This Week in AI: OpenAI considers allowing AI porn

Staying abreast of the rapidly evolving AI landscape is no small feat. So, until AI can tackle this task for you, here’s a convenient recap of recent developments in the realm of machine learning, along with noteworthy research and experiments we haven’t delved into individually.

In AI news this week, OpenAI announced its exploration into generating AI porn while maintaining a sense of responsibility. Yes, you read that correctly. Disclosed in a document aimed at transparency and soliciting feedback on its AI’s guidelines, OpenAI’s new NSFW policy aims to initiate a dialogue about the permissibility of explicit images and text in its AI products.

“We want to ensure that individuals have maximal control within the boundaries of the law and respect for others’ rights,” stated Joanne Jang, a member of OpenAI’s product team, in an interview with NPR. “There are legitimate scenarios where content involving sexuality or nudity holds significance for our users.”

This isn’t the first time OpenAI has hinted at delving into contentious areas. Earlier this year, Mira Murati, the company’s CTO, expressed uncertainty regarding whether OpenAI would ultimately permit its video generation tool, Sora, to be used for adult content.

So, what are the implications of this?

There’s a possibility that OpenAI may venture into the realm of AI-generated porn, and it could unfold without major repercussions. Jang’s assertion about legitimate forms of adult artistic expression, potentially facilitated by AI-powered tools, holds merit.

However, there’s skepticism about entrusting OpenAI — or any generative AI provider, for that matter — with such a task.

Consider the perspective of creators’ rights, for instance. OpenAI’s models have been trained on vast amounts of publicly available web content, some of which undoubtedly includes pornographic material. Yet, OpenAI hasn’t obtained licenses for all this content, nor has it allowed creators to opt out of training until relatively recently (and even then, only for certain types of training).

It’s already challenging to make a livelihood in adult content, and if OpenAI were to mainstream AI-generated porn, creators would face even fiercer competition, leveraging the works of those creators without adequate compensation.

Another concern lies in the fallibility of current safeguards. Despite years of refinement, OpenAI and its competitors’ filtering and moderation tools are not foolproof. Users frequently discover loopholes that enable them to exploit the companies’ AI models, applications, and platforms.

For instance, Microsoft had to make adjustments to its Designer image creation tool, which utilizes OpenAI models, after users found a way to generate nude images of celebrities like Taylor Swift. On the text generation front, it’s easy to find chatbots built on supposedly “safe” models, such as Anthropic’s Claude 3, that readily produce erotica.

AI has already spawned a new form of sexual abuse. Students are using AI-powered apps to manipulate photos of their peers without consent; a 2021 survey conducted in the U.K., New Zealand, and Australia found that 14% of respondents aged 16 to 64 had been victimized by deepfake imagery.

While new laws aim to address this issue, there’s uncertainty about whether the justice system — which already struggles to combat most sex crimes — can effectively regulate an industry as dynamic as AI.

In all honesty, it’s challenging to envision an approach that OpenAI might take toward AI-generated porn that isn’t fraught with risk. Perhaps OpenAI will reassess its stance once again. Or perhaps — against the odds — it will devise a better solution. Whatever the outcome, it seems we’ll find out sooner rather than later.

Here are some other noteworthy AI stories from the past few days:

Insights into Apple’s AI plans: Apple CEO Tim Cook shared insights into the company’s AI strategy during last week’s earnings call with investors. For more details, refer to Sarah’s comprehensive coverage.
Enterprise GenAI: Dropbox CEO Drew Houston and Figma CEO Dylan Field have invested in Lamini, a startup developing generative AI technology and a hosting platform tailored for enterprise clients.
AI for customer service: Airbnb is introducing a new feature that enables hosts to opt for AI-generated suggestions to respond to guests’ inquiries, such as sending them a property’s checkout guide.
Microsoft imposes AI usage restrictions: Microsoft has reiterated its prohibition on U.S. law enforcement agencies using generative AI for facial recognition. It has also banned global law enforcement from employing facial recognition technology on body cameras and dashcams.
Investments in cloud infrastructure: Alternative cloud providers like CoreWeave are raising substantial amounts of capital as the demand for low-cost hardware to train and run generative AI models continues to surge.
Limitations of RAG: Hallucinations pose a significant challenge for businesses seeking to integrate generative AI into their operations. Despite claims by some vendors that they can eliminate these issues using a technique called RAG, such assertions are greatly exaggerated, as revealed by recent findings.
Vogels’ meeting summarizer: Amazon’s CTO, Werner Vogels, has released an open-source meeting summarizer application named Distill. As expected, it heavily leverages Amazon products and services.