Last year, San Francisco artificial intelligence lab OpenAI upended the tech world with a chatbot that passed the Turing test and the bar exam. It was a bot built using large language models, mathematical representations of how the human brain processes text and speech. These models are so powerful that if used for nefarious purposes, they could do everything from promoting violent video games to spreading misinformation about elections and humanitarian crises.
As a result, they have become a focus of public concern and scrutiny. Companies that develop these models, including Alphabet, Twitter, and Facebook, have set up teams to monitor them for suspicious activity. In some cases, they have censored content that violates their terms of service or have removed accounts that use them to engage in hate speech and other harmful activities.
But as these teams grow, they’re facing new challenges. They need help to make changes quickly enough to keep up with the rate of activity on their platforms and the proliferation of tools for creating harmful content. They also need help finding people with the right skills to fill open roles.
And this is making some leaders in the field uncomfortable, leading them to question whether these departments can be trusted to do their jobs. Inbal Goldberger, VP of Trust and Safety at ActiveFence, has seen the effects of these pressures firsthand. “No one stays in trust and safety who isn’t doing it for authentic reasons,” she says. “It’s not the most glorious or prestigious job, but it’s an important and necessary function for our industry.”
In this episode of AI Now, Goldberger joins host John Montgomery to discuss the state of trust and safety in the digital media ecosystem. They also discuss how a shift toward closing down access to large language models impacts the industry and what it means for the future of AI.
