🖥️ A new computer in space and are all our datasets garbage?
🍩 Okily Dokily, we might have The Simpsons forever because of AI
The Good
I wrote about a machine that got shot off to the ISS last month and is the very first edge computer in space. Edge means that instead of sending data to the cloud to be parsed and sorted out, the computer or device has enough compute to do it right there. This allows quicker results, especially in areas with minimal bandwidth. Usually, that is a place like an oil rig or a rural area, but the ISS has only one downlink to Earth, and it is usually tied up.
Edge devices can be pretty small, but this computer, called Spaceborne-2, is the size of a few suitcases and weighs the same as “average-sized panda,” according to the guy in charge of the program (a colorful description from an engineer, you love to see it). Anyway, it is up there to help space experiments do science faster, but is also a test run for Mars, which will require a lot of edge capability as the distance from Earth causes a lag in communication. Also, stress testing edge capabilities in space will probably mean the next generation of devices here on earth will get a major upgrade.
The Bad
The huge datasets that AI systems need are often given labels through Amazon Mechanical Turk. The platform has thousands of people willing to do these labels cheaply enough that it’s financially feasible to label a dataset with millions of examples. While investigations have called out Mechanical Turk for exploiting workers, turns out there is a whole other issue. Many workers are labeling these datasets however they think the powers that be will like. According to one Turk in this Vice article:
“I sometimes find myself thinking like, I think this is a wrong answer ... but I know that if I say what I really think I will get booted from the job, and I will get bad scores and I'm like, okay, I will just do what they want me to do. Even though I think it's a shitty choice.”
This is bad for bias reasons, but also for the sake of accuracy (which, I guess, also turns into a bias issue; it’s bias all the way down). Researchers are spending a lot of time trying to build tools that can learn from fewer examples, that would mostly get around this, as well as other problems. In the meantime, this article into question every dataset labeled by Mechanical Turk.
More News
The Simpsons have been just been renewed for seasons 33 and 34, and the show could go on forever with deepfake technology.
File under I don’t quite understand but sounds neat: using light to run AI chip processes.
Forget adding some noise or distorting images in some way, if you just add a sticky note to an image that says “iPod” some computer vision systems will “see” an iPod.
An algo was used to spot untreated and possibly illegal waste discharges in England.
The Myanmar military, you know, the one that just did another coup, is setting up a face rec tech camera system across the country.
America, meanwhile, has private companies blanketing the country with surveillance cameras with very little oversight.
***
Until I get through a rewatch of I Love Lucy (it’s on Hulu!),
Jackie