How AI image generators can help robots
Every corner of the internet is flooded with AI image generators that create fantastical images at the intersection between dreams and reality. The entertainment value of these images is shown by the ever-growing treasure troves of random and whimsical images that serve as indirect portals into human designers’ brains. Simple text prompts produce an almost instantaneous picture, satisfying our primitive minds, which are hardwired to seek instant gratification.
The field of AI-generated artwork is not new. It dates back to the 1960s, when early attempts were made using rule-based symbolic approaches for creating technical images. The evolution of models to untangle words and parse them has become more sophisticated. However, the explosion in generative art has created controversy and hype around copyright and disinformation.
Yilun Du is a Ph.D. candidate in the Department of Electrical Engineering and Computer Science and an affiliate of MIT’s Computer Science and Artificial Intelligence Laboratory. He developed a method to make models such as DALL-E 2, more creative and better at understanding scenes. Du explains how the models work and whether or not this infrastructure can be used in other domains. He also explains where we draw the line when it comes to AI and creativity.