By Dan McCreary, Distinguished Engineer, Optum
Our CIO recently told me he wants everyone in our division to know at least the basics of AI. And he means everyone, at all job levels, no matter what role they’re in. So I shared with him an idea I’d been toying with: the AI Racing League.
Some people are scared about AI because of what they see in the movies. The AI Racing League will help them learn about the reality of AI in a non-scary way. The goal is to make AI fun!
The first AI Racing League events
In August of this year, Optum Tech University (OTU) invited teachers, community stakeholders, and members of local tech organizations to participate in the first “race.” The OTU set up a second event in September for Optum employees, and the slots filled up in a couple of days. Another event at the end of October is already full, with more planned for the future. We’re even developing a curriculum to bring the AI Racing League to schools.
We want everybody here to understand the potential of what AI can bring to health care. We also know that today’s students are future employees, and we want to get them started early. The only thing keeping us from growing even bigger is finding enough mentors to accommodate everyone who wants to sign up.
Three learning tracks (but one race track)
At the events, stations are set up around a large conference room. Each team gets a preassembled car kit and color-coded instructional cards to guide them through the technologies they’ll be using. The cards’ colors are similar to ski-slope ratings: the green cards have beginner-level information, the blue cards cover intermediate skills, and the black cards are for experienced folks ready for advanced concepts.
Participants learn hands-on about UNIX, SSH, Python, Raspberry Pi and Nvidia Nano boards, electric motors, computer vision, and machine learning. Teams are guided through the stations, with a dedicated mentor to help answer questions. Once the teams make it through all the learning stations with their cars, they end up at the final spot, called the “calibration station.”
How to train your self-driving car
To teach the cars how to drive, the teams drive their cars around the track — 10 times in one direction, and 10 times in the other direction. The camera on each car captures 30 images per second, which is recorded to the on-board flash storage.
After the training runs, the cars are plugged into a hefty GPU server (sitting quietly on the side of the room), and the images are uploaded. This is where the magic happens. The machine-learning model analyzes what it sees in the images, and then loads instructions back onto the car’s computer so it can drive itself.
One of our AI Racing League mentors, Robert Rossmiller, explained what happens:
“We don’t know what it’s actually seeing and how it’s learning,” he said. “We’re not saying, ‘Here’s the yellow line, stay on that line.’ We’re saying, ‘Here’s how to drive, now go learn how to do it. This is what it looks like to go forward. This is what it looks like to turn left.’ We’re giving the car everything and saying, ‘You figure it out.’”
It’s all about real-time predictive analytics. The cars are essentially predicting how best to get around the racetrack. Based on what the camera sees right now (the real-time data inputs), and what it knows from past experience (the training runs and analysis), the on-board model is predicting whether to turn the wheels right or left and at what speed to go down the path (the recommended actions).
What does it have to do with health care?
When a patient walks into a clinic, we want to be able to predict in real time what an appropriate care path would be. In a car, the data is relatively simple. Every pixel in the image is the same distance apart from the others. With patients, the distance between their multiple conditions is not the same. But the algorithms are the same. Health care data is more complicated, but we are applying the same concepts.
The ultimate goal is to recommend the right care path for a patient, personalized to their current health situation, just like the cars are guided around the race track in real time. There are still huge hurdles to overcome, but the AI Racing League is setting the foundation. We’re getting everyone familiar and comfortable with AI concepts, because there’s a lot more for all of us to learn after that.
If you want to become a mentor or learn more about the AI Racing League, contact us at email@example.com.
And to learn more about technology innovation at Optum, visit optum.com/technology.
About the author
Dan works on innovative database architectures including document and graph databases. He started the NoSQL Now! Conference (now called the Database Now! Conferences) and co-authored the book Making Sense of NoSQL. He is involved in the CoderDojo program and is the co-founder of the AI Racing League. You can read Dan McCreary’s full profile on our People page.