American magazine MIT Technology Review unveiled its prestigious list of innovative people whose work has the potential to change the world on Tuesday morning. Adopted Montrealer Sasha Luccioni is among the 35 selected personalities.
It is a great recognition to be included in this list. I don’t know about other professions, but in research we have a bit of imposter syndrome. We always ask ourselves: “If my article is published, will people recognize me?” explains Sasha Luccioni, climate manager at Hugging Face.
How did she manage to attract the attention of MIT Technology Review? With his research on the ecological footprint of machine learning, a branch of artificial intelligence (AI).
His methodology suggests casting a wider net than has already been done in the past, for example by addressing the environmental costs of training an AI model as well as its entire lifecycle.
We are not aware of that [des systèmes tel ChatGPT] It consumes so much energy because it is so dematerialized and democratized.
Sasha Luccioni was born in Ukraine and studied computational linguistics and human cognition in Paris, South Korea and other places around the world. But it was only in Montreal that she decided to settle down.
Now to my Twitter bio [aujourd’hui appelé X] Identifies with Montreal, I can’t move anymore!, she jokes.
Initially drawn to a multidisciplinary doctoral program at the University of Quebec in Montreal (UQAM) that combined computer science, AI and human cognition, she decided to stay in the metropolis because of its healthy climate.
This was at the time when deep learning was on the rise, and Montreal was one of the best places to be at the time, she notes, pointing to the important contribution of Yoshua Bengio, a Montreal AI luminary .
In Montreal we created something caring and harmonious that I love. I could see that it was more stressful and competitive elsewhere in the world, especially in San Francisco.
Sasha Luccioni specialized in the environment on a whim. When she entered the job market, she quickly discovered that she lacked social awareness.
I taught my children how to make compost from a young age: my daughters already knew where to put the apple peel when they were 2 or 3 years old, she says, but she wanted to fight harder against the climate changes.
Her husband warned her: why not combine business with pleasure, combine social consciousness and work? After all, she didn’t study for nothing.
At the time, the Quebec Institute of Artificial Intelligence (Mila Quebec), founded by Yoshua Bengio, was looking for postdocs who wanted to use AI as a tool to increase environmental awareness. She jumped on board, most notably developing This Climate Does Not Exist, a visualization tool that uses deep learning to generate climate empathy.
Open in full screen mode
Yoshua Bengio received the 2019 Turing Prize from the Association for Computing Machinery (ACM), considered the Nobel Prize in computer science.
Photo: (Christinne Muschi/The Canadian Press)
To this day, she says, she works as a climate manager for Hugging Face, a young start-up that is taking action against fashion regarding corporate secrets and the lack of transparency of platforms. Its customers include major technology companies, including Facebook.
It’s a hub: people share AI models and data sets. It helps create more robust systems and democratizes responsible AI. Because our platform has many transparency mechanisms and we also offer responsible AI licenses with open access, she explains.
9 questions for Sasha Luccioni
Collecting environmental data linked to AI is child’s play?
AI is about the egg and the chicken: the big tech companies – for example Google, Microsoft, Amazon – will provide AI services, but since they are not the ones using them, this is not reflected in their carbon footprint .
The AI falls into a crack [pour ce qui est du calcul de l’empreinte carbone].
The numbers could come from data centers, but they don’t calculate what percentage of machines belong to AI and what percentage serve Netflix instead. One of the reasons mentioned is concern for the confidentiality of the machine.
Generative AI requires the use of tens of thousands of graphics processing units (GPUs), but we do not provide carbon footprint data, such as how much water is required to produce them.
Open in full screen mode
Chipmaker Nvidia has reached a stock market valuation of $1 trillion (over C$1.35 trillion) along with other giants such as Alphabet, Apple and Microsoft.
Photo: Getty Images / Justin Sullivan
The company that makes the most GPUs, Nvidia, hasn’t released any numbers. It outsources manufacturing to other companies and it becomes difficult to keep track and connect the wires.
There is therefore a gap in AI lifecycle analysis that cannot be filled.
Why is it important to obtain this information about AI’s environmental footprint?
We need this data to be able to say that we will prefer this or that technology because, for example, its footprint is smaller.
And the more energy-intensive models are not used to combat climate change.
There’s a race for GPUs right now. We want more and more of it without knowing the cost to the environment. There definitely needs to be more transparency in the supply chain.
Do you have any possible solutions for AI management?
First of all, I think that we should not deploy mass-produced systems without checking their energy efficiency or without carrying out a minimum of evaluation tests.
We could therefore have a rating system that determines, for example, what percentage of the AI under study is racist, sexist, effective and robust, and how it reacts in a different context for which it was developed.
We set scales that state that for a planned use in education or medicine the system must have a rating of more than 90%, but for customer service there are less than restrictions.
For example, Facebook had trained its Galactica model – which was decommissioned after three days – on scientific articles, but when asked questions outside the scope for which the tool was created, it ended up spreading stereotypes and making homophobic comments. This is one of the things we should check before starting a system.
Will we be able to keep up?
Six months ago, things were moving in the right direction, particularly with the Montreal Declaration [dont le but est de poser des principes éthiques et responsables au développement de l’intelligence artificielle].
But given this trend towards generative AI and the acceleration that comes with it, it is very difficult to track this progress. Science cannot keep up with the use of these technologies. And in another six months, the laws written today may not make sense.
I have no solution to this problem other than once again advocating for transparency. You have to show your cards: If you know your cards, you can make more informed and flexible decisions.
Are we demonizing AI too much or, on the contrary, not enough?
Right now it’s moving so quickly and it’s very opaque. We will never truly demystify AI algorithms, we will call them magic.
People are afraid of things they don’t understand. There is an unease, a panic, associated with the way AI is practiced. We don’t realize the extent to which this technology is already ubiquitous: classifying resumes, determining mortgage rates, etc.
Personally, I cannot trust systems that I do not have enough detail about to make an informed decision.
ChatGPT works very well, but has many limitations, especially when it comes to subtle culture-related knowledge. But the more people use it, the more they tend to trust it.
It makes us dependent on these technologies that are not yet mature enough to become dependent on them.
An AI tool that you particularly like?
The Merlin app! You put your phone on the table and it detects the birds flying near your house. You can take it on a journey and learn more about the different species that cross your path.
Start of the widget. Skip widget?
End of widget. Back to the top of the widget?
In addition, at Espace pour la vie (Biodôme, Insectarium, etc.) we use AI to monitor biodiversity. Researchers are working to track down moths. They are even able to identify not only the species but also the individual insect.
Are women attracted to the AI industry?
It’s really difficult to find a place in a field where we’re estimated to be only 11% women.
I am involved with Women in Machine Learning, an NGO dedicated to promoting the status of women in machine learning. We do mentoring, organize events and have a fund to send women to conferences. We are trying to build a community of mutual help, of people who can support each other.
But there are many so-called tech bros. Guys who create a website where you can automatically generate women and mark them as sexy or not sexy, so that an hour later they tell you that you like big tits…
Every week, sites emerge that use AI in sexist or misogynistic ways and objectify women.
In short, feminism in AI still has a long way to go.
What sign would you like to have in the world?
My greatest wish would be to better connect AI with society. What I have observed so far is the gap between the two. When it comes to engineering and mathematics, for example, we often forget one important aspect: the human aspect.
Open in full screen mode
Researcher Sasha Luccioni sees Montreal as a technology hub, particularly for a number of SMEs and large AI companies based in Montreal, but also for government support of this industry with tax credits and research grants.
Photo: Radio-Canada / Ariane Labrèche
I see AI as a technology that exists in a context that cannot function without people. The more contacts we make between people working in this area, the more we will build bridges and integrate diversity into the dialogue.
To achieve this, we need to continually spread the word and raise awareness in society.
Now that you’ve been added to the 35 Under 35 list, what’s next?
With this award, I look forward to meeting people from the world of climate change to discuss issues and projects. In particular, I will be attending conferences at MIT in the spring where we will present our work and exchange ideas.
Because the idea behind this list is to create cross-connections and not to do so in isolation. It’s about building a community, a network, and especially in the area of AI it’s important not to work in silos.
I also plan to give a Ted Talk at the annual TedWomen event in Atlanta, USA in October. The effects of AI on the climate are discussed, but also the way in which AI is changing our society, its risks and consequences.
I’m trying to approach the topic in a more constructive and non-alarmist way because I want people to know what AI is without panicking. The more they understand how it works, what it can and can’t do, the more it will help us move forward.