Uncensored AI art model raises ethical questions – TechCrunch

A new open source AI image generator capable of producing realistic images from any text message has been taken up astonishingly quickly in its first week. Stability AI’s stable diffusion, high fidelity but capable of running on off-the-shelf consumer hardware, is now being used by art generator services such as Artbreeder, Pixelz.ai and more. But the unfiltered nature of the model means that not all use has been entirely over the line.

For the most part, the use cases have been above board. For example, NovelAI has been experimenting with Stable Diffusion to produce art to accompany the AI-generated stories created by users on the platform. Midjourney has launched a beta that taps into Stable Diffusion for greater photorealism.

But stable diffusion has also been used for less palatable purposes. On the infamous discussion forum 4chan, where the model leaked early, several threads are dedicated to AI-generated art of nude celebrities and other forms of generated pornography.

Emad Mostaque, CEO of Stability AI, called it “unfortunate” that the model was leaked on 4chan and stressed that the company was working with “leading ethicists and technologists” on security and other mechanisms around responsible release. One of these mechanisms is a customizable AI tool, the Safety Classifier, included in the general Stable Diffusion software suite that attempts to detect and block offensive or unwanted images.

However, the Safety Classifier – while on by default – can be disabled.

Stable diffusion is very new territory. Other AI art generating systems, such as OpenAI’s DALL-E 2, have implemented strict filters for pornographic material. (The Stable Diffusion open source license prohibits certain applications, such as exploiting minors, but the model itself is not bound at the technical level.) Also, many do not have the ability to make art of public figures, unlike Stable Diffusion. These two characteristics can be risky when combined, allowing bad actors to create pornographic “deepfakes” that – in the worst case – can perpetuate abuse or implicate someone in a crime they did not commit.

Stable diffusion

A deepfake of Emma Watson, created by Stable Diffusion and published on 4chan.

Unfortunately, women are most likely far from being victims of this. A study conducted in 2019 revealed that of the 90% to 95% of deepfakes that are non-consensual, approximately 90% are women. That bodes poorly for the future of these AI systems, according to Ravit Dotan, an AI ethicist at the University of California, Berkeley.

“I worry about other effects of synthetic images of illegal content — that it will exacerbate the illegal behavior being portrayed,” Dotan told TechCrunch via email. “For example, synthetic children want [exploitation] increase the creation of authentic children [exploitation]? Will it increase the number of pedophile attacks?”

Principal researcher of the Montreal AI Ethics Institute Abhishek Gupta shares this view. “We really need to think about the lifecycle of the AI ​​system that includes use and monitoring after deployment, and think about how we can envision controls that can minimize harm even in the worst case,” he said. “This is especially true when a powerful ability [like Stable Diffusion] comes out into the wild that can cause real trauma to those against whom such a system can be used, for example by creating offensive content in the victim’s likeness.”

There was something of a preview last year when a father, on the advice of a nurse, took pictures of his child’s swollen genitalia and sent them to the nurse’s iPhone. The image was automatically backed up to Google Photos and was flagged by the company’s AI filters as child sexual abuse material, resulting in the man’s account being disabled and an investigation by the San Francisco Police Department.

If a legitimate image could trigger such a detection system, experts like Dotan say, there’s no reason why deepfakes generated by a system like Stable Diffusion couldn’t — and at scale.

“The AI ​​systems that people create, even when they have the best intentions, can be used in harmful ways that they don’t expect and can’t prevent,” Dotan said. “I think developers and researchers often underestimated this point.”

Of course, the technology to create deepfakes has been around for a while, AI-powered or otherwise. A 2020 report by deepfake detection company Sensity found that hundreds of explicit deepfake videos featuring female celebrities were uploaded to the world’s largest pornography sites every month; the report estimated the total number of deepfakes online at around 49,000, of which over 95% were pornographic. Actresses including Emma Watson, Natalie Portman, Billie Eilish and Taylor Swift have been targets of deepfakes since AI-powered face-swapping tools entered the mainstream several years ago, and some, including Kristen Bell, have spoken out against what they see as sexual exploitation .

But Stable Diffusion represents a newer generation of systems that can create incredibly – if not perfectly – convincing fake images with minimal user effort. It’s also easy to install, requiring no more than a few installation files and a graphics card that costs several hundred dollars on the high end. Work is underway on even more efficient versions of the system that can run on an M1 MacBook.

Stable diffusion

A Kylie Kardashian deepfake posted to 4chan.

Sebastian Berns, Ph.D. researcher in the AI ​​group at Queen Mary University of London, believes the automation and the possibility to scale up customized image generation are the big differences with systems like Stable Diffusion — and the main problems. “Most harmful images can already be produced by conventional methods, but are manual and require a lot of effort,” he said. “A model that can produce near-photorealistic footage can give way to personal extortion attacks on individuals.”

Berns fears that personal images scraped from social media could be used to condition Stable Diffusion or such a model to generate targeted pornographic images or images depicting illegal acts. There is certainly precedent. After reporting on the rape of an eight-year-old Kashmiri girl in 2018, Indian investigative journalist Rana Ayyub became the target of Indian nationalist trolls, some of whom created deep-fake porn with her face on another person’s body. The deep falsification was shared by the leader of the nationalist political party BJP, and the harassment Ayyub received as a result became so bad that the United Nations had to intervene.

“Stable diffusion offers enough customization to send out automated threats against individuals to either pay or risk having false but potentially harmful footage published,” Berns continued. “We’re already seeing people being pushed out after their webcam became remote. That infiltration step may not be necessary anymore.”

With Stable Diffusion out in the wild and already being used to generate pornography – some without consent – ​​image hosts may be required to take action. TechCrunch reached out to one of the major adult content platforms, OnlyFans, but did not hear back by the time of publication. A spokesperson for Patreon, which also allows adult content, noted that the company has a policy against deep fakes and does not allow images that “reuse celebrities and put non-adult content into an adult context.”

If history is any indication, however, enforcement is likely to be spotty — in part because few laws specifically protect against deepfaking when it comes to pornography. And while the threat of legal action drags down some sites dedicated to offensive AI-generated content, there’s nothing stopping new ones from popping up.

In other words, says Gupta, it’s a brave new world.

“Creative and malicious users can abuse the features [of Stable Diffusion] to generate subjectively offensive content at scale, using minimal resources to run inferences – which is cheaper than training the entire model – and then publish them on places like Reddit and 4chan to drive traffic and hack attention,” Gupta said . “There is a lot at stake when such capabilities are released ‘into the wild’ where controls such as API rate limits, security controls on what kind of output is returned from the system are no longer applicable.”

Leave a Reply

Your email address will not be published.