A team of three researchers at MIT are attempting to prove that "An artificial intelligence algorithm would be influenced by the kind of content fed to it." Their method for proving this was to make their algorithm read images from a subreddit dedicated to footage of people dying. After the algorithm read the images it was shown randomly generated ink blots. When standard AI saw the above image it thought it was a person holding an umbrella in the air; When the teams AI algorithm was shown the ink blot it said it was a man shot in the head in front of his screaming wife. This research was done to help show that machine learning algorithms are generally not biased, however, the data they are shown is usually biased. [1 quote, 1 link, 1 image, ]
No comments:
Post a Comment