Hi all, 
Over the past few years, I have been developing tools for health informatics research (specifically observational health research) within Julia. My personal reasons for this has been personal productivity, higher baseline performance in software I build for this space, and stronger composition between existing work across the Julia ecosystem. However, a fear has crept into my thinking recently about this investment which is the shadow of Jevons Paradox. I wanted to share some thoughts about how I think Jevons Paradox both does and does not apply in the application of Julia to health informatics research and wanted to hear what you all think of it.
What Is Jevons Paradox?
When technological progress or government policy increases efficiency of resource usage, but the falling cost of this resource results in increases of its usage, rather than a reduction.
Put more plainly, the paradox emerges when you improve the efficiency of a resource to decrease its use but instead, the improved efficiency increases the demand of the resource.
The most canonical example comes from environmental economics wherein a city with a vehicle traffic problem wants to reduce traffic in their city. To remedy this problem, they think that increasing the number of lanes in their main city highway will reduce traffic. So, they build additional lanes. But rather than improve traffic, traffic becomes even worse as a result of more people wanting to drive through the city.
A Hypothetical Julia Health Informatics Research Scenario
Let’s assume the following hypothetical scenario:
Suppose a healthcare organization has access to a data asset with millions of patients that is hosted on their local data server. Their team of researchers is responsible for conducting a variety of studies across these patients to answer numerous questions they have on their compute server. The analyses that they run often take several days or even a week at worst and can heavily tax system resources in that time. As a result, this limits how many analyses researchers can be running concurrently.
A certain analyst advocates for Julia to be used within the organization. They show that they can run similar analyses in drastically reduced time frames. Additionally, these analyses were shown to be far less resource intensive than prior state of the art. The other analysts saw this and soon after, started running analyses in Julia.
The healthcare organization was then, as a whole, able to run even more analyses concurrently. Even though the efficiency of using their compute server was increased to run more analyses, the researchers still saw their systems being heavily taxed. Does Jevons Paradox apply in this scenario?
This hypothetical scenario is what I have been mentally grappling with as I seek to bring Julia (and high performance computing methods for that matter) into the realm of health informatics.
Ongoing and Open Discussion
I discussed this hypothetical with my friend E, who is an expert and researcher in technology policy. She thought that Jevons Paradox could apply here but rather to look at the application of paradox through the framing of “health informatics” more broadly rather than strictly programming. She thought that since a programming language and code itself can be viewed as a near “infinite” resource, the question of value and efficiency would more so come from within the “health informatics” domain more strictly.
I recently made the acquaintance of Phil Vernes (@pmodv) at JuliaCon 2023 where he attended my talk on using Julia for the analysis of huge patient populations (on the order of tens of millions of patients). We excitedly discussed the potential for my work at JuliaCon 2023. More recently, I posed my hypothetical to Phil and I thought his response was quite apt in thinking about the problem. Here is his response paraphrased:
In the hypothetical, I think Jevons Paradox here does apply, but is it not indicative of the success of the healthcare organization in achieving its mission?
To Phil’s point, he was more advocating that rather than having to avoid Jevons Paradox, we are actually using the paradox to our advantage by inducing more utilization of the data asset to answer even more research questions. Because, if the goal of the healthcare organization in the hypothetical is to truly answer research questions, would we not want to encourage the investigation of as many research problems as possible? I discussed this notion with E some more and we both agreed that, although Jevons Paradox is more commonly seen negatively, in this scenario, the increased utilization would actually be an asset. Some outstanding questions that I have though is:
-
Does this perspective make sense?
-
Is the hypothetical an oversimplification of this problem space?
-
Can Jevons Paradox actually be seen as a “good” thing or am I confusing concepts?
What do you all think?
Thanks!
~ tcp 