According to Sara Hooker, languages are what truly connect us as humans. The former Google Brain researcher since 2022 has led Cohere for AI, the nonprofit research lab of Cohere, the Toronto-based AI model company valued at $5.5 billion. And Hooker speaks from experience. Her Irish mother and British father met in Sudan, and Hooker grew up moving to Mozambique, Eswatini, Kenya, and Liberia. That meant she heard and had to learn many different languages—she even attended middle school in Mozambique where lessons were only in Portuguese.

“Language is super personal for me to work on,” she tells Fortune. “There’s this really powerful idea that when you speak in the language of someone, you really connect with their heart, not their head.”

As it turns out, language is also at the heart of Cohere for AI’s mission to do fundamental research into machine learning problems, through open collaboration and democratized access. The organization’s Aya initiative, launched two years ago, collaborated with over 3,000 researchers from 119 countries to expand cutting-edge multilingual AI research, which is considered critical work because most AI models today are heavily skewed toward English and a few dominant languages like Chinese, Spanish, and French. This creates significant limitations in AI’s accessibility, fairness, and usability for the majority of the world’s population, who speak less widely represented languages.

“From the beginning, we doubled the number of languages that were covered by generative AI,” Hooker says.

Last week, Cohere for AI also released Aya Vision, an AI model that integrates both language and images capabilities, supporting inputs in 23 different languages for everything from image captions or answers to questions based on images. One example: Imagine entering an image of an artwork you see while traveling overseas. The model can understand not only what style was used but what region it originated from.

“One of our big pushes with this recent release has been making things accessible,” she says. Since so many people in Africa use WhatsApp, she explains, Cohere for AI is enabling free access to its Aya models on the app, allowing users to access the model on a platform they already use to communicate every day.

Not in her plan

Hooker emphasized that she never set out to work on AI. When she earned a full scholarship at age 19 to attend Carleton University in Minnesota, her dream was to eventually work at the World Bank. But when she headed to the Bay Area after graduation to work at startup Udemy in 2015, she realized pioneering research on deep learning provided a powerful set of tools to solve problems.

“I didn’t set out wanting to do artificial intelligence, but I had always wanted to work on interesting problems,” she explained. That led to a position as a researcher at Google Brain, which had only recently become an industry AI lab. In 2017, she began working with computer scientists Geoffrey Hinton, considered one of the “godfathers” of AI, and Samy Bengio, who is now director of AI research at Apple. “They were very sincere and ambitious, which is a rare combo,” she said. “At a certain point in your career, what matters most is you’re working with real people, and there’s an authenticity about the mission.”

That authentic mission was also clear when Cohere co-founders Aiden Gomez and Nick Frosst invited Hooker to lead the new Cohere for AI. One big bet for the new research lab was to recruit talent from other parts of the world besides a handful of top cities like London and San Francisco. Now, she said, Cohere for AI senior research staff is across Germany, Denmark, the U.S., Brazil, and Colombia.

“Our North Star has been very clear,” she said, though it took other companies a while to get on board with the idea that countries want AI that works for their languages. “Our efforts were really ahead of the curve,” she added, pointing out that Cohere for AI offered some of the first AI models for languages like Korean, Vietnamese, and Hindu, as well as African languages like Swahili. “We chose a problem which was critically underserved, but now everyone else is scrambling to catch up,” she said.

Back to her roots

It has been a full-circle moment for Hooker, who says growing up in Africa made it obvious that AI models for more languages was a massive need. “I think people forget—we can build the best model in the world, but what matters is how people feel about it and whether it works for them,” she said.

Her connection with Africa remains strong: Hooker says she stays involved with the machine learning community there, including attending an Africa-hosted annual conference that has taken place in Tunisia, Ghana and South Africa, and will be held this year in Senegal.

It’s a long distance to travel from her current home in Santa Monica, California, where she moved after spending a decade in the Bay Area. While she loved Silicon Valley for its “intoxicating” energy and level of talent, Hooker admits that the region can also foster mediocre ideas with so many people attending the same dinner parties and listening to the same tastemakers.

Spectacular ideas often spin out of areas beyond Silicon Valley, she says—though those areas may struggle to raise the same amount of capital that can be had in the Bay Area. “The most radical ideas come when people feel like they don’t have much to lose,” she explained. “Once you’re in a quarterly planning cycle, and your main ingredient is throwing more computing power at a problem, that rarely brings the type of radical innovation that you need in the long term. You need to step outside to go forward.”