We all have been pretty much warned about the boiling dangers of AI (Artificial Intelligence) and the so-called “AI apocalypse,” which is changing (and ruining) our lives as we know it.
But, before A.I turns into Norman, the psychopath, we believe there is no harm in watching AI make hilarious attempts on its voyage to conquer our world.
Like this new AI called AttnGAN, which is trained to produce accurate images based on text input. The researchers trained the bot on a specific dataset like the images of birds or cat. But when it was presented with some diverse set of database, the results went something like this –
The Generative Adversarial Network (GAN) algorithm was initially published by Microsoft Deep learning group which, at the moment, came up with blurry images of birds. And the work is also published in a research paper on arXiv.org.
According to the Microsoft blog: “GANs work well when generating images from simple text descriptions, but the quality stagnates with more complex text descriptions.”
In other words, the algorithm performs decent with one-word text, but it seems to get confused while entering a full sentence. And the detailed information of the images is lost. Hence, all the blurry and horrifying images.
Cristóbal Valenzuela, technologist and a researcher from New York, has created an AttnGAN demo so that you can test out some cool text inputs yourself.
Although the AI appears to be pretty naive at this text to image thing, the overall picture is pretty useful since a working GAN can have lot practical applications like a sketch assistant to painters and interior designers, or as a tool for voice-activated photo refinement.