Stability AI CEO steps down ‘to fix concentration of power in AI’

News

Mar 25, 20243 mins

Generative AIIT GovernanceTechnology Industry

AI governance is a hard problem, and Emad Mostaque wants to spread the workload.

Emad Mostaque stepped down as CEO of Stability AI, the developer of the Stable Diffusion image generation tool, over the weekend, saying he wanted to “fix” the concentration of power in AI.

The company said he had also quit its board of directors, which had appointed COO Shan Shan Wong and CTO Christian Laforte as interim co-CEOs while seeking a replacement for him.

Board chairman Jim O’Shaughnessy thanked Mostaque for “his leadership and relentless commitment to Stability AI and the open source movement,” in a statement confirming the changes.

On Twitter/X, Mostaque said that his shares of Stability have a majority vote and give him full board control.

Having one person hold this kind of power in a company, or a having few companies dominate the industry, seems to bother him, though.

“We should have more transparent and distributed governance in AI as it becomes more and more important,” he wrote. “The concentration of power in AI is bad for us all. I decided to step down to fix this at Stability & elsewhere.”

“It’s a hard problem, but I think we can fix it,” he wrote, hinting that he has a plan by adding, “Will be sharing more soon.”

Mostaque’s concerns about the centralization of generative AI are understandable, but not everyone is convinced that’s really what’s going on in the industry now.

Diversity in AI

“All of these companies have their own models,” said NYU Tandon school of engineering associate professor Chinmay Hegde, “so it’s a very big and diverse landscape — I wouldn’t call the situation tending toward centralized.”

Many of the most prominent AI models are closed source, with Meta being a notable exception among the major players but the number of AI models available, coupled with the relative novelty of the technology on the commercial market, creates a certain amount of diversity.

“This is all very nascent,” said Hegde. “We’re about 18 months removed from the first wave of [generative AI] coming out.

The closed-source nature of many of the most popular AIs — Anthropic’s Claude, OpenAI’s ChatGPT, and so on — may be closer to the heart of Mostaque’s concerns about the state of the industry. Hegde noted that it’s important to retain a diverse landscape of AIs, particularly those that are open source, to address many of the key issues facing generative AI.

“There are issues surrounding copyright which are being debated in the courts right now, [and] issues related to bias and other harms,” he said. “These questions are very much part of ongoing research in the community, and removing access to these models means that fewer resources are being devoted to these very important problems.”

Stability isn’t without other issues, according to Hegde, making it far from unique among AI startups. Reports indicate that a number of researchers have left the company recently, and smaller AI companies are facing some common headwinds.

“These smaller AI startups are burning a lot of money and there’s not a lot of differentiation between the companies,” Hegde said. “So it’s hard to see where the long-term revenue is going to come from.”

Mostaque could not immediately be reached for comment.

SUBSCRIBE TO OUR NEWSLETTER