SoylentNews is people
SoylentNews
[Important Note: Some links (especially Ars Technica) are NSFW for US government employees as they contain slides that are marked "Top Secret". Exercise discretion/caution in this story's comments, too. -Ed.]
Yes, it is cloud-based, yes, it does decide about the fate of hundreds of humans, and yes, ultimately it does direct robots to kill innocent humans.
SKYNET is a system created by the NSA that applies machine learning algorithms to supposedly determine the likelihood of someone turning into a terrorist based on mobile phone metadata. According to slides published at Ars Technica, evil acts like switching off your mobile phone (= evading surveillance), switching SIM cards (= trying unsuccessfully to evade surveillance, thanks to IMEI, etc.), swapping phones with others (= trying unsuccessfully to evade surveillance, thanks to other surveillance data) will be taken together as indicators of your evil intentions.
Patrick Ball—a data scientist and the executive director at the Human Rights Data Analysis Group—who has previously given expert testimony before war crimes tribunals, described the NSA's methods as "ridiculously optimistic" and "completely bullshit." A flaw in how the NSA trains SKYNET's machine learning algorithm to analyse cellular metadata, Ball told Ars, makes the results scientifically unsound.
(Score: 0) by Anonymous Coward on Sunday February 21 2016, @08:09PM
I wonder whether it is anymore possible to hide something with volume.
Sure it is. So long as they aren't looking for something very specific.
If they are just looking for people who might be terrorists, for example, then the number of false positives will overwhelm them. Let's say they have data on 10 million people, and their algorithm is very good and only has a false positive rate of 0.1% (which would be very low), that means they would have 10,000 innocent people falsely flagged as terrorists which they would have to manually examine and exclude (or put under surveilance just in case), on the flip side you also have to consider false negatives, where they don't identify actual terrorists, the lower the false positives, the lower the false negatives, so if they had a false positive rate of 0.1%, they might have a false negative rate of 50%, which would be no good, you could be exceedingly generous and assume they could get it down to 10%, but that would still be missing quite a lot of terrorists and including far too many innocent people. If there were 100 terrorists in that data set, then they would flag 10,090 people as terrorists and then somehow have to determine out of those which are actually terrorist, and yet there are still 10 of them they have failed to spot. In all likelihood here, I am being far too generous with my numbers and they would have both more false positives and more false negatives. If they want to decrease the false negative rate, they will also increase the false positive rate.
This is a nutshell is the problem with trying to find something rare in a large population.