In our on-demand webinar, Too important to fail: How to bring better AI to healthcare, IMO Health’s Chief Strategy Officer, Dale Sanders, and Marc d. Paradis, Vice President of Data Strategy at Northwell Holdings & Ventures, discuss the status of AI in healthcare today.
In this chat they cover the importance of data quality for building better AI, methodologies for successful AI, and more.
Continue reading for a snippet of their conversation where Marc explains why AI is considered a wicked problem. Or, you can watch the full webinar here.
Marc d. Paradis:
In case you weren’t aware, AI in healthcare is a wicked problem.
Wicked is actually a technical term. For a problem to be classified as such, it is characterized by the following 10 features. Wicked problems are:
- Unique
- Lacking a clear definition
- Multi-causal, multi-scalar, and interconnected
- Important to multiple stakeholders with conflicting agendas
- Straddling organizational and disciplinary boundaries
- Connected to other problems
- Impactful throughout the system
- Defined by solutions that are not right or wrong, but rather better or worse
- Time-intensive to address
- Never completely solved
Indeed, wicked problems are these slippery, multidimensional, cross disciplinary, and often vague, system-oriented problems that tend to fold deeply into these grey areas of human judgment. That’s why they’re wicked and that’s why they’re so hard [to solve].
Let’s say we have a chest X-ray. On the basis of this chest X-ray, our first question might be: Does the patient have pneumonia? Second question: Is it COVID? It’s not a straightforward question to answer.
Alternatively, consider you have a set of regular vitals that have been taken from patients. [Then] your questions may be: Does the patient need to be woken up in the middle of the night to take another set of midnight vitals? Or: Can they be allowed to sleep through the night and have that restful, restorative sleep that will help them recover and get out of the hospital sooner with better outcomes? It’s not easy to see how the data in front of you directly leads to those kinds of questions or answers.
Despite these difficulties, I want to emphasize that AI in healthcare is saving lives today — whether in radiology, oncology, cardiovascular risk segmentation, or intensivists for hospitalist care. [In this webinar we mention a handful of papers], each with a company behind them. [These companies’ AI solutions are] in production and being used — AI in healthcare actually saves lives.
[And AI in healthcare is] is not just an insoluble wicked problem, it is an importantly impactful one, but it is not magical. You don’t simply sprinkle the pixie dust of AI on your hardest problems and see them suddenly transform from wicked problems into easy problems.”
