“Pick Important Problems, Fix Them”

Defining problems that can be fixed by AI

In his seminal book about regulation, The Regulatory Craft (2000), Professor Malcolm Sparrow from Harvard University, coined the phrase “Pick Important Problems: Fix Them” as a methodology for regulators when dealing with complex regulatory challenges, such as tax defaulters. Professor Sparrow’s approach to solving regulatory challenges can equally be applied to the field of Artificial Intelligence (AI) by making sure that you select the ‘right’ challenge and consider the best AI solution to address it.

The topic of where to start on the AI journey comes up during most conversations with clients. This implies that there is a journey with a start and a finish (which is not necessarily always true), and, it also implies that the client and AI provider are able to work together and develop an AI application that can solve one of the business challenges.

Put differently, ‘human intelligence’ is required to define what could be considered an important challenge. Once you have a clear view of what’s important, you still need to figure out whether an affordable AI solution exists that can help you fix the problem. However, the Maslow’s Hammer principle applies to AI as well - If all you have is a hammer, everything looks like a nail (Abraham Maslow, 1966). AI can help solve many challenges but it may not always be the best way to solve all business challenges. 

If all you have is a hammer, everything looks like a nail

So before looking at important problems, we need to clarify what AI is good at doing in healthcare.

AI, using advanced technology that enable machines to do complex tasks which would otherwise require human intelligence, is probably one of the most promising tools that can help humans overcome some of the world’s most pressing problems. Thanks to recent advances in the collection, storage, processing and analysis of data, we have access to some life changing solutions.

AI in healthcare is particularly useful when it comes to discerning patterns that would be otherwise almost indiscernible even to a trained human brain. This pattern recognition can assist healthcare professionals at all stages: diagnosis, treatment and prognosis of diseases, as well as speeding up the process of discovery. Of course, there are multiple success stories when it comes to AI in healthcare, more of which later. For now, AI solutions have successfully been implemented to assist physicians in their clinical decision-making, helped to accurately diagnose diseases and make solid predictions when it comes to operational issues such as waiting times and length of stay.

When it comes to selecting a problem that can be ‘fixed’ by an AI solution, we should really focus on problems with the following characteristics:

1.      You are confident that a solution exists to your problem. For example, even though a lack of medication adherence poses a big problem when it comes to the efficacy of treatment, it may be outside of the control of the provider to effectively deal with this challenge. However, long waiting times constitute an important problem and, to a certain extent, can be addressed by a healthcare operator. Similarly, even though I am pretty confident that an accurate AI based symptom checker may support the clinician’s diagnostic work, I am not sure if this is a solution to your problem;

2.      Good, reliable data exists and you have access to this data. One of the main prerequisites to a successful implementation is the availability of accurate, timely, valid and reliable data.

Data should be accurate, timely, valid, reliable and available

3.      There are problems with the current interventions: they are hard to implement, do not work very well or cost a lot of money. Taking the example of waiting times again, this is a particular challenge with lots of interventions but they are all cumbersome to implement. A better approach may involve using AI tools to create real time alerts when there is a risk of overcrowding.

4.      There are patterns out there but they are not visible to the human eye. This is not because you or your team are not smart enough or you haven’t received the right training. It is just because the volume of data is so big and there are so many variables at work that it becomes almost impossible to see the wood for the tree.

5.      Fail to prepare, prepare to fail: make sure you articulate what you want to achieve and monitor your progress over time.

Next time:

 Focus on the important stuff