Breaking the Stereotypes in the Development of AI
Looking back on the historical trends, could you chart out a brief overview of the development of AI over the years?
It is interesting to note that the concepts of both video games and machine learning emerged side-by-side, around 70 years ago. If we look back at the history of AI developments, games served as a playground to measure the performance of AI algorithms against real humans. An example would be the chess match between Garry Kasparov and Deep Blue 21 years ago. The modern video games have many features and potential interactions—with vehicles, pedestrians, people, and more—and are systemic, relying on many complex systems that interact with each other, thereby simulating an enticing and true-to-life city or country. This is precisely what we do at Ubisoft, and it has been made possible by advancements in tools such as AI and machine learning.
When we refer to modern AI, we are addressing machine learning or deep learning, which has evolved exponentially over the years. The basic idea of deep learning or machine learning is to use data and find an underlying behavior, mimicking this behavior when new input data sets are shown. It means that these data sets are crucial for AI, whether you produce the information by yourself or already have it at your disposal. At Ubisoft, we have both. We have specific data on animation, textures, and dialogues and also game engines that are excellent simulator for believable worlds. Instead of programming AI with conditions in the traditional way, we train the AI model using rewards and punishments, allowing it to learn the simulation by itself. In the last five years, the improvements in video games have led to increased quality levels that make the development of AI more convenient than implementing it in real life. For the same reason, AI has dramatically improved our production pipelines and by extension our simulations in video games.
From your personal experience, what are some of the prominent challenges in the advancement of AI that developers and manufacturers face today?
To build useful AI and machine learning based systems, you need three different things—data, expertise, and valuable application. The absence of any of these three dimensions poses the most significant impediments in the maturity of AI right now. The first issue is technical; although there have been tremendous improvements in deep learning, the expected accuracy and explainability of the system is a critical aspect that developers need to focus on while building an efficient system. Deep learning based AI can make correlations into complex structures of data, if not yet able to explain or infer causality between elements like the human brain does. Using such statistical AI models on a real life application is a substantial technical challenge to overcome for making it reliable and explainable. The second challenge is in terms of the regulation that comes with the usage of AI. When AI is deployed in self-drive cars or creates art, music, images, models, or can reproduce voices or videos of real people, there arise many ethical and copyright issues. This is an essential factor to take into account because it’s a way to decide the kinds of usage that are acceptable in our societies. Finally, the third challenge is change management and adoption. To be useful, people from the core business and activities need to adjust their tools or processes, requiring a total switch of mindsets. For a programmer, this approach means finding the most relevant data that characterizes an activity or business requirements or resolves hindrances instead of implementing rules based on knowledge. So, change management is a vital factor for injecting any AI-driven application into their business operations.
How do you differentiate augmented and artificial intelligence?
As someone who’s been working in this field for a long time, I feel that AI is not the best-chosen expression. Because it’s misleading to think that those systems will soon be good enough for creativity, judgments, and creating links in terms of causality. From a personal standpoint, let’s talk more about the augmenting intelligence part of a work, and then really start talking about AI, per se. The first and most important application of AI and machine learning is that it can help us automate repetitive tasks. Also, if we’re able to free humans from those tasks along with empowering them creatively through innovative tools and techniques, it will lead to more amazing creations. In terms of creation, it must be the human who must always be at the center of it, and we should work on AI with the objective to augment the creators’ capacity rather than substitute it. Firstly, on an ethical and philosophical standpoint, it’s not the right path; and secondly, on a pragmatic standpoint, the technology is still not matured enough to achieve that.
What are your thoughts on the development of AI in future, and how can it be utilized for the betterment of humanity?
We have many opportunities now and can witness exciting results as long as we’re able to harness the power of AI; for example, the technologies that we use in the gaming industry to create different avatars. As both AI and video game technologies have reached an exciting point of believability, they also find a wide range of applications in other industries such as healthcare. Currently, we are in discussion with the University of Montreal to explore the possibilities for treating schizophrenia by transferring gaming technologies into an innovative application. The developments in AI and machine learning force us to go out of our silos and find new opportunities to create more appealing solutions that can help real-life people. Whether it’s in healthcare or mobility, I see numerous ways to make our world more powerful and accessible for many people.