SUBSCRIBE ON:
Introduction to Guest
Daeil Kim is the technical founder of AI.Reverie, a simulation platform that trains AI to understand the world and make it a better place. Formerly, Daeil was a data scientist at The New York Times, and his research/learnings have been published in several of the top machine learning conferences (NIPS, ICML, AISTATS). Daeil received his PhD in computer science from Brown University, focusing on the development of scalable machine learning algorithms.
AI.Reverie was also ranked by Forbes as one of the top 25 Machine Learning Companies to look out for in 2020.
The Call to Action
At first, Daeil was not quite a technical person. In fact, he studied literature in college. From there, he transitioned into neuroscience – researching schizophrenia and neuro psychiatry, which led Daeil to start thinking more about the brain and fundamental aspects of intelligence. Daeil’s innate curiosity led him to pursue a PhD in machine learning, a transition that happened over a period of five years. After completing his PhD, Daeil wanted to apply machine learning to areas that weren’t being applied to. One of those areas was journalism. Daeil started his career at The New York times as a data scientist, working on solving their problems for over two and a half years.
During Daeil’s time at The New York times, he really wanted to go back to the idea of simulation. And with his advisor being a computer vision professor, he was aware of the fundamental problems they were dealing with. At the time, the big problem was data, which was a laborious process of labeling data. If Daeil was to think about supervised learning, as a way to train the state-of-the-art vision algorithm, the process within itself was inefficient. Daeil was determined to find a better way.
Coming Up with The Idea
Daeil was aware of these particular problems early into his PhD, and synthetic data was not a new concept. People in the academic world were trying to solve this labeling problem in computer vision. Others in the industry were able to create prototypes in academia around trying to use synthetic data to improve the training of computer vision. However, no one actually tried to make it a production system, meaning, how can we solve a lot of the world’s problem with it? This was an obstacle that Daeil was determined to tackle. Even though Daeil loved working at The New York Times, he knew branching out on his own was the best way to go.
Daeil spent several months learning game engine programming and how to build virtual worlds. He also met his co-founder during his stint at The New York Times (they both led international expansion). As a former entrepreneur and CEO of Giga home, Daeil’s co-founder was thrilled about Daeil’s vision. He, like Daeil, also left The New York Times to pursue his dream. Together, the dynamic duo founded AI.Reverie.
Testing the Product & Validating the Market
Daeil always knew there are ways to scale world creation, so the first thing he did was focus on the idea of taking geospatial data and recreating the world from it. From ingesting OpenStreetMap, satellite images, terrain data, for example, he could fuel the process of creating a world on a one-to-one scale in a very large area.
An example is the famous video game Grand Theft Auto. GTA spent over $100 million just making a small part of the world in terms of art, design, etc. in the video game. Daeil imagined something similar but with the mission to solve actual real-world problems in agriculture, retail, and government. Daeil was also able to use machine learning to accelerate the process.
The earliest adapters in AI.Reverie was the government, which surprised Daeil. The government was the one most open to the idea of using synthetic data. At the time, very few people believed in synthetic data as an actual solution to a product. Daeil was just curious to see who was interested in using synthetic data to solve computer vision problems. AI.Reverie learned that there are some interesting government applications around safety and started to understand how they could create the right product market fit. However, the market was not too prepared for AI.Reverie, it was a little too soon, ahead of its time. Companies barely had computer vision teams and were barely trying to understand how to set up those teams. It took a lot of phone calls and figuring out where the opportunities lie for AI.Reverie, all while creating an infrastructure that allows the company to work at scale.
One of the most important problems to solve is how to make synthetic data batter? It’s about laying out the infrastructure to pipeline, while trying to create a business. Daeil mentions that at some point, it’s a matter of getting the right clients, the right partners to work with you, and winning those contracts. However, this was not an easy process. It’s hard for every entrepreneur to find the product market fit, and it’s especially hard when you’re building deep tech.
Using Gaming Techniques & Machine Learning to Develop Virtual Worlds
Generating virtual worlds is an algorithmic process, and there are many things you must create before generating virtual world. However, using gaming technology to develop virtual worlds is different than the typical ways to build video games. The engineering process is very different from what you have to do for computer vision. For example, in video games you’re often focused on narrative structures, on storytelling, on creating an experience that people come back with. There’s a beginning, middle, and end. For computer vision there’s no need for that. What you really want to figure out with synthetic data is how do you create diversity? How do you create as many variations as possible of the vehicle? How do you create as many variations as possible in the world itself and the structure of the world?
AI.Reverie is using machine learning to figure out the right engineering pipelines of using gaming technology moving forwarding.
Challenges Encountered
The two main challenges encountered are: finding the product market fit and teaching/educating people about synthetic data, which can be a difficult concept for one to wrap their head around. Daeil did not think it would be so hard to convince people that synthetic data was the way to go. The main reason this is difficult being there aren’t any examples of synthetic data, or how it could potentially work and be an asset. That evangelism is still quite challenging.
Future Endeavors
AI.Reverie’s goal with synthetic data in general is to solve the world’s problems in multiple verticals, creating technology that’s more horizontal. To continue to create solutions for problems in agriculture, consumer packaged goods, retail, government (to name a few). Computer vision is a main focus because Daeil is thinking in terms of the future of robotics and Artificial Intelligence. You have to solve the perception layer first. If you don’t solve vision, robotics becomes a lot more challenging, for example, if you’re trying to ask a robot to retrieve a beer from your fridge. The robot at least needs to know what the fridge looks like. This is a huge ambition for AI.Reverie.
Advice for Aspiring Entrepreneurs
On a personal level, really think about your convictions and your beliefs in what you are doing and trying to accomplish. Spend time meditating on this because at times you will doubt yourself. You will wonder if it’s all worth it. When you feel this way, always go back to your first principles of why you think your idea or company is important and envision how to make it work in order to keep going. There will be sacrifices to be made, too, including financial sacrifices. You are taking a big risk. At the end of the day, be kind to yourself, and also be kind to others. Really believe in yourself and your company.
Mindfulness Principles
Become aware of you who are as a person. Self-awareness is crucial. What normally ends up happening is your own emotional triggers can become reactive, which is something not many people understand. If you don’t look at your past and your own past traumas, you won’t have the clarity to make the right decisions. It’s key to learn about yourself and to be kind and honest with yourself. Always be willing to do the inner world. For if you’re not personally aware of your triggers, receiving criticism or things not working out can cause you to put yourself down. This can lead to a downward spiral of negativity and self-doubt.
It is also important to note that if you’re not making mistakes, you’re not (in some ways) progressing. It’s ok to make mistakes – learn from them and keep trying to progress. Again, self-awareness is key.
About the Host
Ari Yacobi is a data scientist, a teacher and a storyteller who has spent his career at…Read the Bio