One of the main issues with artificial intelligence (AI) is that it requires a lot of data. That, in turn, requires data scientists who know how to work with all of that data. The problem is, data scientists are in very high demand, and there’s an incredible amount of competition for their skills (and not just within health care).
For health-care CIOs who want to use AI to solve real business problems, the lack of internal talent can pose a big challenge. In this article, two data-science experts from Optum® offer a few insights into fixing the data-science talent issue.
Why the right AI skillset is so rare
AI is not a single algorithm or technology, and it’s more than just math or computer programming. AI includes many different kinds of machine learning, natural language processing, image recognition, etc., all using very large data sets. So you need people who can work with the tools and the data, but there’s more to it than technical skill.
When you talk about data science, some people will tell you the lack of skill is in math and computer programming. But that’s not always true. There’s no shortage of quantitative talent. A lot of people can do math and programming. The skills that are lacking are in the methodology, the thinking process and the business context behind what technology you use.
For example, if you’re framing a house, you can use a nail gun and that’s great for that job. But you probably don’t want a nail gun if you’re building a doll house for your daughter. You want a different kind of nail and a different kind of hammer. In the same way, with AI you need to understand how to use data science to solve real business problems and deliver real value to your customers and patients, so it’s not just a case of “We have AI, too!”
Two heads are better than one
Good data scientists also need a measure of humility. Quite frankly, this is something that’s often lacking. We emphasize very heavily the importance of humility. Having one really smart person with a PhD is not going to work. No one knows everything, and very complex algorithms don’t always follow the rules or assumptions.
We teach people to be open, to ask for help, and to work in teams. You want to start with an absolute minimum of two data scientists. In the technology world it’s called “pair programming.” We actually like the pod concept of four or five folks working together, and we use the two-pizza rule (never have a team larger than can be fed with two pizzas).
Choose “all of the above”
There are different approaches to getting the right AI talent and skillset you need in your organization. You can hire some PhDs, you can train and retain your own talent, or you can partner with an outside group.
The first option requires an economics conversation. For example, a data-science PhD coming out of Stanford, you’re talking about salaries potentially in the million-dollar range. That’s not going to work for most companies. We looked at this challenge and said, “Okay how do we also grow our own? How do we build this organically, using what’s out there?”
You don’t necessarily need to hire the hottest algorithm person out there. But you also won’t be able to 100% build your own team either. Start with a core of servant leaders, build your teams around them, and have an open environment where it’s safe to make mistakes and learn from each other.
Everyone needs to be trained
It’s going to take some time to train people from within. One of the interesting things we’ve discovered is that you can find very smart, talented, capable people outside your organization, but when you bring them in, they’ll still have to learn your tech stack, processes, policies, maybe even industry. That can be a long learning curve, too.
When we hire, it’s someone who brings a very specific skillset and capability, and ideally someone in healthcare with some history with us, too. In our experience, we have gotten more value more quickly out of the people we have trained internally than we have out of large-scale hiring.
What’s worked well for us is to have teams with mixed levels of experience (from new graduates to more experienced), give them a problem, give them the freedom to solve the problem within guidelines, let them fail, and then have the theory and in-classroom training alongside that practical training.
Collaborate to survive
No matter what you do, you’re going to run into unknowns when you release AI models into the wild. Test data is static and safe, but live data morphs and changes every day, and you’re going to need to retrain your AI models.
We teach people to open up and pay it forward to get paid back, to contribute your learning, expertise, knowledge, and time to help others. If you do that, you’ll find that when you get to a problem that you don’t know much about, you will have a community of people around you who will reach out to provide the information and support to help you.
Article by: modernhealthcare.com