Foundational research in the age of AI

Éric Moulines is one of the world’s leading researchers in computational statistics and machine learning and he is taking on a new role as Dean of the Computing and Mathematical Science Division at MBZUAI at a critical moment, when the University is expanding and artificial intelligence’s (AI) place in society is rapidly changing.

Moulines’ research focuses on the theoretical foundations of machine learning and probabilistic computation – work that spans methods such as high-dimensional Monte Carlo sampling, and stochastic optimization. Much of his work aims to solve challenges related to computational statistics in an era of ever-larger datasets.

Moulines originally joined MBZUAI four years ago, splitting his time between Abu Dhabi and France. Over that period, he has witnessed the University evolve as it has added new faculty, expanded its graduate cohorts, and launched an undergraduate program. “Our faculty is top tier, we have attracted very good students, and the changes to the University have been massive,” he says.

His hope for the Computing and Mathematical Science Division is that it will soon become one of the world leaders in computer science and math. It’s already well on its way to achieving that status. “MBZUAI was one of the top institutional contributors to last year’s NeurIPS conference,” he says, adding that it surpassed many other top-tier global institutions. “In the coming years, we will continue to increase our impact.”

Why MBZUAI?

Moulines sees significant benefits to pursuing research at a new and growing institution. Unlike a more established, traditional university, MBZUAI provides an energetic environment for faculty, students, and postdocs – giving them the opportunity to grow and quickly advance their research.

MBZUAI faculty often have more support in terms of doctoral students and postdocs than they would have at other institutions, allowing them to pursue what Moulines calls “curiosity-driven research.” MBZUAI also provides student researchers with important advantages, such as access to compute that’s greater than what is often available at other schools.

These benefits go beyond the University itself. The effort to build a great research university in MBZUAI is an important part of the larger project to expand the science and technology ecosystem in the Emirates and the region. Moulines notes that universities in the Bay Area played a foundational role in the establishment of Silicon Valley. Researchers at those institutions made important discoveries and developed new technologies that they translated to startups. Investment followed the critical mass of brainpower, resulting in a center of innovation that has changed the world. “It’s clear that Stanford and Berkeley created Silicon Valley and generated the right signals for companies to be founded there. We are working to create a similar kind of momentum,” he says.

Education in the age of AI

The current revolution in AI is having a major impact on many different fields, including education. And many educators are rethinking how they will teach students following these changes.

Large language models have found their way into classrooms and labs. While these systems can be valuable tools to assist with coding and math, Moulines says that they shouldn’t be seen as a replacement for students’ learning these important skills. That said, they can’t be ignored by educators either.

The answer lies in using AI for tasks that it is truly good at. Moulines points to the tedious work that comes along with research, such as checking the detailed, step-by-step reasoning of a complex mathematical proof, or finding a bug in a section of code. That kind of work is important but it’s not where original thinking happens. AI can take on these tasks, freeing students and researchers to focus on the more creative work of asking hard questions and generating new ideas.

“I strongly believe that we need to teach with AI, but at the same time, students need to get basic knowledge of math, computer science, formal languages, and other disciplines,” Moulines says. Knowing the fundamentals provides students with a deep understanding of basic principles of science and technology and can help them drive their research forward faster than relying on AI ever would.

The role of the university in AI research

In an era where frontier AI labs have made major advancements in the performance of large-language models and have built products that tens of millions of people engage with every day, AI research conducted at universities remains extremely valuable. And academic research should continue to follow a path that is distinct from the work being done by frontier labs, says Moulines.

The potential directions of AI research are extremely broad and can’t be reduced solely to the type of research that is being conducted at frontier labs. These labs are working on solving a class of important problems, Moulines says, but there are many other challenging and important questions that need to be pursued. “The work we do at MBZUAI is not only of commercial value, but also relates to topics that are often overlooked by large AI companies, such as cultural alignment and under-resourced languages, including Arabic,” he says. Universities offer an environment in which researchers can think creatively about the directions they want to take their work in, and provide the time, space, and resources to pursue them.

As an analogy, Moulines refers to the aerospace industry, which is led by giant companies such as Boeing and Airbus who build some of the world’s most advanced air- and spacecraft. But just because these companies are at the forefront of what is possible today, it doesn’t mean that students should stop studying computational fluid dynamics or carrying out basic research in the field. The same is true for AI.

“Frontier labs may be building the best AI models of today,” Moulines says. “But at a university such as MBZUAI, we are building the new fundamentals of what can be done with AI in the future.”