Dall-E 2’s Images are not so Delightful, Thanks to the Prevailing Bias
OpenAI’s Dall-E 2 generated illustrations from written descriptions are not so delightful images
Dall-E is the newest AI technology that both inspires and concerns researchers. It is a new AI system that can create realistic images and art from a description in natural language and a neural network that can take any text and turn it into images. It is a technology made by OpenAI, an artificial intelligence research lab co-founded by Tesla CEO Elon Musk and many others. It highlights the problem of AI bias and the need to change incentives in the industry.
OpenAI’s Dall-E is learned by another OpenAI model called Contrastive Language-Image Pre-training. CLIP is trained on hundreds of millions of images and their associated captions. It has learned the relationship between images and the text used to describe them. The CEO of OpenAI, Sam Altman, created DALL-E 2 as the most delightful thing to play with. Apparently, these Dall-E 2 images weren’t drawn by any human illustrator.
OpenAI’s Dall-E 2 produces fantastical images:
Just write down what you want to see, and the AI draws it with vivid detail, high resolution, and, arguably, real creativity. This new AI system can turn textual descriptions for a wide range of concepts expressible in natural language into images. It is a 12-billion parameter version of GPT-3 trained to generate images from text descriptions, using data of text-image pairs.
Dall-E can make realistic edits to existing images from a natural language caption and also can add and remove elements while taking shadows, reflections, and textures into account. It can take an image and create different variations of it inspired by the original. It generates more realistic and accurate images with 4x greater resolution. It is also capable of generating a lot of images that are not delightful.
This feature of OpenAI is applied in painting, which works as a more sophisticated version of Photoshop’s content-aware fill, realistically adding or removing elements from a selected section of the image while taking into account shadows, reflections and textures. And it is also recognized that AI inherits biases from the corpus of data used to train it, millions of images scraped off the internet and their corresponding captions resulting in reinforcing social stereotypes.
DALL-E’s capacity to create such pictures, in any case, would be restricted. Everything unequivocal or roughly happy was eliminated from its preparation information, so it has little openness to these ideas. And One of OpenAI’s different ventures is Dactyl, which included preparing a robot hand to agilely control objecting human-like developments it instructed itself.
OpenAI scientists made a few endeavors to determine predisposition and decency issues. However, they couldn’t root out these problems in an effective way because different solutions result in different trade-offs. Furthermore, it is a long way from the main artificial intelligence organization managing predisposition issues and compromises. It’s really difficult for the whole AI group.