Artificial intelligence has become quite an important part of our life, as we use such machines for all kinds of purposes. However, researchers at MIT decided to build a robot that will bring a few chills down your spine. In an attempt to see what happens when machines go wrong, they created the first psychopath AI called Norman.
The psychopath AI perform a Rorschach test after viewing gruesome image captions
Whenever computers get access to the wrong data, they mind end up processing it in an improper manner. Many people are afraid that, at some point, machines might go mad. While an apocalyptic scenario is still relatively far away, researchers decided to perform a case study. This way, they developed Norman the psychopath AI and put it to some tests.
At first, they started with image captioning, where machines generate text based on an image. This is a deep learning method and it’s quite common among AIs and similar mechanisms. Afterwards, they fed Norman with death-related images on an extremely dark subreddit and their captions.
The second trial consisted of a Rorschach test where both the psychopath AI and a friendly AI had to interpret the inkblots. Of course, there was a huge discrepancy between the answers. Whenever the friendly AI saw a regular image of people sitting at a table or birds on a fence, Norman perceived the inkblots as vicious killing scenes.
This was a case study on the dangers of improper AI use
The psychopath AI wasn’t specifically created to be like that, but researchers encouraged these tendencies. By feeding it with gruesome image captions and death-related content typical of Reddit, the machine developed these scary traits. This sounds really disturbing, and researchers felt the need to explain themselves.
If you’re feeling scared, that’s exactly what researchers wanted. Norman the psychopath AI was created to warn us of the dangers of giving improper data to machines. Artificial intelligence can be helpful for us, but things can go really wrong if such machines access certain types of data.
Image source: Pixabay