The video game sector has experienced significant growth in recent years. We take stock of some fundamental issues, this time considering a major Montreal example of research and development in artificial intelligence.
Yves Jacquier is Ubisoft’s chief wordsmith. That is to say, he runs La Forge, the research and development laboratory of the video game mega-company with nearly 20,000 employees spread across approximately 45 studios and 30 countries.
The teams of scientists are active in Montreal, where Mr. Jacquier and his HQ are based. But also in Toronto, Bordeaux, Annecy and Shanghai. La Forge brings together around fifty employees permanently supported by around 30 graduate students.
“Our mission is to bridge the gap between academia and the creation of games,” explains the scientist, himself a doctor in particle physics and electronics from CentraleSupélec, an engineering school attached to the University of Paris. -Saclay.
“Researchers come to us to carry out their work and we give them access to everything, exactly as we do for our employees: our data, our experts, our infrastructures, etc. The objective is to start from concepts coming from academia to create prototypes that could possibly help us create better games or better operate existing ones. »
The lab has forged around 80 prototypes. Research also leads to scholarly publications. Since 2017, Smiths has published nearly 70 articles in peer-reviewed journals on topics ranging from character animation and graphic environments to sounds and software. Mr. Jacquier repeats it: his lab is active at the cutting edge of technoscientific trends. “We really focus on what is disruptive. »
Generative AI already old
Smiths therefore did not wait for the very recent popularization of conversational robots like ChatGPT to take an active interest in generative artificial intelligence (AI).
“When we started La Forge seven and a half years ago, it was still a fairly marginal subject,” says the director. AI produced interesting results in a laboratory, but was not yet usable in production. In the long run, we managed to understand the possible applications for our professions and what the limitations of these new means are as well. Because it’s not magic either. »
The very particular nature of the sector, straddling art and technology, also calls for adapted solutions. “We are more than multidisciplinary,” says Mr. Jacquier. We are interdisciplinary. At La Forge, blacksmiths come from all walks of life. When working with AI, we want to make sure we cover the constraints of the work of both artists and programmers, including legal issues and ethical questions. »
Artificial intelligence has concrete repercussions on all professions that create or modify computer data – and therefore, in truth, on the vast majority of links in the production chain of a game. New machines can automatically generate worlds , levels, scenarios; create effects; improve graphics; provide advice to artists as well as programmers or gamers, etc.
When working with AI, we want to make sure to cover the constraints of the work of both artists and programmers, including legal issues and ethical questions.
“But to achieve something that is not either a gadget, a threat or even, in some cases, completely illegal, we have to work and develop our tools ourselves,” warns the chief scientist.
It gives the concrete case of project ideation. Creative teams use mood boards (mood boards), trend and inspiration boards that help define the broad aesthetic lines of a game in development by retrieving reference images from Pinterest or Google. Generative Artificial Intelligence makes it possible to create even better adapted images and produce infinite versions. The proposals can then inspire artists to push their imagination even further.
Experiment responsibly
These new means herald many, many changes in the world of work. Some jobs will inevitably disappear, and others will be transformed. Unions and associations of cultural professionals are calling on governments to protect jobs and copyright in this context of accelerated technological change. So how do Ubisoft employees react to La Forge’s disruptive research?
“I do not pretend to speak for almost 20,000 employees at Ubisoft,” replies the scientific director. “But I remind you that in the last year, we put a lot of effort into listening to the concerns of our people. And when I say concerns, I am talking about opportunities, fears and questions at the same time. »
The Ubisoft-Mila Research Chair has just been created at McGill University to explore the ethical use of AI in the creation of content for games. The company has also put in place a structure to encourage responsible experimentation with these new tools. Research and development are thus framed by what Mr. Jacquier calls “an editorial vision” postulating that generative AI “is there to assist the creator, and not to automate creation”.
This is the case for character motion capture. Standard production mechanics use sensors placed on humans that activate according to the scenario. It is then necessary to clean the data to correct errors – if a sensor falls after an actor rolls, for example –, what professionals call occlusions. This cleaning was done with a mitt for years. La Forge succeeded four years ago in having this tedious task carried out by AI.
“Artist technicians can therefore work on greater added value,” says Mr. Jacquier. For the same initial investment, we become able to deliver data much more quickly which is then used for finishing work. What makes our games great is our people. New tools, new ways of doing things, if we use them well, will continue to bring us a technological advantage and a creative advantage, but on condition that we ensure that the contribution is at the service of human genius and not at its expense. detriment. »
This time, the concrete example comes from the management of toxic slippages in discussions between online players. The sector has set standards to encourage positive and inclusive practices. Ubisoft has also been using AI for a year to try to detect slippage.
The creation of the digital censor requires capturing inappropriate speech, but also taking into account the context of the speech to judge it. In a war game, saying “I’m going to shoot you” or “I’m going to kill you” is still within the norm. A linguist participated in refining the model.
“Commercial tools have between 20 and 40% detection efficiency,” explains the director of La Forge. “Our prototype is almost 83% efficient. On the other hand, a lot of data is required to be able to train it reliably and reliably classify toxicity issues. The results have been published, and we hope that other companies which have the same problems will add a stone to this building. »