Submitted via IRC for AndyTheAbsurd
The UK-based company ASI Data Science unveiled a machine learning algorithm Wednesday that can identify terrorist propaganda videos with 99 percent accuracy.
This development marks one of the first instances of a company successfully using A.I. to flag extremist propaganda. The Islamic State group is notorious for its social media recruiting efforts, and this algorithm could help curtail them.
While the researchers at ASI wouldn't discuss any technical specifics of the algorithm, it appears to work like other kinds of A.I. recognition software. The algorithm can examine any video and determine the probability that the video is a piece of extremist propaganda. According to the BBC, the algorithm was trained on thousands of hours of terrorist recruiting videos, and it uses characteristics from these videos to assign probability scores.
Source: https://www.inverse.com/article/41273-uk-company-creates-algorithm-to-flag-propaganda
(Score: 3, Informative) by TheRaven on Thursday February 15 2018, @10:31AM (1 child)
I would love to see school mathematics curricula replace a load of the calculus that they teach with more stats. Calculus is very useful in a few fields, but even there most of what you're doing is constructing a differential equation that models something and then asking a machine to solve it, so practicing solving them until you're reasonably fast for a human, but still orders of magnitude slower than a computer, is not very useful. In contrast, statistics is useful in every experimental subject, including physical and social sciences.
Most of the papers that I review are related to compilers and even at the top conferences it's still common for people to run a benchmark n times, sometimes discard outliers (but not explain why) and take the mean of the 'before' and 'after' versions and then say 'we have an X% speedup!'. Often X is less than the performance variation that you get by randomising the layout of functions in a binary. It's exceedingly unusual to see anyone talking about the distribution, any kind of confidence interval, or what their experimental error was.
sudo mod me up
(Score: 3, Interesting) by opinionated_science on Thursday February 15 2018, @11:45AM
*all* mathematics is better than intuition. Learning how to apply known methods is a key way to investigate the world around us.
"soft" subjects like sociology, psychology and economics, get disproportionate political weight specifically because the mathematical treatment is so weak, and the general public is generally innumerate.
We use differential equations to find possible solutions for given systems. We use statistics to evaluate the quality. The two are not separable, part of the same toolbox.