Artificial intelligence has made spectacular progress in fields from medicine to the production lines of factories in recent years. One gap that’s never been overcome is that regarding voice commands. But this gap too may be one about to be surpassed.

Programmer Cris Valenzuela, a resident researcher in the interactive telecommunications program at New York University, created an image generator activated by words as they occur to users (and so long as they’re in English).

The image generator works in real time as you write the words, so each image is slowly “translating” users’ statements and phrases into images. The results may not seem spectacularly realistic at this point, but they are quite evocative.

For example, by entering “football players on a field,” the resulting image looks like an abstract painting, but one which shows colors recognizable as those of a football field during a sporting contest. The field is green below, and the movement of someone wearing a red and white jersey is apparent above. There are even bits of a face and flesh-colored shapes resembling arms and legs in movement.

The project is the result of a research project called Attention Generative Adversarial Network (AttnGAN), based on a graphics processor that learns, little by little, through interactions with users.

The AttnGAN may not give you accurate images in a very short time, but by playing a bit with the program, a user will create images that seem to come from a dream, images that vaguely recall phrases, or perhaps, images as they’re seen in our memories.

Is it possible that these images express, even involuntarily, something of the way the human mind processes images? Are we expecting too much from artificial intelligence at this point? We don’t know for sure. What is true is that the AttnGAN is a beautiful page that allows users to enjoy introducing creative and eccentric phrases and to then witness how artificial intelligence will translates the very dictates of the imagination.

unnamed
unnamed-1
 

 

 

Images: Cris Valenzuela

Artificial intelligence has made spectacular progress in fields from medicine to the production lines of factories in recent years. One gap that’s never been overcome is that regarding voice commands. But this gap too may be one about to be surpassed.

Programmer Cris Valenzuela, a resident researcher in the interactive telecommunications program at New York University, created an image generator activated by words as they occur to users (and so long as they’re in English).

The image generator works in real time as you write the words, so each image is slowly “translating” users’ statements and phrases into images. The results may not seem spectacularly realistic at this point, but they are quite evocative.

For example, by entering “football players on a field,” the resulting image looks like an abstract painting, but one which shows colors recognizable as those of a football field during a sporting contest. The field is green below, and the movement of someone wearing a red and white jersey is apparent above. There are even bits of a face and flesh-colored shapes resembling arms and legs in movement.

The project is the result of a research project called Attention Generative Adversarial Network (AttnGAN), based on a graphics processor that learns, little by little, through interactions with users.

The AttnGAN may not give you accurate images in a very short time, but by playing a bit with the program, a user will create images that seem to come from a dream, images that vaguely recall phrases, or perhaps, images as they’re seen in our memories.

Is it possible that these images express, even involuntarily, something of the way the human mind processes images? Are we expecting too much from artificial intelligence at this point? We don’t know for sure. What is true is that the AttnGAN is a beautiful page that allows users to enjoy introducing creative and eccentric phrases and to then witness how artificial intelligence will translates the very dictates of the imagination.

unnamed
unnamed-1
 

 

 

Images: Cris Valenzuela