Unfortunately, OpenAI quickly dashed hopes that they would relax their policy on generating NSFW with their tools. They published a "Model Spec sheet" on Wednesday outlining ethical behavioral expectations for their systems. NSFW content creation came up in one paragraph, and it mentioned that the company was exploring whether it could allow such use "responsibly." This was quickly widely reported in the mainstream media as OpenAI was thinking about allowing AI porn. Unfortunately, it seems that an OpenAI representative ruled this out in an email sent on Thursday (presumably to Quartz magazine).
An email from OpenAI on Thursday stated:
We have no intention of creating AI-generated pornography. We put the safety of children first and have robust measures in our products to avoid deepfakes, which are unacceptable. In addition, we believe it is crucial to thoroughly consider sexuality in discussions with people of all ages.
Therefore, it is no surprise that OpenAI has ruled out using Sora to generate pornographic videos in the future.
Or does it? Decrypt.co claims that Jenna Jang, the product lead at OpenAI, informed them that the definition of porn varies widely.
Concerns arose, however, when it was suggested that OpenAI was considering loosening pornography production laws. An Open AI product leader tells Decrypt that, in hindsight, the company should have done a better job of explaining what it was "investigating."
Joanna Jang, the product lead for model behavior at OpenAI, stated, "I think what I'd like to do based on the feedback and response, in the next version we share-is be more precise about what some people mean by NSFW content and the taxonomy here." Jang clarified that NSFW can refer to anything from deepfake images created by AI to written profanity.
As for the kind of content that would fall under a more liberal definition, Jang told NPR that it "depends on your definition of porn."
However, she tries to make it clear that OpenAI is "absolutely not in the business of making deepfakes or AI porn," and she goes on to explain why the Model Spec Sheet has caused so much controversy.
According to Jang, OpenAI released the specification to define optimal behavior of models, focusing on compliance, avoiding NSFW material and valuing openness.
The question "Should AI make porn or not?" was asked before the model spec, so we want to add more subtlety to this conversation, Jang said. And I hope the model spec will allow us to have similar discussions in real life.
"Again, this is the reason I wish I had really set up [a framework] so that we could have started this discussion even a day earlier.
she continued.
So it still seems pretty ambiguous to me. I think we can conclude that AI porn is undoubtedly banned, but ChatGPT may be allowed to use more vulgarity in the future, and I wouldn't be surprised if it produces "female erotica." But don't start talking about porn!
Update: Reason magazine has an excellent article on this topic written by Elizabeth Nolan Brown. Click here to read it.
As in this situation, the solution is for regulators to step aside and allow a wide variety of AI products to become widely available. Then individuals with different levels of sensitivity can use the product or products that most closely match their requirements and ideals.
However, based on past events and the reactions of political figures so far, it seems that we will have ideas that are universally applicable. Anything seems like a pretext for a Very Important Congressional Hearing and tech-throttling legislation if all you have is moral panic.
As usual, those who use sex work and have already embraced AI technology will lose the most from this hysteria.
