It's recommended, only due to their most recent project in artificial intelligence: Norman, the AI who's also a psychopath. The MIT researchers were trying to prove the point that "the data that is used to teach a machine learning algorithm can significantly influence its behavior", and therefore, if you'll use it to make any important decisions, the data you feed it matters.
However, after plundering the depths of Reddit and a select subreddit dedicated to graphic content brimming with images of death and destruction, Norman's datasets are far from what a standard AI would be exposed to.
In order to test Norman's psychological status after his Reddit binge, the researchers used Rorschach inkblots, which they claim "is used to detect underlying thought disorders". Nice, eh? Per MIT, Norman's psychopathic tendency "represents a case study on the dangers of Artificial Intelligence gone wrong when biased data is used in machine learning algorithms". After it was trained through this data set, as shown in the link they compared the Norman AIs captioning with the regular AI's and the vast difference in which they classified the image could be noticed without any problem.Читайте также: Flying cars could soon become a reality
"When people talk about AI algorithms being biased and unfair, the culprit is often not the algorithm itself, but the biased data that was fed to it", the researchers say. Where one AI saw a vase with flowers, Norman saw a man shot in front of his "screaming" wife.
The researchers trained Norman to caption images, which means creating short descriptions for what they see. While a typical AI might see a close-up of a wedding cake on a table, Norman will look at that same inkblot test and see a man who had been murdered by a speeding driver. Norman, however, saw "a man is electrocuted and catches to death". AI can also be used for good, like when MIT managed to create an algorithm called "Deep Empathy" previous year, to help people relate to victims of disaster.При любом использовании материалов сайта и дочерних проектов, гиперссылка на обязательна.
«» 2007 - 2018 Copyright.
Автоматизированное извлечение информации сайта запрещено.
Код для вставки в блог