A sit down with Danny Lange, creator of ML systems at some of the world’s leading tech companies
A sit down with Danny Lange — VP of AI and machine learning at Unity Technologies, and formerly of Uber, Amazon and Microsoft. Danny is a computer scientists and researcher with a passion for teaching machines to talk…properly…with a little more charisma than Siri, and few fewer memes than Alexa.
1. Danny, how did you became the person heading up the development of AI at some of the most prestigious tech companies of our time?
From the very beginning, I have been guided by two core principles. The first is that autonomous systems are needed to successfully address increasingly complex system challenges, and the second is that real impact is achieved through the enablement of broad developer adoption. Let me give you some concrete examples of that from my past. I spent years in two startups creating voice-enabled virtual assistants just like today’s Siri and Alexa. The vision was to achieve dynamic dialog generation driven by the feedback between with the human user and the computer. With a developer friendly and service-oriented architecture we were ahead of ourselves and our time. Siri and Alexa are not even close today. While at Microsoft the Big Data revolution finally caught up with me. Suddenly, over a very short period of time, data became a great enabler. My team created desktop machine learning tools and later on cloud-based tools that would help all the developers of the company to benefit from joining their data with great algorithms. Scale and performance became a big deal and we jumped into Hadoop and deployed GPUs for Deep Learning. Yes, Deep Learning abruptly entered our vocabulary. I decided to jump ship and joined Amazon since there was an organization that Jeff Bezos had primed for AI automation. I rolled out a company-wide ML platform that served many teams across the company. I also rolled out the first public ML services on Amazon Web Services. At Uber I repeated the story of rolling out a company-wide ML platform called Michelangelo. It is all about enablement.
2. You’re best known for very complex and large-scale implementations of machine learning. What motivates you to solve those, what are the requirements for data scientists, AI engineers and others applying AI these days?
Machine Learning is a revolutionary technology with the potential to disrupt the way businesses operate. But to achieve true impact the technology needs to be deployed across many parts of the organization and integrated deeply into the business processes. If you move into the AI field today as either a Data Scientist or as a Software Engineer you should be oriented toward automation of ML and AI.
The human-in-the-loop is disappearing as the AI feedback loop is powered by technologies such as reinforcement learning. Their primary role will be towards operationalizing AI and covering ever increasing areas of the business. Only then will we see true impact Machine Learning and AI.
3. As humans we’re often tempted to overestimate the short term. So, how will AI & AR transform our lives & environments in the coming 3–5 years, and how contingent is one on the other for ensuring broad adoption?
The only real intelligence we know is the biological one surrounding us. Nature created intelligence for living organisms to thrive and multiply. Eat, avoid getting eaten, multiply, and beware of physics are the four key principles of the eco-system in which intelligence as we know it has evolved. Now take Unity, your spatial real-time development engine with built-in physics.
In what better place than Virtual Reality (VR) can you imagine exploring Artificial Intelligence. This is your private synthetic Biodome. I firmly believe that it is in this environment we will take the first significant steps toward Artificial General Intelligence.
Now throw Augmented Reality (AR) into this and you can link up the virtual with the real. The result of this will be new and improved learning architectures and methods that will also strongly benefit business applications.
4. On your website, you’ve stated: “In Machine Learning you need to think more like Werner Heisenberg than Isaac Newton.” Was there a defining moment that inspired this analogy and what does it mean for others applying AI today?
I have always been fascinated by the evolution of the science of physics from the absoluteness and determinism of Newton’s age to the abstract and statistical world of Heisenberg. It was a fantastic journey for humanity from late 1700’s to the early 1900’s. I can see clear signs of current software and systems development going through a similar process where conventional code is developed for an ideal, but non-existent world is being displaced by learning-based and adaptive systems much better suited for dealing with a real world with Heisenberg’s uncertainties everywhere. When a company like Amazon tries to understand its several hundreds of millions of customers shopping in a product catalog with over a Billion products, the law of big numbers kick in and statistics become king.
5. We’ve seen you promoting the paradigm: from ‘computer programming’ to ‘computer training’. According to this, what are the biggest changes in computer science AI engineers will be facing in the coming years?
The AI Engineer’s biggest challenge will be to move from the fringes of the organization to becoming mainstream. Here you have to scale and be dependable. New software development and maintenance methods need to be developed as AI is inherently dynamic in its nature. Model performance monitoring and management become central as does data privacy issues. These changes will demand new skillsets to be learned by the AI Engineer with an emphasis on the softer cross organization and cross functional skills. Expectation management will be another challenge and organizations will have outsize expectations to the short-term benefits of AI while underestimating the long-term benefits.
6. You’ve been once quoted that AI capable of killing people can technically already exist. What are the ethical & moral consequences of systems actively influencing our decision-making processes in daily life? Can we ensure to outweigh AI’s potential risks of which some are warning?
AI is a highly adaptive learning technology that puts its owners in a new, precarious role. Let us take a look at conventionally designed systems. In general, these systems demonstrate consistent programmed behavior in accordance with the specifications. Implementation and algorithms are fuelling these systems. When these systems misbehave it is either deficiencies in the specification or bugs in the implementation. Let us now take a look a systems driven by AI. Objectives, rewards, and the AI feedback loop drives the continual learning and never-ending optimization of AI-based systems. An innocent objective such as maximize revenue may allow a highly capable AI learning system to develop deep and hard-to-detect ways to deceive the user into additional spending. When is the line between acceptable business practices crossed? This is first and foremost a question of an AI Ethics Code. At Unity we are pushing ethics guidance to our developers to educate them about the potential pitfalls of letting the AI optimize for just one aspect of the customer interaction.
7. While at Uber, your vision was to bring machine learning to every corner of the company. With the technologies and skill sets available today, what can people making use of AI do to achieve the same?
While my vision at Amazon and Uber was to bring Machine Learning to every corner of the company it has only grown since I joined Unity.
My team’s mission at Unity is to democratize AI. I want to see AI in the hands of more than a few large corporations.
AI is a game-changer and when we unleash the mass of Unity developers on it, we are guaranteed to see innovation and new use-cases that will reach far beyond gaming. The potential of large-scale simulations with a rich set of advanced learning algorithms will create new breakthroughs in robotics, design, industrial applications, and even AI itself. The game is on.
Thanks you Danny for you time & insight!
