We've all heard about the dangers of AI (Artificial Intelligence) and the so-called "AI Apocalypse," which is changing (and destroying) our lives as we know it.
But before AI turns into the psychopathic “killer” we’ve been waiting for, there’s no harm in watching the hilarious attempts of technology on its journey to conquer our world! Like this new AI technology called AttnGAN, which is trained to produce accurate images based on text input. The researchers trained the bot on a specific dataset, such as images of birds or cats.
The Generative Adversearial Network (GAN) algorithm was originally published by Microsoft's Deep Learning Group, and so far, it creates blurry images of birds, while the work was also published in a research paper on arXiv.org.
According to Microsoft's blog: "GAN works well when generating images from simple text descriptions, but the quality stagnates when it accepts more complex descriptions.".
In other words, the algorithm works decently with single-word texts, but seems to get confused when a full sentence is entered, losing detailed image information. As a result, the images it creates are blurry and a bit “scary.”.
Cristóbal Valenzuela, a technologist and researcher from New York, created a demo of AttnGAN, so you can try out some texts yourself.
Although AI seems to be quite weak to handle this task, the overall idea is very useful, given that a working GAN could have many practical applications, such as helping painters and interior designers with their sketches or as a tool for voice-activated photo taking.
