Predictions Series 2022: AiThority Interview with Peter Stone, Executive Director at Sony AI
Hi, Peter. Welcome to our Interview Series. Please tell us a little bit about your journey and what inspired you to start at Sony AI.
A pivotal moment in my early career as a Ph.D. student at Carnegie Mellon University was when I saw a demonstration of the first soccer-playing robots in the summer of 1994. They were from Alan Mackworth’s lab at the University of British Columbia, and I became immediately inspired to try to understand the intelligence required to play soccer. At the time, most AI researchers were focused on much more abstract planning tasks, or short-time-duration skills for individual robots. I saw the opportunity to use the game of soccer to, for the first time, investigate new methods for enabling collaborative (with teammates) and adversarial (against opponents) multi-robot planning in relatively complex domains. It so happened that one of the few other people in the world at the time who was thinking about robot soccer as a challenge domain for AI was Hiroaki Kitano at Sony (currently the CTO of Sony and CEO of Sony AI). So we quickly became connected and as a result, I have always thought of Sony as a visionary player in the AI space.
Shortly thereafter, Sony introduced the Aibo robot, which reinforced my view of the company as being well ahead of its time. I used the Aibo as my first robot research platform when I joined The University of Texas at Austin as a Professor in 2002, and as a result interacted extensively with Masahiro Fujita, who was a leader of the Aibo project at Sony.
Fast forwarding to 2015, I co-founded a startup company, called Cogitai, that focused on continual learning. Due in large part to my prior relationships with people at Sony, Sony became an early investor, and eventually became the core of Sony AI when it was founded in 2020. Overall, I’ve always found Sony to be a company focused on inspiring, future-looking problems, and I was honored to play a part in establishing Sony AI.
What are the types of AI does Sony AI focus on?
The goal of Sony AI is to unleash human imagination and creativity with AI. We focus on very ambitious, large-scale projects pertaining to such topics. Our flagship projects are AI for gaming, AI for imaging and sensing, AI for gastronomy, and AI ethics. We have particular expertise in robotics and machine learning – especially reinforcement learning. But, we are interested in all forms of AI as they pertain to enhancing human creativity.
Please tell us more about your AI and Deep Learning projects at Sony AI. What are these capabilities directed at (industry, business processes, etc)?
Our biggest success so far has been Gran Turismo Sophy (“GT Sophy”), an AI agent trained with end-to-end deep reinforcement learning in collaboration with Polyphony Digital Inc. (PDI) and Sony Interactive Entertainment. GT Sophy is able to outrace the very best human drivers at the PlayStation game, Gran Turismo. When we started the project, it wasn’t clear whether this would be possible to achieve, and if so, many people did not think that reinforcement learning would be the most effective approach. This result is very interesting from an academic perspective, and it has been garnering a lot of attention in the field. But of course is it very relevant to the Gran Turismo game itself, and we are now working very closely with our colleagues at PDI to make GT Sophy available to players in the future as a part of Gran Turismo 7.
From your point of view, how has deep learning evolved over the last 2-3 years?
Deep learning has solidified its place as one of the most important and powerful tools in the AI and Machine learning toolbox, especially for problems for which there is lots of training data available. It is being deployed in more and more products, especially related to computer vision and natural language processing. Meanwhile, there continues to be research interest in exploring the power of large language models (and other foundation models), and generative models such as generative adversarial networks, and we continue to see more and more interesting success stories for deep reinforcement learning, such as GT Sophy.
What is on the horizon for deep learning in the next few years to come? Is there anything that you believe we’ll see next year?
I expect we will continue to see more and more creative uses for large language models in various application areas, including robotics. I hope we’ll see more exciting successes. And I also hope that we will begin to get a better handle on their limitations. I don’t expect that deep learning will solve all problems in AI and ML, so it is important that the field continues exploring alternative methods, and also exploring methods for combining deep learning with other AI paradigms. For example, there is an exciting thread of research on “neurosymbolic” methods that is investigating ways to effectively merge symbolic AI with deep learning.
What is the current state of robotics?
That’s a very broad question! The field of robotics spans everything from development of new sensor and actuators to intelligent interaction with a dynamically changing, human-populated world. Robots remain most widely deployed in controlled environments, such as factories and warehouses, where they can be separated from people as they perform the exact same task repeatedly. In more open and uncontrolled environments, robot navigation is becoming fairly reliable (though still far from being a completely solved problem), while robust robot manipulation remains quite challenging, especially when it comes to deformable objects.
I am currently serving as the Director of Texas Robotics at the University of Texas at Austin, where we are focusing especially on three areas. First, we are interested in long-term autonomy, which is the quest to enable robots to remain operational without human intervention in complex, open environments (such as homes, office buildings, and university campuses), for long periods of time (hours or days), while performing useful service tasks. Second, we are pursuing advances in human-robot interaction, which is a necessary for long-term autonomy in many domains. Human-robot interaction includes everything from medical assistive and rehabilitation robots to enabling robots to converse with people and navigate effectively through crowds. Third, we are pursuing a variety of approaches to robotic manipulation, including of deformable objects. To me, these are among the most exciting current directions in robotics.
Thank you, Peter! That was fun and we hope to see you back on AiThority.com soon.
[To share your insights with us, please write to firstname.lastname@example.org]
Peter is the Executive Director of Sony AI America. He is also the founder and director of the Learning Agents Research Group (LARG) within the Artificial Intelligence Laboratory in the Department of Computer Science at The University of Texas at Austin, as well as associate department chair and Director of Texas Robotics. In 2013 he was awarded the University of Texas System Regents’ Outstanding Teaching Award and in 2014 he was inducted into the UT Austin Academy of Distinguished Teachers, earning him the title of University Distinguished Teaching Professor. Professor Stone’s research interests in Artificial Intelligence include machine learning (especially reinforcement learning), multiagent systems, and robotics.
As a wholly owned subsidiary of Sony Group Corporation, Sony AI was established in April 2020 to accelerate the fundamental research and development of AI and enhance human imagination and creativity, particularly in the realm of entertainment.
We believe in AI that empowers the imagination and creativity of artists, makers and creators around the world. Our aim is to advance AI so that it augments – and works in harmony with – humans to benefit society.