In shorts and sandals, participants sit at picnic tables that could use a good coat of paint. Many have wet hair after swimming in the Caribbean Sea.
PHOTO PHILIPPE MERCURE, THE PRESS
Around 6 p.m., everyone naturally converges on the beach to watch the sunset, a beer or a glass of rum punch in hand.
Around, night has fallen. The sound of insects adds to that of the waves rolling on the beach. A breeze stirs the coconut palms, driving away some of the heat that has accumulated during the day. Nearby, something is stirring in the groves – perhaps the monkeys who raided lunch, stealing two bananas.
PHOTO PHILIPPE MERCURE, THE PRESS
A monkey stole a banana during a lunch raid.
We are here at McGill University. This is not a joke. More precisely at the Bellairs Institute in Barbados, a place with a fascinating history (see capsule).
PHOTO PHILIPPE MERCURE, THE PRESS
An evening workshop
As for the seminar I am attending, it is unlike any other.
The thirty participants who take part have been handpicked and carefully chosen to generate a clash of ideas.
Half of them come from companies active in artificial intelligence, such as OpenAI, Google DeepMind, Microsoft or ServiceNow. The others occupy prestigious positions in the academic world: Cambridge, Harvard, Carnegie Mellon, Mila (Montreal), Vector Institute (Toronto).
PHOTO PHILIPPE MERCURE, THE PRESS
Coming from companies at the cutting edge of artificial intelligence and the academic world, around thirty of them are participating in the seminar.
They are computer scientists, mathematicians, neuroscientists, lawyers, economists, sociologists. Their countries of origin: the United States, Great Britain, Croatia, India, Poland, Italy, France, Israel, China, Canada…
For artificial intelligence at the service of humans
Their mission: to think about the risks of what we call major language models. We are talking about these conversational robots like ChatGPT, capable of assimilating practically all of what humans have produced as knowledge, of digesting this knowledge, then of using it to dialogue with humans with sometimes astonishing results. precision, sometimes frankly eccentric.
A single rule was imposed on researchers: emphasize the “immediate, predictable and catastrophic” risks of language models.
Exit, therefore, the scenarios where artificial intelligence one day acquires a will of its own and decides to exterminate humans. It’s not that the organizers consider it impossible. They simply want to direct thoughts towards more concrete and pressing problems.
In front of the group, lit by a few lamps attached to the roof overlooking the picnic tables, Sylvie Delacroix speaks about the impact of major language models. This Franco-Belgian researcher is professor of law and ethics at King’s College London, England.
PHOTO TAKEN FROM SYLVIE DELACROIX’S WEBSITE
Sylvie Delacroix, professor of law and ethics at King’s College London
Language is not just a way of describing the world. It’s also a way of building it. Language shapes reality.
Sylvie Delacroix, professor of law and ethics at King’s College London
The researcher invites participants to suggest examples of this.
“A religious leader who calls for war against the infidels,” suggests Nicolas Chapados, vice-president of research at ServiceNow in Montreal.
Sylvie Delacroix approves. His point: by using language, conversational robots like ChatGPT (OpenAI), Llama (Meta) or Gemini (Google) are starting to shape the world. For better and for worse.
Since the public unveiling of ChatGPT3 in November 2022, the world is indeed no longer quite the same. For the first time, masses of humans interact with a machine that maps their thought processes. A machine capable of summarizing an entire conference in five points. Or transform a large document into a PowerPoint presentation.
A machine which is also fundamentally imperfect and whose operation remains a black box even in the eyes of its designers. The interactions of millions of humans with this robot launched a gigantic scientific and social experiment whose consequences are still poorly understood.
PHOTO PHILIPPE MERCURE, THE PRESS
In the center, organizer Denis Thérien on an Adirondack chair
In the very center of the assembly, reclining on an Adirondack chair, sits the architect of this seminar: the Quebec researcher Denis Thérien.
PHOTO PHILIPPE MERCURE, THE PRESS
Denis Thérien, researcher at ServiceNow and organizer of the seminar
Wearing sandals and a backwards cap from which white hair escapes, this professor emeritus of McGill University displays the rebellious demeanor of an adolescent. After a stint at the defunct Quebec company Element AI, Denis Thérien is now a researcher at the American multinational ServiceNow.
It’s the 35e seminar he organized at the Bellairs Institute for 35 years.
“This place completely changed my scientific life. I became the most connected guy in the world thanks to that,” he says, waving around the place.
Forget luxury right away. Although Bellairs is nestled in paradise, the place feels more like a scout camp than a five-star hotel. Participants often sleep two by two in spartan rooms without private toilets and air conditioning.
“You need a certain form of asceticism to stay here,” admits Denis Thérien.
PHOTO PHILIPPE MERCURE, THE PRESS
Participants in the artificial intelligence seminar enjoy the beach at the Bellairs Institute of McGill University.
But at the end of the property, there is the selling point that convinces the greatest minds on the planet to stay there: a golden sand beach overlooking turquoise waters filled with tropical fish.
Between mathematics and philosophy
Throughout the seminar, I saw the magic of Bellairs working. After two days, it already feels like we’ve been here much longer. Habits are established, connections are forged.
PHOTO PHILIPPE MERCURE, THE PRESS
Ana Marasović, Siva Reddy and Gillian Hadfield chat in front of a sunset.
The schedule only provides for one workshop in the morning and another in the evening. In the afternoon, participants continue discussions in small groups or put on a mask and snorkel to explore the coral reefs.
Around 6 p.m., everyone naturally converges on the beach to watch the sunset, beer or glass of punch in hand.
PHOTO PHILIPPE MERCURE, THE PRESS
Hattie Zhou, Ryan Lowe, Neil Lawrence and Siva Reddy in discussion
In a regular conference that has hundreds of attendees, you’re lucky if you have a 10-minute conversation with someone. Here, we can carry on conversations over several days. It takes things to another level.
Hattie Zhou, doctoral student at Mila
Young researcher Hattie Zhou is pursuing a doctorate at Mila in Montreal, after working for Uber in San Francisco.
These discussions can switch from technical aspects to philosophical questions in seconds. At one point, participants analyze mathematical methods for increasing the security of language models. The next moment, they wonder about the impact that artificial intelligence will have on human identity.
After five days of discussions, the moment of truth has come: we must agree on the main conclusions of the exercise. And we feel that these are deeper questions that are in the minds of the participants.
“We need to democratize these technologies,” says Neil Lawrence, a computer scientist at the University of Cambridge, in England. There is an urgent need to ensure that lawyers, accountants, teachers, nurses, normal people are empowered by these technologies, instead of having them imposed on them by very large entities who cannot understand the whole picture. . »
“We feel a disconnect,” observes Eszter Vértes, researcher at DeepMind – a British artificial intelligence company bought by Google. We, the machine learning community, are trying to find technical ways to fix things – identifying biases introduced by models and then correcting them, for example. Several participants here reject this approach to reparations. They want to design more holistic processes that will naturally lead to positive results. »
We also emphasize the need to use the power of artificial intelligence to tackle the major problems of humanity: access to health care, the fight against climate change, the reduction of inequalities.
“Currently, we use artificial intelligence in art. But what is the problem to be solved there? “, says Eszter Vértes for example.
Next step: write these conclusions into a position paper that will be submitted to a scholarly journal for publication.
“In several respects, I can say that this is the most interesting seminar that has taken place in Bellairs in 35 years,” says Denis Thérien at the end of the stay to a round of applause. I have the impression that, this time, we have tackled problems that affect everyone, and not just 1% of the population. »
A gift from Commander Bellairs
PHOTO PHILIPPE MERCURE, THE PRESS
The Bellairs Institute of McGill University, Barbados
How the hell did McGill University end up with an institute in Barbados? The story dates back to Commander Carlyon Wilfroy Bellairs, a British naval officer elected to the United Kingdom Parliament in 1915.
Upon his retirement, Commander Bellairs settled on the west coast of Barbados. In 1954, a year before his death, he bequeathed his property to McGill.
Why did you choose the Quebec university? One theory (unverified) is that Commander Bellairs had a grudge against the British who had driven Winston Churchill’s conservatives from power in 1945. A donation to Canada would have been a snub to England.
The Bellairs Institute has long been a center for marine biology research. In the mid-1990s, McGill considered selling it. Professor Denis Thérien, who has already started to organize seminars there, then intervenes and promises to make it profitable by increasing the number of scientific meetings.
Today, seminars are held there from January to May. McGill agriculture students also spend a session there in the fall. Very recently, Bellairs set up a scientific committee to evaluate and choose seminars.