make them good at what we want them to do... which may make them not so good at what we want them to do.
I think we may also have to deal with the equivalent of mental illness in AIs, and be unable to fully diagnose and/or fix the problems because the system is complex enough to be opaque to us, the same way we don't (and may never) fully understand our own brains and their malfunctions.
Intelligences general and flexible enough to do anything may do everything somewhat badly, and systems sufficiently specialized to do a task extremely well may be unable to cope with the unexpected. Or not. The next few decades are going to be very interesting.
I know whining about common core is a popular pastime among people who have an irrational fear of change,[...]
Actually, most of the Millennials who learned math via Common Core have an irrational fear of change.
For example, I tried to give this young woman at Panda Express 12 dollars and 12 cents, because the bill was 6 dollars and 87 cents, so that I could get a $5 bill and one quarter back from the transaction so I wouldn't have to carry around so many separate bills or extra coins, and she looked apoplectic.
I thought she was going to cry.
She simply could not cope with the change...
Because she could not do simple math in her head.