Affectivism and Predictive Analytics at the Learning and Performance Crossroads

A basic tenet of learning and performance (L&P) is that it subsumes the cognitive, psychomotor, and affective domains. The most mature of the three in modeling human behavior have traditionally been behavioral science and cognitive science. In the July 2021 issue of Nature Human Behaviour (VOL 5 | JULY 2021 | 816–820 | www.nature.com/nathumbehav), authors of “The Rise Of Affectivism” ponder “…whether we have moved beyond the eras of behaviourism and cognitivism, into the era of affectivism.” Indicators cited include the increased volume in funding and publications around emotional content and affect in memory, behavior, attention, decision-making, and perception, from 1980 to the present. Relevant content domains in which affect now plays a major role include “law, education, environmental research, and conflict and reconciliation research.” It wasn’t until the 1990s that research confirmed that human emotion is “an objective, measurable, and scientifically accessible phenomenon.” All quotes are from the cited article.

I thought about this while reviewing current L&P articles and studies that aim to overlay and map the affective domain onto more traditional taxonomy models. The earliest manifestations of Bloom’s taxonomy (1950s) focused almost exclusively on cognitive complexity, but more recent revisions to the classical model incorporate affectivism. A key motivator is artificial intelligence and its application to affective computing, machine learning, and robotics. A very early and successful application is the transition from classical linguistics to contemporary information technology. Phonetic transcription that transforms phonemes into diphonemes through acoustics has made text-to-speech plausible in a variety of domains, including L&P. This advance has much more to do with eliciting positive affect and emotion in learning than simply striving to make an artificial voice sound real. Through my years in L&P, there were many instances in which the cost of a human voiceover was prohibitive, but text-to-speech was not yet up to the task. In such cases negative affect rendered voiceover not an optional learning strategy.

A key goal of incorporating affect into L&P is to better predict how learners and performers will respond to L&P content and knowledge support. It is not only about matching levels of content complexity with corresponding human competency, but also framing content around an affective taxonomy. An example that comes to mind from the travel industry is the enormously complex process of airline ticket exchanges and attempts by travel software vendors to automate the process. Because of inherently high failure rates in computing exchanges and their corresponding penalties (“debit memos”) for making errors, the affective domain needs to be considered. No matter how much effort is put into how instruction is framed around cognition, there is prevalent fear, frustration, and anger experienced by agents because they are frequently penalized with debit memos. These emotions result in an unpleasant affect. To address it, some travel software vendors have agreed to cover debit memos costs rather than have travel agents avoid their products. This is baked into both instruction and practice, primarily to eliminate the unpleasant affect that most all travel agents experience when faced with calculating ticket exchanges.

The more mature endeavors seem to be around socio-affective modeling in competency-based L&P. Very simply, it’s not just about mapping L&P objects to skill/competency levels of learners, but to also include socio-affective attributes into the mappings and predictions. Affects such as being excited to meet learning challenges, discouraged to try, indifferent, absent, and others are the stuff of socio-affective modeling around prediction.

Now consider predictive analytics. It is all about identifying behaviors of like-minded folks when presented with a given stimulus. It narrows the target and predicts socio-behavioral outcomes, giving you just a few things to consider. Today’s predictive analytics use models rather than just brute-force data crunching. So the problem in L&P is how to frame the models and measure affect. Predictive analytics applied to advertising looks at the stimulus being the wares you are selling and determining the best consumers to target for maximum return on marketing dollars. For L&P, the “wares” are L&P objects, and the “consumers” are learners and performers. Predictive analytics tell you who to target to get the best return on your L&P dollars in terms of improved organizational performance.

I believe the real challenge is around quantitative measures of affect in what has traditionally been regarded as qualitative. Can you really call “excited” or “discouraged” metrics? How are they measured? Has L&P really made the affective domain for learners and performers “objective, measurable, and scientifically accessible” as has been done in other fields? Have advances in affective analytics around memory, behavior, attention, decision-making, and perception made their way into L&P science? In my opinion, the answer is somewhat, and at a very early stage. As I noted in my last blog, L&P content science lags data science, especially around the affective domain.

Peruse The Learning Ideas Conference schedule and you will find many presentations and papers addressing affectivism and predictive L&P analytics. But once again, cross-discipline cooperation and collaboration are key to evolving our ability to predict L&P outcomes. Watch this space for further developments and gems as they emerge at an ever increasing pace.