Blog Series: PUBLIC IMPACT ANALYTICS SCIENCE (PIAS)
Photo Credit: DALL.E 2025
While computers in the age of AI possess the ability to follow data-driven problem solving approaches, they largely lack what is needed to solve complex and lingering societal problems: an important skill that I have termed insight-driven problem solving.
René Descartes—The 17th-century French philosopher, scientist, and mathematician who revolutionized much of Western thought—is perhaps best known for declaring “cogito ergo sum,” meaning “I think, therefore I am.” Less known are his important contributions to problem solving. He advocated for a new paradigm in problem solving that emphasized the reliance on empirical evidence, logic, and analytical reasoning. His approach demanded a scientific resolution to solving problems which entailed diligent study and meticulous analysis. He was one of the proponents of what we know as reductionism. In his own words, to solve difficult problems, we must:
“Divide each difficulty into as many parts as is feasible and necessary to resolve it.”
Analytical thinking is firmly tied to the doctrine of reductionism. As Russell L. Ackoff—an American scientist and a pioneer in the field of operations research, management science, and systems thinking—put it in his 1974 book “Redesigning the Future:”
“Analytical thinking is a natural complement to the doctrine of reductionism. It is the mental process by which anything to be explained, hence understood, is broken down into its parts. Explanations of the behavior and properties of wholes were extracted from explanations of the behavior and properties of their parts. The temperature of a body, for example, was explained as a function of the velocity of the particles of matter of which it was composed. An automobile’s behavior was explained by identifying its parts and explaining the behavior of each and the relationship between them.” [1]
Analytical thinking is also what arguably resulted in the first ideas for creating a machine capable of problem solving—what we know today as a “computer.” The Analytical Engine, which is generally considered “the first computer” in the history of computing, was proposed in 1837 by Charles Babbage—the English polymath and inventor who is referred to as “the father of the computer” by some scholars. Computers’ analytical thinking and problem solving are, however, all based on data-driven approaches. But, as I mentioned, to solve complex societal problems we often need to go beyond data-driven approaches and make use of insight-driven problem solving.
To better understand what insight-driven problem solving refers to, we should first note that solving societal problems often requires understanding various aspects of the complex phenomena at hand. Because not all such complexities can be included in the analyses, one needs to use the doctrines of reductionism and analytical thinking to divide the problem into as “many parts as is feasible and necessary to resolve it.”
Reduction must be done at a suitable level—not so much as to compromise the integrity of the problem and not so little as to lose hindsight. Reductionism at the right level allows analytics scientists to obtain insights powerful in understanding the underlying phenomenon and, hence, offering useful solutions.
Let us, for example, consider the need to find policies that can effectively introduce transparency to the quality of outcomes in the healthcare sector. As Wally Hopp of the University of Michigan and I discussed in a piece published by the National Academy of Medicine [2], the lack of information about the quality of outcomes in healthcare for years was blamed on peculiarities of the healthcare sector and was regarded by policymakers and others as another example of “healthcare exceptionalism.” In the past two decades, however, governments and some private organizations have made conscious efforts to increase quality transparency in the healthcare sector to enable consumers to make more informed decisions. The launch of Hospital Compare in the U.S. by the Center for Medicare and Medicaid Services (CMS) and similar efforts by National Health Service (NHS) in the U.K. are just some examples of such efforts.
In 2005, for example, the U.S. CMS launched a website called Hospital Compare, which provides information about the quality of care at more than 4,000 Medicare-certified hospitals. In 2011, United Kingdom Prime Minister David Cameron pledged that the NHS would make performance data publicly available and announced that “information is power, and by sharing it, we can deliver modern, personalized, and sustainable public services.” [3]
Unfortunately, however, the macro-level data and related empirical studies suggest that these efforts have not been successful. For example, several data-driven studies show that the launch of the Hospital Compare website and other efforts to increase quality transparency have not resulted in tangible improvements in outcomes. Why? And what can be done so public reporting efforts do not fall short of their promise? What do policymakers and other involved authorities need to do differently?
These are the type of questions Wally Hopp and I were curious to answer. Questions are sometimes easy, but answering them is often not. For us, this was especially the case because answering these seemingly simple questions requires understanding the impact of policymakers’ interventions in increasingly creating transparent quality information on choices made by various other decision-makers, including patients and providers. For starters, we needed to understand how different patients respond differently to increasingly transparent quality information. For example, which patients are most/least likely affected by this information? Then we needed to figure out answers related to providers and the healthcare market as a whole. For example, how will increased levels of transparency in quality information and the patient behavior it promotes impact competition among providers? Will providers respond by becoming more specialized regarding either specific medical procedures or geographic markets? Answering these intermediate questions could then help understand the effects on social welfare and the inequality implications of different choices available to policymakers. But, again, these questions are not easy to answer because of various interdependent decisions across patients, providers, policymakers, and various other entities involved in the healthcare sector.
By designing simple yet powerful mathematical models and calibrating them with data, we were able to provide some useful insights [see, e.g., [4]]. Then, we worked on transferring these insights into solutions for policymakers and other authorities. These steps are essential for insight-driven problem solving and significantly differ from traditional data-driven problem solving approaches widely incorporated in (and used in training and developing) AI systems.
Can AI tools obtain insight-driven problem solving capabilities in the future? Well, it is fair to say that it is hard to predict. But what we know today is that they currently lack it despite possessing various outstanding skills, including those required to pass some human-level cognitive psychology tests [5], find personalized dynamic treatments for patients [6], or solve the most difficult mathematical problems such as those from the International Math Olympiad [7]. In essence, current AI systems have demonstrated that they can work well with data and provide solutions to well-specified data-related problems. However, going beyond this to generate actionable insights and transfer them to valuable solutions for addressing complex problems (e.g., societal issues) using careful analytical thinking, reductionism, and other steps is not yet within their scope.
References
- Ackoff., R.L. (1974). Redesigning the future: A systems approach to societal problems. J. Wiley and Sons.
- Saghafian, S., & Hopp, W. J. (2019). The role of quality transparency in health care: Challenges and potential solutions. National Academy of Medicine (NAM) perspectives.
- Henke, N., Kelsey, T., & Whately, H. (2011). Transparency—the most powerful driver of health care improvement. Health International, 11, 64-73
- Saghafian, S., & Hopp, W. J. (2020). Can public reporting cure healthcare? the role of quality transparency in improving patient–provider alignment. Operations Research, 68(1), 71-92.
- Saghafian, S., & Idan, L. (2024). Effective generative AI: The human-algorithm centaur. Harvard Data Science Review, special issue 5.
- Saghafian, S. (2024). Ambiguous dynamic treatment regimes: A Reinforcement Learning approach. Management Science, 70(9), 5667-5690.
- Google DeepMind (2025). AI achieves silver-medal standard solving International Mathematical Olympiad problems, https://deepmind.google/discover/blog/ai-solves-imo-problems-at-silver-…