In case you want a picture of a wizard corgi playing the guitar. Image: Google
A team of Google researchers has built an AI capable of generating eerily realistic images from text prompts, once again raising questions about the ethical nature of developing an AI that can interfere with our shared reality.
The AI is called Imagen and the images its creators have chosen to show the world are surreal and breathtaking.
These systems stand out for their ability to parse outlandish natural language prompts and create consistent, instantly recognizable images.
Some of the images chosen to illustrate Imagen use simple prompts where subject characteristics are swapped as in the prompt, “a giant cobra snake on a farm”. The serpent is made of corn’.
Imagen’s version of a snake made out of corn. Image: Google
Others are more contextually complicated, such as with the “An art gallery featuring paintings by Monet” prompt. The art gallery is flooded. Robots circle the art gallery using paddle boards.”
Robots enjoying a gallery exhibit. Image: Google
Imagen is the latest in an image-generating arms race between AI companies after OpenAI revealed last month DALL-E 2 and began providing limited access to it for public testing.
According to a guide research article on ImagenGoogle’s AI is easier to train and the resolution of its images can be increased more easily than its competitors.
In some cases, Imagen also demonstrates better understanding of detail than DALL-E 2, especially when building from prompts that include embedded text.
Imagen works well when it comes to including text in images. Image: Google
Unlike OpenAI’s DALL-E 2 – which is onboarding 1,000 users per week on its demo platform – Google has no plans to allow public testing of Imagen.
As the researchers explained, while generative AI has the potential to “supplement, extend and augment human creativity”, it can also be used maliciously for harassment and harassment. spreading false information.
Part of the problem with generative images is that training data tends to be pulled from the open web.
This leads to consent issues for people who may appear in the dataset, but also significant concerns about data containing stereotypes and “oppressive viewpoints”.
“Training text-image models on this data risks reproducing these associations and causing significant representational harm that would disproportionately impact individuals and communities already experiencing marginalization, discrimination and exclusion within society. society,” the Google researchers wrote.
“We strongly caution against using text-to-image generation methods for any user-facing tool without special care and attention to the content of the training dataset.”
Generated images of people are particularly problematic, the creators of Imagen found, and have a demonstrated bias “in favor of generating images of lighter-skinned people” while also tending to reinforce gender stereotypes around occupations.
OpenAI is also aware of these issues and has integrated a filter for public use of DALL-E 2 designed to stop certain content to be created, such as violent or sexual images, and those that may be linked to conspiracy theories or political campaigns.
In a recent blog postOpenAI said 0.05% of user-generated images were automatically flagged, with less than a third confirmed by human reviewers.
While encouraging users to flood social media with AI-created images, OpenAI has urged early adopters of DALL-E 2 “not to share photorealistic builds that include faces.”
#Googles #generate #strange #photos