[LUM#22] “Everyone has their own definition of artificial intelligence!”
While AI is increasingly present in our lives and imaginations, raising a host of legitimate questions and fantasies, its technological reality remains opaque to many of us. Neural networks, deep learning, algorithms... A brief review of concepts with Anne Laurent, director ofthe Montpellier Institute of Data Science and vice president for open science and research data at UM.
Technically, how could we describe how AI works?
The generic concept of AI refers to a machine's ability to replicate our cognitive abilities (reasoning, learning, recognizing, creating, etc.). But we often focus on learning. The idea is to teach a machine a concept by training it to distinguish between situations based on examples, as humans do. There are many learning methods. The input learning data, or examples, can be photos of moles, for example. The system must then distinguish whether or not it is melanoma.
But what happens between this entry and this exit?
The most powerful learning methods currently available are based on neural networks with highly complex systems and processes. The basic concepts are similar to how our neural system works. The input signal activates a neuron before being transmitted to the next one, and this happens on several levels, known as layers of neurons. Between the input and output, the system will decide whether or not to transmit to the next neuron, when to transmit, etc., in order to accomplish its task. This is called neural network learning or deep learning. Deep means that there are many layers of neurons and connections that need to be configured. For the machine to perform its task successfully, it must therefore be trained with very large amounts of data.
Where does this data come from?
When we talk about AI in research, we are referring to research data generated by scientific imaging instruments, DNA sequencing, and so on. This data is produced either by the researcher using AI or by other teams who have worked on the subject and agree to share it for training purposes. This is where open science comes into its own.
Many mathematicians and computer scientists are working on AI by creating new algorithms. What is an algorithm and what is its role?
When you want a machine to do something, you have to speak to it in its language, a programming language: Python, Scala, or Java, for example. But you don't code or program without thinking. You start by writing an algorithm, which means writing in a formalized way, but in a language that humans can understand, a conceptual description of the data used and the operations you want the machine to perform, and the conditions under which you want it to perform them.
If we had to define artificial intelligence, what could that definition be?
The term AI was coined in 1956, and since then it has had a history and evolved, with certain sub-themes developing more than others. In my opinion, to define AI, we need to return to the initial vision of its founding fathers, who believed it had the ability to reproduce our cognitive skills: planning, reasoning, deciding, learning, sensing the world, etc. But we must also take into account the recent breakthroughs made by commercial generative AI systems. So for many people, AI equals generative AI; for others, AI equals statistics... Everyone has their own definition of AI!
For several months now, AI news has been dominated by Chat GPT, a generative AI. What is generative AI?
It is a form of AI that can generate content: text, images, videos, etc. It relies on the input it is given, to which it adds everything it has learned in its model. Users can guide it by providing context and a purpose. For example, they might ask it to provide the causes of melanoma, but in a way that is appropriate for teenagers who are going to spend their summer at the beach, or for their parents so that they can better educate their children. Generative AI contextualizes and adapts the discourse to the requested purpose. AI fascinates us because it speaks to all of us.
What other types of AI are present in our daily lives?
Some AI systems are dedicated to tasks known as "classification." I mentioned moles, but I could also have talked about Pl@ntNet, a plant recognition app developed in Montpellier that can classify plants among thousands of existing categories in plant taxonomy. There are also AI systems known as segmenting AI, which are widely used in marketing to categorize customers without preconceived ideas. There are AIs that plan and, for example, help organize schedules by solving very complex combinatorial and constraint problems. Others make recommendations, on online video platforms for example... There are a huge number of subcategories, and all these tasks are becoming increasingly hidden. AI is using more and more models, and sometimes we don't even realize it...
One of the central questions in the development of AI is that of control. Can we verify and explain the work of AI?
This is a very topical issue because we need to act responsibly, and this is now required by the European regulation passed this summer. That said, we need to differentiate between different levels of control. The challenges and methods will differ, for example, depending on whether we want the end user to be able to understand the AI's decision, or whether this explainability is more relevant to experts, for example in a court of law. I always have a little voice in my head telling me not to ask more of AI than we ask of humans. We shouldn't expect it to be error-free, not because it's imperfect, but because life isn't binary.
Bias in artificial intelligence results is often discussed. Can it be limited?
This usually involves representation bias in the input data. If we want to teach AI to recommend study sections to high school students, for example, and we simply reproduce existing statistics, the AI will never or very rarely recommend the math section to young girls. To correct these biases, we need to increase the amount of data to give the machine data where there are as many girls as boys. This is also why France wants to create its own AI models, so as not to be subject to cultural biases that are not our own.
So we can't feed raw databases to the machine; we have to process them first?
Yes, and that's just the tip of the iceberg. It's a job that isn't talked about much, but it's very tedious and meticulous, and it makes the data as usable as possible. That's one of the missions of the ISDM, which I head up.
What exactly is the role of the ISDM?
The ISDM, Montpellier Institute of Data Science, supports researchers' activities in terms of research data management and processing. It provides tools, infrastructure, training, and, through its data clinic, advice and expertise. We also organize the AI Market, a university initiative to bring together stakeholders in AI and data, and we do a lot of work on data storage and security. In short, we are a gateway for anyone who wants to jump on the AI bandwagon, and they should...
You are also Vice President for Open Science. What are the challenges facing the development of artificial intelligence?
Open science is the fuel for this learning, but sharing does not mean open bar! Opening up to everyone is one way of sharing, but it is not the only one. Conditions and restrictions can be imposed, and consideration given to security, protection, and the promotion of scientific heritage.
Is AI essential for science today?
Yes, absolutely. AI will transform the research profession by increasing our ability to produce state-of-the-art work, to verify, structure, or challenge our ideas, etc. It will also speed up the preprocessing of our data, the management of certain administrative tasks, and our search for funding.
How is the UM positioning itself in response to this development?
Everywhere! The UM has ambitious projects. The aim is both to give researchers the ability to take control of all forms of AI and to showcase the power of AI to acculturate and gradually change our practices. All while remaining as ethical as possible, without "leaking" our ideas or destroying the planet. The UM also strongly supports AI research and the development of new algorithms. Cutting-edge methods are being developed in Montpellier. This is the case, for example, with AI methods in healthcare, such as federated AI, in collaboration withInria. There is no huge gap between use and development; the two feed into each other.
We were not included in the final selection for the AI Cluster tender. Is this a hindrance to the development of AI in Montpellier?
No, because the AI cluster dynamic is continuing with great energy. Stakeholders in the socio-economic world have reaffirmed their commitment to continuing this dynamic. The Metropolis is extremely motivated, the Region voted this summer for a very strong AI strategy, and our financial backers are behind us... The next challenge will be to structure this dynamic and bring it to life. We are also fortunate to be working with the University Hospital, which is one of the most dynamic and advanced institutions in the field of AI and health data processing. We have all the pieces of the puzzle in Montpellier, with a very, very favorable set of players that I don't see anywhere else.
Find UM podcasts now available on your favorite platform (Spotify, Deezer, Apple Podcasts, Amazon Music, etc.).