shaping AI as the world grapples with its potential

Joelle Pineau first encountered the concept of bias in artificial intelligence a quarter of a century ago as a student at the University of Waterloo, Ontario. She had been recruited to a project at the university to help train a voice recognition system for helicopter pilots.

To build an unbiased system, they wanted female pilots to train the technology but could not find any available. So it was Pineau, an engineering graduate, sitting next to a pilot in the cockpit, whose voice was recorded to help train the system on the stress levels a pilot might typically experience.

“I embraced the challenge,” she recalls. “It was very rigorous science, this experience of calling out bias in your data and finding a solution around it. Not that we had enough [of a] sample to really make a strong case but, at least, we could have a measure of the bias and the performance of the system to report on the gap.”

Pineau, now 49, is speaking virtually from a room in Meta’s New York offices, named “Reproducible Research”. She is now vice-president of AI research at Meta, the tech company that owns Facebook, Instagram and WhatsApp.

Meta is a leader in the development of AI. Its Fundamental AI Research (Fair) lab conducts Meta’s long-term research and aims to achieve scientific breakthroughs in the tech, which are implemented into its products across platforms.

Chief executive Mark Zuckerberg has outlined his vision to build artificial general intelligence — where systems have a level of intelligence equal to most humans. Meta says it wants to build this on open source, making the tech readily available to developers and academics instead of behind closed proprietary systems.

Llama, its series of open-source large language models, rivals closed models from competitors such as OpenAI, Google and Anthropic. It has been used in Meta AI, an interface that users can chat with in WhatsApp and Facebook, as well as for image generation.

Meta’s advanced research is done via Fair, where Pineau leads around 1,000 people across 10 locations. “It is a big responsibility,” she says. “Sometimes, I see things in our research lab, and then it’s just a few months before it’s in the hands of millions, if not billions, of people.”

What was the pathway from budding musician and maths whizz to shaping some of the most powerful technology in the world?

Pineau’s fascination with complex subjects was born from a love of maths “from a very young age”, she says. “I remember being fascinated by basic mathematical concepts like the power laws or the logarithmic base.” A keen interest in music — playing piano and viola in orchestras — ultimately remained a hobby.

At 19, Pineau’s first job was working for Canada’s ministry of natural resources, building models relating to solar energy in fish farms. Later, she worked at a fire research lab in her home town, Ottawa. She has a PhD in robotics from Carnegie Mellon University, where she built roving robots that could detect and navigate obstacles and assistants for nursing homes.

It was then that she discovered AI and machine learning, building algorithms and conversational chatbots — primers for the advanced technology behind Meta AI and Microsoft-backed OpenAI’s ChatGPT bot, today. “It was very primitive at the time, the late 1990s, but it was fun.”

There came a time when it was pretty obvious that a lot of the biggest innovation in AI was going to happen in industry

While teaching computer science at McGill University, Montreal, “there came a time when it was pretty obvious that a lot of the biggest innovation in AI was going to happen in industry”, Pineau says. So, in 2017, she joined Meta, then known as Facebook, to open its Fair lab in Canada, alongside existing labs in Paris and New York.

“I jumped in with curiosity as to what you can do in research when you have industry-level investment,” she says. “Meta was the only [company] that had a commitment to open science and open research. And so I didn’t bother interviewing anywhere else.”

Pineau did not set out to become a leader, she says, and acknowledges that “it is a little bit difficult to find people who both have the ability to do research and an appetite to have more management responsibility”. But she is now part of a leadership team that is more female than male.

“I didn’t hire most of these people into [Fair], but I created opportunities and supported their roles,” she says. Then she adds: “It is also easier as a woman leader to draw other women in . . . so that’s also another important part of the equation.”

During her university career, Pineau was one of 15 women in a cohort of 75 at Waterloo, and one of six in 20 in robotics at Carnegie Mellon. Today, the number of women joining the tech sector remains low and, in 2019, professional services firm Accenture found that half were leaving by age 35.

“Anything I say, you have to put through the filter of the ‘survivor bias’ . . . I’m still in this field probably because I actually had a number of quite positive experiences,” she says. Pineau cites the support she has had, over time, from a number of generous colleagues and contacts. But she was also “quite aware” of the imbalance of men and women in the field.

Pineau believes that academia has made “remarkable strides”, however, and that measures such as allowing students to specialise or change their subject during their course could help to improve gender balance.

She is still affiliated with McGill University and lives in Montreal, where playing music, running and spending time with friends outside the tech world “gives me a very different perspective” from counterparts in California.

That bit of distance, alongside a long history of working in AI, has given her valuable perspectives as the world grapples with hype and concern over the potential of this powerful technology.

“No one can really predict the future, but there is definitely a moment [happening now],” she argues. “Since the launch of ChatGPT, for the first time, every day, anyone, anywhere, can experience AI in a completely different way. Previous to that . . . it was sort of invisible.”

For Fair, there are still “a ton of open research problems” and, more broadly at Meta, questions on the product and regulation side.

“There is a lot of scrutiny but also a lot of really good, important questions that we need to decide as a society,” Pineau says. “How are we going to position ourselves? There’s no right or wrong answer. What are the ways in which we want to work together to create an environment that favours innovation, that keeps people safe, and respects the work of creators?

“We’ve got to find the right ways to weave all of that together.”