What’s next for the Allen Institute for AI: CEO Ali Farhadi charts a course for broader impact

Last year was one of the most productive in the history of the Allen Institute for Artificial Intelligence (Ai2).

The non-profit Seattle-based institute developed and released 111 artificial intelligence models in 2024, while making the underlying training data, code, model weights, and other components available to outside researchers and developers as part of its open-source commitment.

Next up: translating all that work into real-world solutions.

“The dream for us is when these kind of technologies help solve a real problem, and that’s what Ai2 is also moving towards — to learn more about what it means to solve big problems in the world,” Ai2 CEO Ali Farhadi said on the latest episode of the GeekWire Podcast, describing Ai2’s goals for the current year.

He explained, “We’re going to continue to innovate in the whole stack — data, models, post-training, pre-training — but we’re also trying to learn about what it means to fill in the gap from a model to a solution.”

One big focus is Ai2’s role as the AI partner in the Cancer AI Alliance announced last year by Seattle’s Fred Hutch Cancer Center, along with other major cancer research institutes and tech companies.

“Any step in the right direction in that space is going to be a risk worth taking, and that’s the kind of impact that we’re after,” Farhadi said, describing the plans to use AI to address problems faced by oncologists.

“We need to learn a lot about this domain and what technologies apply,” he explained. “What are the AI gaps? What are the problems that cannot be solved? What are the problems that can be solved?”

Ai2, founded in 2014 by the late Microsoft co-founder Paul Allen, is funded by the Allen estate and other donors.

Its board includes Jody Allen of the Paul G. Allen Trust; outgoing University of Washington President Ana Mari Cauce; Hope Cochran of Madrona Venture Group; venture capitalist Steve Hall; former Vulcan CEO Bill Hilf (chairman); Ed Lazowska of the UW Allen School of Computer Science and Engineering, and Microsoft Research President Peter Lee.

A computer vision specialist, Farhadi founded and led Ai2 spinout Xnor.ai as CEO, and sold the AI startup to Apple in 2020 in an estimated $200 million deal that represents one of the institute’s biggest commercial successes to date.

After leading machine learning efforts at Apple, Farhadi returned to Ai2 as CEO and board member in July 2023, succeeding Oren Etzioni in the role. Farhadi is also a professor at the UW Allen School.

At some point in the process of developing and releasing its open-source AI models, Farhadi said, Ai2 realized that being open wasn’t enough — that it was also important to make the technology more accessible and usable.

As one step in that direction, for example, the institute started releasing public demos based on OLMo, Tulu, and other models created by its researchers. Last week, Ai2 released its first on-device AI app leveraging a version of its open-source OLMoE model that can run offline on Apple iOS devices, promising new levels of security and privacy.

Ai2’s work hasn’t received the national attention that has been given to OpenAI or, more recently, DeepSeek. But when asked why we haven’t seen an “Ai2 moment” on the global stage, Farhadi offered a different perspective.

“First of all, we’re a nonprofit,” he said. “We’re not selling anything. We’re not hoping to increase our market value or market price. We are a science-driven institute after doing the right thing … We’re after impact, making the world a better place with AI technologies. The moment that we’re after is when a problem is solved.”

Listen to the full episode above, and subscribe to GeekWire in Apple Podcasts, Spotify, or wherever you listen.