Google DeepMind COO Laila Ibrahim talks about the responsibility of building AI

london
CNN
—

Laila Ibrahim’s first love wasn’t a computer. Somewhat surprisingly for Google DeepMind’s first COO, it’s not artificial intelligence, and it’s not coding.

Ibrahim’s first love was engineering, and she says that background is what makes her work in computing so valuable.

“I became an (electrical) engineer because I thought it was a combination of math, art, and science. And along the way, I really enjoyed working with real people. What I enjoy about my career is being able to bring it all together and bring a unique perspective to everything I do,” Ibrahim told CNN’s Anna in a recent interview at Google DeepMind’s headquarters in London. -Told Stewart.

“Being an engineer has led us to ask questions of what, why, and what are we trying to accomplish. So if we can truly understand the problem, we can simply add a ton of solutions to undefined problems.” You’ll be able to figure out what the right solution is instead of throwing around.

As a professional problem solver, part of her job “is to worry,” Ibrahim said. “What are the risks? How can we mitigate them? But also think about the opportunities and how can we support them? … It’s almost a moral imperative to play this role. Suddenly, my very strange and circuitous background made some sense to where I was sitting now.

Ibrahim said one thing she has learned is that she is not very good at predicting the future. “But I’m good at building it.”

The daughter of immigrants who grew up with English as her second language, Ibrahim spent her childhood and adulthood first in the American Midwest, then as an exchange student in Japan, and by the time she entered college, she was an “outsider”. He said that he often felt that way. I studied electrical engineering at Purdue, Indiana.

“There weren’t that many women,” Ibrahim recalls. “In fact, I could count them on my fingers at the time.”

By then, in her early 20s, she added, “I was used to bringing a different perspective to everything.”

While that “outsider” mentality felt like a hurdle at first, Ibrahim says the biggest lesson she learned was to embrace it as a superpower, and she wishes she had accepted it sooner. .

She built a career, first at computer chip maker Intel and then at a venture capitalist firm. Ibrahim became the first president and COO of online learning platform Coursera.

In 2018, an exciting opportunity came across her desk. DeepMind, an artificial intelligence research institute founded in 2010 and acquired by Google in 2014, was looking for its first COO.

“After 30 years of my career, I wanted to think very carefully and intentionally about what the next chapter would be,” Ibrahim said.

“But how can you say no when you have the chance to work on innovative technology like this? So I actually joined the conversation, and I did it very slowly and deliberately. We wanted to understand what the founders’ vision was for what could be achieved and what the risks were.”

Ibrahim says he spent a total of 50 hours interviewing for the position, considering the possibility of entering the exciting but often controversial world of AI.

“I went home, tucked my daughters in at night, and said, ‘What legacy can I leave in the world?'” After all, how to build AI responsibly? I felt there was no better place than DeepMind,” she said.

Ibrahim’s love for engineering was influenced by his Lebanese father, who was orphaned by the age of five and grew up to become an electrician.

“I remember when[I was]a kid, he had these beautiful paintings on his desk at home,” Ibrahim said. “And we’re going to see these photos become microchips that go into things like heart pacemakers. So this orphan from Lebanon is going to actually save people’s lives through engineering pacemaker research. It should have been possible.”

Motivated by his father’s example, Ibrahim views his work in terms of impact. Perhaps the most important example at Google DeepMind is AlphaFold, the company’s AI program that can solve what are called protein prediction problems.

“Proteins are the basic building blocks of life,” Ibrahim explained. “And if we can understand how a protein folds, we can understand the function it has and what goes wrong when it folds incorrectly. Alzheimer’s disease, Parkinson’s disease. Diseases are all protein-related problems.

What would normally take human researchers years to study a single protein can now be completed in minutes. The company also made AlphaFold open source. This means it can be accessed by researchers anywhere in the world (according to Google DeepMind, it is used by more than 2 million people in 190 countries, and that number is growing). In October, two of Ibrahim’s colleagues won the Nobel Prize in Chemistry for their work on the program.

“We weren’t expecting[a Nobel Prize],” Ibrahim said, “definitely not this year,” noting that Alphafold is only four years old.

“AlphaFold is just the first step,” she added. “We have a portfolio of research being done not only in biology, but also in chemistry, physics and other areas.”