Blog: An Interview with Dr. Danny Lange, VP of AI at Unity Technologies
It was while writing my 5th book that Dr. Lange had contacted me through LinkedIn. We had connected through discussions on my previous book on Unity ML-Agents. So I was very much elated with getting assistance from the team at Unity. It was from those suggestions that I feel altered my book for the better and gave me a peek under the covers at Unity. I also felt that it would be fun to share some of those insights with others through my blog. This led to me previously interviewing Dr. Arthur Juliani, head AI researcher at Unity:
Late last year, Dr. Danny Lange, the VP of Unity’s AI/Machine Learning division contacted me through LinkedIn and…medium.com
Continuing with that theme I requested that Dr. Lange also provide me with an interview and he so graciously accepted. If you are not sure who Dr. Danny Lange is, I have provided his bio below.
Who is Dr. Danny Lange
Danny Lange is Vice President of Artificial Intelligence at Unity Technologies where he leads efforts to advance the capabilities of AI through scalable game simulations. Prior to his role at Unity, Danny was the head of Machine Learning at Uber where he lead the development of the company’s Machine Learning platform. Previously, he was General Manager for Machine Learning at Amazon where he managed Amazon’s internal Machine Learning platform as well as launched the first AI product for Amazon Web Services (AWS) known as Amazon Machine Learning. Danny has also lead Machine Learning efforts at Microsoft and started his career building autonomous agents as a Computer Scientist at IBM Research.
Q1: Previously you were at Uber and before that Amazon. What brought you to Unity and game development and how has your experience been thus far?
Danny: I am a visionary person, but the practical application has been my ultimate goal. This is clear from my work at visionary yet game-changing companies such as Uber and Amazon. At Unity I can take it a notch up. Here I live and breathe in the virtual world of a spatial game engine which I often call the Biodome of AI. My abilities are only limited by the computing power of the cloud services I can access. I can push AI to the very limit of its capabilities without injuring people or wasting thousands of physical products. At Unity I can supercharge AI development, not only for my team, but also for millions of researchers and developers across the globe.
Q2: How is the games industry different from your experience developing enterprise software? How are they the same?
Danny: Both industries are extremely customer and human experience centric. Whether it is getting someone safely to their destination, same-day delivery of a package to their doorstep, or providing them with a truly enjoyable gaming experience, you have to put the customer first. The fantastic advantage of gaming is that we can manage human experiences in real-time. It is here that AI’s ability to operate at extreme scale and at the speed of light creates new and previously unimaginable opportunities to thrill.
Q4: You mention in your talks that Reinforcement Learning and ML-Agents are not just for games. Can you discuss other industries, perhaps such as Uber, that could benefit in using Reinforcement Learning?
Danny: My vision is to simulate the world. Through large-scale simulations we can surpass the experience space of the physical world and make products that are much smarter than the ones we are surrounding ourselves with today. New products that have experienced everything and seen every possible challenge in their environment before they have ever materialized. Reinforcement learning will in this way bring intelligence to everyday products whether that is your vacuum cleaner, door bell, or autonomous vehicles.
Q5: You recently extended entries to the first phase of the Obstacle Tower Challenge by a month. What is the feedback you are getting from those that want to enter the challenge?
The Obstacle Tower Challenge has been extremely well received. In the first round of the challenge we saw over 2,000 submissions with close to 60 teams reaching level five or higher. It is a tough challenge, at least for deep reinforcement learning systems. It features a combination of visual, physical, and cognitive challenges that makes it nontrivial for most common algorithms to get very far. Yet teams are making headway with one team getting as far as to the 16th level. I hope that we will help furthering the state of the art of reinforcement learning through this challenge.
Q6: What comes after the Obstacle Tower Challenge? Are you considering new and different challenges researchers/developers will be able tackle?
We spent almost a year designing and implementing the obstacle tower; the first video game designed just for computers to play it. It has been a big investment and we plan to open-source it with the objective of providing the global AI community with an elaborate yet extensive training platform for the years to come.
Q7: Do you think a fully self-reasoning general artificial intelligence will be discovered in the next 5, 10 or 15 years and why?
I would like to avoid the discussion of “when” we will reach the point of AGI and instead turn the focus on the journey there. On the road to AGI we will invent new technologies and learn new techniques that will be a great assistance to humanity. We are going to need all the help of these tools and skills to fight off the very real and immediate existential threat of Climate Change. This is a much greater threat to humanity that any singularity that may or may not happen.
Q8: What can you tell us about what to expect this year from Unity?
We will continue with the second round of the Obstacle Tower Challenge over the summer and in the fall Unity will announce an entire range of new AI products for game development and large-scale simulations for industrial applications such as synthetic data generation, robotics, and autonomous vehicle training. Our world famous ML-Agents Toolkit will keep evolving and you will see us put a toe in the water with multi-agent systems.
Q9: As someone who has been doing AI for a while now, what can you tell us about the growth in the industry? Is it what you expected or is it exceeding your previous expectations?
The growth of the industry keeps — and I am repeating — keeps exceeding my wildest expectations. Major innovations are happening at breakneck speed. And we are just at the beginning. CNN, RNN, and GAN are just some of the basic building blocks of AI. I like to compare these network architectures to IF, FOR, and WHILE statements in structured programming languages. In the world of AI we have barely started writing the programs yet. I have very high expectations to multi model systems such as hierarchical and multi-agent systems exploring emergence, language, collaboration, and anticipation. It is going to be so exciting times.
I could likely go on with at least 20 more questions but I need to stop it there. Thank you so much for your time Danny and for agreeing to do this.
It is indeed a very exciting and dangerous time to live in. AI will certainly shape our continued existence on this planet in the coming years. This will require us to muster more self aware visionaries and leaders that can guide us through this age of change. I certainly see Dr. Danny Lange as one of those visionaries and leaders that understands the need for balance in our changing existence.