Nope, not Dungeons & Dragons (though AI-based D&D could be fun), let’s talk Diagnostics & Decisions.

Before we talk about AI-based decisions, let’s review human-based decisions, shall we?

One of the hardest things to do as an adult is decide on something that we really don’t feel certain about. Deciding what to eat for supper can be comically difficult when including friends and family but that is nowhere near the level of difficulty as making a decision that can be life changing. For example:

• Should I pursue college/grad school?
• Should I get married?
  • To him/her?
  • Now?
• Should I buy a new or used car?
• Should I change jobs?

Dr. Maidenberg gives some tips and tricks on how to approach such a big decision on her blog. Interestingly, tip #6 suggests that you create a table and map the pros and cons to the map creating a weighted table, and then sum those weights on a scale to objectively provide a suggested solution based on logical reasoning (Maidenberg, 2021).

Why is that interesting? Because that is exactly what fuzzy logic does.

Lotfi A. Zadeh introduced fuzzy sets and fuzzy logic to the world in 1965. Fuzzy sets are mathematical values bound on either end of the scale by 0 and 1, with 0 of course being the minimum and 1 being the maximum (Herrera-Viedma, 2015). The beauty of this scale is that it extends the previous logic sets derived by George Boole that make up the Boolean Logic system perhaps most famously employed in the digital logic circuits of electronic systems in general and computers in particular (Zohuri & Moghaddam, 2017).

In Boolean logic, something is either true (1) or false (0), ON (1) or OFF (0), or it is in a set (1) or it isn’t (0) depending on context. But what about temperature? How does something like temperature fit into Boolean logic, or does it? Think about this, everything exists at some temperature that is observable above absolute zero, right? So everything is either above absolute zero (1) or it isn’t (0). While this “fits” within the scope of Boolean’s binary logic, it isn’t very helpful because it doesn’t allow for us to find out where on the scale above absolute zero something falls. Something that is hot is just as much a member of the set of things that is cold so long as cold is above absolute zero, both items would be represented by a “1” in Boolean logic. Not very enlightening is it?

Humans tend to express thoughts and concepts linguistically rather than numerically. We say something is “hot” or “cold” much more often than we say something is “102.4 degrees Centigrade” or “3.8 degrees Centigrade” which is how a computational machine would describe temperature. So how can we reconcile these disparate expressions of the temperature of something? Enter fuzzy logic, an artificial intelligence (AI) technique that allows for reasoning with and overcoming incomplete, vague, and imprecise information, especially linguistic information (Chrysafiadi, 2023).

Concepts that exist on a continuum such as temperature can be represented linguistically by terms that imply a relative value that humans tend to understand empirically and intuitively. The terms can then be further refined via descriptive adjectives. For example, we get a certain concept in our minds when we hear the term “hot” or “cold” which can then be modified with descriptors such as “very hot” or “extremely hot” and “very cold” or “extremely cold.”  A process called fuzzification can then be applied to convert these linguistic values into numeric values by placing them on a weighted scale. See Figure 1 below for a visualization of this process (Wolf et al., 1996).

As you can see in Figure 1, a temperature of 23.5° C (the vertical dashed line) is fuzzified by being a member of two sets, it is 0.18 “hot” and 0.79 “warm” simultaneously. This collection of sets (very cold, cold, warm, hot, very hot) would then process the weights of “0.18 hot” and “0.79 warm” in a method called “inference” whereby rules are applied, the rules are then combined via Boolean logic operators (AND, OR, NOT, etc.) in a process called “defuzzification” to arrive at a “crisp” (i.e., singular) output value. This crisp value is the final result of the fuzzy logic. Because there can be multiple inputs that are processed and compiled together via the inference and defuzzification engines, fuzzy logic is considered a multi-input multi-output (MIMO) controller.

In any real-world decision-making task, information is frequently incomplete and imperfect (i.e., fuzzy). Think about the bulleted list of questions above and the information on hand to advise those decision-making processes. In order to have an AI help with any of those scenarios, the AI must be able to handle the incomplete and imperfect information.

Educators, trainers, and administrative staff have various levels of information on hand to work from when making decisions. Regardless of what data collection techniques are employed, there is never a complete picture of the situation presented. The educator only sees the student in the context of the campus so factors that influence performance such as home or social life (among other factors) are unknown with respect to deciding what may fully be encouraging or impeding the performance of the student in their studies. The administrative staff only sees the aggregated data of the class, cohort, or school and, like the instructor, doesn’t have access to the full scope or extent of what outside factors are having an impact on the performance of the group. In both cases, there is extrapolation that must take place or the lack of data will create a lack of decision-making which can paralyze and cripple an educational process.

Automated intelligence gathering that harvests relevant data from a number of resources both within and without the classroom can be performed by an AI tool that can help to fill in the knowledge gaps and provide a clearer picture of what is going on to better support the decision-making process. These multiple sources of information (multi-inputs, or MI) may have disparate data formats (numerical, textual, mixed) and if so, normalizing that data to produce useful information would be difficult if done manually, but with proper rules in place for how to parse that data, an AI should be able to compile and synthesize the inputs in a format that can infer what is going on and can inform the decision assistant AI to help the educators and administrators arrive at the most informed decision possible.

Of course, a potential pitfall is the “proper rules” mentioned above. Experts local to the system’s intricacies and nuances must craft the processing rules in such a way that makes sense for the application and the location. For example, rules that are successful in the inner city will most likely not be appropriate for a deeply rural application. This means that while the tool, the AI, can be useful wherever it is applied, it isn’t a “one size fits all” solution out of the box and must be expertly configured to be useful.

Ethically, there are challenges as well. While standards in education and training exist so that assessment and evaluation can be done on even terms, socioeconomic conditions, time allocation, resources available, prerequisites, and community support should all factor into the decision-making process as well. Those standards must be reasonably applied. For example, if measuring to a standard of performance for physical education (P.E.) students with physical impairments cannot be held to the same standard of performance as the athletic team captain. And vice versa. For example, wheelchair-bound students should not be assessed on how many blocks they can circle in the neighborhood, meanwhile, the captain of the cross-country running team shouldn’t be assessed on how long they can circle the track in a wheelchair.

Any decision support system, like any other system, is only going to be as good as it is designed, configured, and used. While artificial intelligence can overcome and compensate for imperfect or incomplete inputs, natural intelligence must still be smart enough to temper the results rationally.

References

Chrysafiadi, K. (2023). The Role of Fuzzy Logic in Artificial Intelligence and Smart Applications. In K. Chrysafiadi, Fuzzy Logic-Based Software Systems. Learning and Analytics in Intelligent Systems (Vol. 34). Springer. https://doi.org/10.1007/978-3-031-44457-9_2

Herrera-Viedma, E. (2015). Fuzzy Sets and Fuzzy Logic in Multi-Criteria Decision Making. The 50th Anniversary of Prof. Lotfi Zadeh’s Theory: Introduction. Technological and Economic Development of Economy, 21(5), 677-683. https://doi.org/10.3846/20294913.2015.1084956

Maidenberg, M. P. (2021, March 16). 6 Tips for Making Difficult Decisions. Retrieved from Psychology Today: https://www.psychologytoday.com/us/blog/being-your-best-self/202103/6-tips-making-difficult-decisions

Wolf, T., Gutmann, B., Weber, H., Fere-Borrull, J., Bosch, S., & Vallmitjana, S. (1996). Application of fuzzy-rule-based postprocessing to correlation methods in pattern recognition. Applied Optics, 35(35), 6955-6963. https://doi.org/10.1364/AO.35.006955

Zohuri, B., & Moghaddam, M. J. (2017). What Is Boolean Logic and How It Works. In B. Zohuri, & M. J. Moghaddam, Business Resilience System (BRS): Driven Through Boolean, Fuzzy Logics and Cloud Computation (pp. 183-198). Springer International Publishing AG. https://doi.org/10.1007/978-3-319-53417-6_6