Instagram starts letting people create AI versions of themselves

Meta is opening up the ability for anyone in the US to create AI versions of themselves in Instagram or on the web, with a new tool called AI Studio.

The pitch is that creators and business owners will use these AI profiles to talk to their followers on their behalf. They’ll be able to talk directly with humans in chat threads and respond to comments on their author’s account. Meta says Instagram users in the US can get started with AI Studio via either its website or by starting a new “AI chat” directly in Instagram.

In a blog post on Monday, the company writes that “creators can customize their AI based on things like their Instagram content, topics to avoid and links they want it to share.” The post goes on to say that creators will be able to toggle things like auto-replies from their AI and dictate which specific accounts it’s allowed to interact with.

AI Studio also allows for the creation of totally new AI characters that can be deployed across Meta’s apps. Here, Meta is coming after startups like Character.AI and Replika, where people already talk to — and even fall in love with — themed chatbots. Like OpenAI’s custom GPT store, Meta will also surface the AI characters people make for others to try.

Meta’s first pass at this concept was having a handful of celebrities create AI versions of themselves with the same likeness but different names and personas. At the time, Meta said it took that approach because it worried about AI versions of celebrities saying problematic things on behalf of their human counterparts. (Even with the controls built into AI Studio, this is still bound to happen. It’s generative AI we’re dealing with, after all.)

It seems that Meta is at least aware that this is dicey territory. The company says that AI profiles are clearly labeled everywhere they appear. A company handbook for creators goes into more detail about the AI creation process, and it looks like the onus is on the creator to list the topics an AI won’t engage on. One of Meta’s example questions that an AI can be told to not respond to: “Should I invest in crypto??”