This post is part of our new Future of Artificial Intelligence series which interviews the leading founders and executives who are on the front lines of the industry to get a better understanding of what problems the industry is facing, what trends are taking place, and what the future looks like.
The following is an interview we recently had with Jackie Lee, PhD, Data Scientist at Virta Health
1. What’s the history of Virta? Where and how did you begin?
JL: The idea for Virta Health started with our CEO and co-founder, Sami Inkinen, who discovered that he had prediabetes in the same year that he won the triathlon world championships in his age group. For someone who was fit and had done all the “right” things to be healthy, this came as quite a surprise. This set off a period of discovery for Sami, which led him to our other co-founders Drs. Steve Phinney and Jeff Volek, who had a combined 70 years of research on metabolic health and insulin resistance. It was clear to Sami, Steve, and Jeff that type 2 diabetes could be reversed. The question was, how do you scale it?
2. What specific problem does Virta solve? How do you solve it?
JL: Virta helps people reverse type 2 diabetes—without drugs and without surgery. To do this you have to start with the right science. Type 2 diabetes is considered chronic and progressive, but we actually repair our patients’ metabolic health using a nutritional intervention focused on carbohydrate intake and nutritional ketosis, changing the paradigm from management to reversal. Scaling our solution is where technology plays a role.
We had to reinvent the diabetes care model from periodic in-person physician visits to on-demand access to physicians and health coaches through a mobile device. We use machine learning and artificial intelligence to individualize treatment and exponentially increase our ability to scale.
3. What’s the future of artificial intelligence?
Prediction #1: Tighter AI integration into decision-making. If we’re talking about applications of AI for the next five years, we’ll probably see more and more intelligent systems integrated into people’s lives to help with decision-making. The scenarios can range from reminding someone to bring an umbrella for the rain to detecting which patients are at higher risk of breast cancer. The key insight is that machines are there to provide assistance to humans instead of replacing them. As humans, we should be thinking about what new values we can create if machines can do all the repetitive and boring work for us.
Prediction #2: AI systems that learn from their environment. Further down the road, we should see AI systems that can learn actively when feedback is provided. For example, if a user says “it’s warm here”, current home assistants probably won’t take any actions based on that comment. However, if the user then turns on the AC manually (a feedback signal to the assistant), over time, the AI system should learn and will turn on the AC next time when someone says “it’s warm here.”
Prediction #3: Unsupervised learning. Machines are very good at learning from structured guidance such as labeled data. For example, by giving computers thousands of images (data) and telling them whether each image contains a cat or not (labels), computers can learn to identify cats pretty well. However, when such labels do not exist, which is the case for most of the data we have, it is still unclear what and how computers can learn. This type of learning is called unsupervised learning, which will probably attract many researchers’ interests for the next decade.
4. What are the top 3 technology trends you’re seeing in the artificial intelligence industry?
Trend #1: Multidisciplinary team. This is not a technical trend, but a practical and important one. In order to make a real impact, computer scientists must collaborate with experts in the domain for which the intelligent systems are built. At Virta, we take this multidisciplinary approach and have data scientists, health coaches, and physicians work together to come up with a solution that can help scale our patient care.
Trend #2: Model Interpretability. To make AI usable, it has to be interpretable. A common critique on deep learning, a successful modeling method, is that it acts like a black box. To inform the users why the models behave the way they do and hence increase machine’s accountability, we need to make machine learning models more transparent. Interpretable machine learning has attracted lots of discussions recently and will continue to be an active research area for the near future.
Trend #3: Reinforcement learning. There have been more and more attempts on modeling problems with the reinforcement learning framework, in which the machine is allowed to interact with its environment and receive feedback to adjust its future behavior.
A classic application of reinforcement learning is to train agents that can play games such as AlphaGo. Recently, reinforcement learning was successfully applied to other tasks such as information extraction and training a dialogue system. It's exciting to see more and more creative ways of applying reinforcement learning. At Virta, we’re also considering using reinforcement learning to make treatment plan suggestions.
5. Why is the AI industry ripe for disruption?
JL: Data is the key. Currently, powerful machine learning models rely on good datasets. We’re living in an era where there’s no lack of data. If you have a good dataset for your problem, chances are that you can build a model to automate the solution. We’ve seen AI already disrupting some domains where there is abundant data such as machine translation and image recognition.
About Jackie Lee
Jackie Lee is a Data Scientist at Virta Health, where she builds machine learning systems to help improve patient outcomes and health coach efficiency. Before her career in building ML-driven systems at startups, she spent many years doing research in the fields of natural language processing, speech recognition, and cognitive science at the Computer Science and Artificial Intelligence Lab (CSAIL) of MIT, where she earned her Ph.D. Her work has been published at top-tier conferences and journals such as ACL, EMNLP, and TACL.