“Creativity is an attribute of people that is hard to fully quantify,” John Smith, an IBM Research fellow who worked on the project tells CNBC Make It. “It requires some kind of leap of thought, imagination or inspiration that is not easily reproducible in machines.”
That’s because artificial intelligence is good at analyzing information (including language and images) and making decisions based on training, according to the IBM research lab’s blog, but creating new content is “a much more experimental activity.”
To enable AI to create the picture, which was used as an illustration in a New York Times series on AI, the IBM Research team first had the system do things like ingest 3,000 Times articles on artificial intelligence, which helped it learn the most important concepts around AI, like “robot, self-driving, and computing,” according to Axios. The team also used technology to score images from past New York Times articles on their strength in representing AI. The most representative image was determined to be one of a robot and human shaking hands.
From there, to create an original image, the team used over 1,000 existing images to train a neural network (basically a statistical model based on the brain that analyzes complex data) to draw new images of human and robot hands, which it did all day and night for almost a week, IBM notes.
Sample imagery was also used so the AI could create an image that matched the style of New York Times’ visuals.