Foundations and Frameworks
How does understanding the differences between symbolic AI, machine learning, and reinforcement learning help you think more critically about AI use in healthcare?
Reflect on how the R.O.A.D. (Requirements–Operationalize Data–Algorithm–Deployment) framework can guide effective development and implementation of these technologies in healthcare settings.
By understanding the underlying mechanisms, we move beyond the “AI hype” to rigorously assess each technology’s risks, reliability, and ethical footprint within the high-stakes environment of patient care. Dr. MaCulloh mentioned utilizing the Design Thinking process at the beginning of each development cycle. This aligns exactly with what we, in tech consulting, most frequently use during the post-strategy realization stage.
The Crucial Roles of AI Paradigms
However, I have found that, most of the time, the solution is not choosing one paradigm over the others. Instead, it is easy to observe that each of the AI paradigms plays a crucial role in a functional AI system. The following evidence reflects this viewpoint:
- Symbolic AI (GOFAI): This approach, often considered rule-based, is very helpful for pre-screening cases against current rules and regulations because it is highly transparent, fully explainable, and auditable. Though rigid, it is perfect for qualifying cases for the minimum requirements set by law or regulation.
- Machine Learning (ML): When processing a case, the algorithm’s confidence is derived from its datasets. In today’s world, we do not lack data from the Health Information System (HIS); we are simply not using it optimally. ML is considered a “Black box,” making it difficult to explain the rationale behind its decisions.
- Reinforcement Learning (RL): RL is ideal for perfecting an AI development project during the system adoption to early majority stages by utilizing the system’s continuous learning and manually adjusting coefficients. This process gradually increases the system’s capability and reliability, allowing it to handle common cases before moving to more complex ones.
Although RL generally raises high ethical and safety concerns, the perfect time to roll out an AI system is often during flu seasons or a pandemic. Patients tend to focus more on treatment outcomes and are more accepting of AI’s ability to provide timely diagnoses and speed up clinical appointments.
The non-Linearity of the R.O.A.D. Framework
Although the distinctions between the paradigms are clear, we must consider several key oversights:
The iterative process is rarely linear, as the R.O.A.D. framework suggests. Besides applying R.O.A.D. to the general healthcare project, it is possible that subsets of the R.O.A.D. process are required within each phase.
- Recursive Loops: As a product manager, it is common to find situations where requirements have been thoroughly collected and evaluated, but the target datasets are incomplete, too costly, uncleaned, or lack consistent labeling. As a result, it becomes necessary to initiate a subset of the R.O.A.D. process within the Operationalize Data phase, because every time we successfully define a solution, we also introduce a new set of problems.
- Changing Requirements: Change requests are filed, and often these requests affect the requirements outcome or push the product principles in a totally new direction.
I suggest that during R.O.A.D. implementation, domain experts should not be confined to the requirement phase, nor data scientists only to the Operationalize Data stage. All roles defined in the framework need to be present for the entire lifecycle as a unified team to validate concepts defined in the previous stages through the define-design-deploy-testing processes.
There is also a need to clearly declare the governance body at the project’s kick-off to handle non-technical issues. Regulation, ethics, or biased treatments and interpretations often arise during the process. Clearly defining who is responsible for mitigating these issues and ultimately distilling the final algorithm is crucial.
Conclusion
In past years, AI was often positioned as optional within the healthcare industry. Many published papers have focused on data usage, ethical concerns, the cost of development, and the need for sufficient computing power to process algorithms.
Today, almost every industry has passed the threshold for automating routine processes. Data centers, robotic production lines, and autonomous vehicles all require sophisticated intelligence. The healthcare industry is on the verge of consolidating data and forming alliances between governmental and private sectors—including financial/insurance institutions and hospitals—to accelerate this momentum.
The R.O.A.D. framework provides a strong foundation for managing the lifecycle of AI development. However, human behavior and psychology also play important roles in interpreting the meaningful decisions computed by the AI system. In cross-cultural contexts, laws, logic, and even favoritism may be treated with different weight. We must constantly be aware of all variables during the construction of a new intelligence.