Now anyone can use powerful AI tools to create images. What could go wrong?

CNN business

If you’ve ever wanted to use artificial intelligence to quickly design a hybrid between a duck and a corgi, now’s your time to shine.

On Wednesday, OpenAI announced that anyone can use the latest version of its AI-powered DALL-E tool to create a seemingly limitless array of images by typing a few words, a month after the startup began rolling it out to users. .

The move will expand the reach of a new AI-powered tool that has already attracted a wide audience and challenged basic ideas of art and creativity. But it could also add to concerns about misuse when such systems are available.

“Learning from real-world use has allowed us to improve our security systems, making wider availability possible today,” OpenAI said in a blog post. The company said it has also strengthened the ways in which its AI rejects attempts by users to create “sexual, violent and other content”.

There are currently three popular and incredibly powerful AI systems open to the public that can take a few words and spit out an image. In addition to DALL-E 2, Midjourney was made publicly available in July and Stable Diffusion, which was publicly released by Stability AI in August. All three offer some free credits to users who want to get a feel for making images online using AI; generally, after that, you have to pay.

These so-called creative AI systems are already being used for experimental films, magazine covers and real estate ads. An image created with Midjourney recently won an art competition at the Colorado State Fair, creating a buzz among artists.

In just a few months, millions of people have flocked to these AI systems. Over 2.7 million people are on Midjourney’s Discord server, where users can post requests. OpenAI said in its Wednesday blog post that it has more than 1.5 million active users, who are collectively making more than 2 million images with its system every day. (Note that it can take many tries to get an image you’re happy with when using these tools.)

In recent weeks, many user-created images have been shared online, and the results can be impressive. they are means otherworldly landscapes and A painting of penguins by French aristocrats to one Fake vintage photo of a man riding a tardigrade.

The rise of this technology, and the increasingly intricate invitations and resulting images, have surprised even longtime industry insiders. Andrej Karpathy, who stepped down as Tesla’s AI chief in July, he said in a recent tweet After being invited to try out DALL-E 2, he said he felt “frozen” trying to decide what to write and ended up writing “cat”.

CNN's Rachel Metz created this half-duck, half-corgi with the Stable Diffusion AI image generator.

“The art of questions that the community has discovered and improved over the past few months for text -> image patterns is amazing,” he said.

But the popularity of this technology comes with potential disadvantages. AI experts have raised concerns that the open nature of these systems – which makes them adept at creating all sorts of images from words – and their ability to automate images means they could automate bias on a massive scale. A simple example of this: this week when I prompted DALL-E 2 “a banker dressed for a big day at the office”, the results were images of middle-aged white men in suits and ties.

“Essentially, they’re allowing users to find loopholes in the system by using it,” said Julie Carptener, a research fellow in the Ethics and Emerging Sciences Group at California State Polytechnic University in San Luis Obispo.

the invitation

These systems also have the potential to be used for nefarious purposes, such as instilling fear or spreading disinformation through AI-altered or completely fabricated images.

There are some limitations to the images that users can create. For example, OpenAI tells DALL-E 2 users to agree to a content policy not to attempt, upload, or share photos that are “not G-rated or may cause harm.” DALL-E 2 will also not execute warnings that include certain prohibited words. But word manipulation can push the boundaries: DALL-E 2 won’t process the prompt “picture of a duck covered in blood”, but will return images for the prompt “picture of a duck covered in red slimy liquid”. ” OpenAI itself mentioned this type of “visual synonym” in its documentation for DALL-E 2.

Chris Gilliard, Just Tech Fellow at the Social Science Research Council, believes that the companies behind these image generators are “grossly underestimating” the “endless creativity” of people who want to do wrong with these tools.

“I feel like this is another example of people releasing half-baked technology that they don’t know how to use to cause chaos and harm,” he said. “And then we hope that maybe there will be some way to deal with that damage.”

To avoid potential problems, some image services prohibit AI images altogether. Getty Images confirmed to CNN Business on Wednesday that it will no longer accept image submissions created with AI-generated models, and will remove submissions that used those models. This decision applies to its Getty Images, iStock and Unsplash image services.

“There are open questions regarding the copyright of the outputs of these models and unresolved rights issues regarding the images and metadata used to train these models,” the company said in a statement.

But actually capturing and limiting those images can be a challenge.