According to their website, IQVIA holds over 60 petabytes (60 million gigabytes) of proprietary data used to provide insights and inform decisions across the healthcare industry. Think market dynamics around a disease, demographics and site selection for clinical trials or pricing for a new therapeutic.
How does one pull the insights out of that mountain of data? Chris Steel was brought on board to drive AI and machine learning into all their products, standardize on platforms and unify the message.
Our conversation went from things AI can do to how to manage AI projects across the enterprise. We talked about creating synthetic data to help diagnose rare diseases, wrangling multiple AI projects, and of course, the potential of ChatGPT.
As I reflect on our conversation, it strikes me that AI broadly is very much like a child. (Keep in mind that I am very much a noob in this world and know just enough to be dangerous.) When you first get one, you have high hopes for its potential. You also have a responsibility to train it well.
Diagnosing rare disease
Children learn a lot about people by observation. Similarly, a deep learning model can figure out on its own what relationships exist in a collection of data.
Rare diseases are often difficult to diagnose (partly because they are rare). How can we identify patients with a rare disease whose diagnosis is currently uncertain?
Training a model on a limited number of positive diagnoses is problematic. AI can be used to generate a set of synthetic data based on known positive cases. That data along with normal data is used to train a model that can then be tested. Finally, the model is used to identify patients with the disease that have been undiagnosed up to this point.
That is pretty remarkable.
Sorry to interrupt here, but you are subscribed aren’t you?
ChatGPT
If you’ve been paying attention, you know that ChatGPT has also delivered some pretty remarkable results and its performance on professional medical and legal exams is stunning. At the same time, it can be a confident liar. In my own experience, I have seen it deliver scientific references that would be completely convincing if I didn’t bother to look them up and find out that they don’t actually exist.
Chris agreed ChatGPT has a lot of potential. He also offered some important caveats. It can record your questions and whatever else you feed it - like your data. Don’t do that. Remember: if it’s free, you are the product. It could share that info with someone you wouldn’t want to have it. Again just like a child!
Beyond testing the free version, get your own instance, be aware of the cost and have a business case because that computing power is not cheap. Chris kindly explained the difference in computing power between Google looking up the information it has indexed and ChatGPT delivering complex answers almost instantly on the fly, (and not paid for by advertising (yet?)
How many projects?
Avoiding duplication of effort across teams in a company as large as IQVIA is a challenge. Chris needed to know who was doing what with AI when he came on board. He asked people to share their projects but didn’t get a lot of response. So he took what he got, put it in a spreadsheet…
And then in the virtual team that I've mentioned where we have all these leaders… I plopped up the spreadsheet and I said, “Here’s what people are doing.
And look, you know this team is doing this and this team is doing that. You know, these guys are the leaders.” And suddenly right after that, as you can imagine, everybody starts dumping all of their projects on me. A lot of them we had to sort of weed out because they actually weren't really AI/ML related, but yeah, turn it into competition and then they are willing to share.
There is your management tip for the day. When you want people to share projects across teams, make it a competition.
This all has me thinking about what the data ecosystem of the future will look like. Where is the dividing line between companies that collect and analyze data for themselves and large repositories that sell the insights to multiple customers? Let me know what you think in the comments.
Your deepest insights are your best branding. I’d love to help you share them. Chat with me about custom content for your life science brand.
Wrangling Multiple AI Projects