How to Build, Publish, and Share Generative AI Responsibly

“If we really want to tackle these issues, we have to be serious,” says Farid. For example, he wants cloud service providers and app stores such as those operated by Amazon, Microsoft, Google and Apple, all of which are part of the IAP, to ban services that allow people to use deepfake technology. for the purpose of creating non-consensual sexual relations. imagery. Watermarks on all AI-generated content should also be mandatory, not voluntary, he says.

Another important thing missing is how AI systems themselves could be made more responsible, says Ilke Demir, a senior researcher at Intel who leads the company’s work on responsible development of generative AI. This could include more details about how the AI ​​model was trained, what data went into it, and whether generative AI models have biases.

The guidelines do not mention that there is no toxic content in the generative AI models dataset. “It’s one of the biggest ways to cause damage to these systems,” says Daniel Leufer, senior policy analyst at digital rights group Access Now.

The guidelines include a list of harms these companies want to prevent, such as fraud, harassment and misinformation. But a generative AI model that always creates blanks also hurts, and isn’t currently listed, Demir adds.

Farid raises a more fundamental question. Given that companies recognize that the technology could cause serious harm and offer ways to mitigate it, “why don’t they ask the question ‘Should we do this in the first place?’ “”