The map is not the territory: How to build machine learning algorithms that better reflect human psychology

Time for a thought experiment: Imagine you had the power to perfectly predict your customers’ behavior. Will your new product flop or be a smashing success? Will a price increase be perceived as fair or make customers run for your competitors? Will customers start using your expensive new chatbot, or will they prefer to call a sales agent? The ability to make these predictions would change everything – from how you staff, to how you plan, to how you invest.

As artificial intelligence (AI) continues to evolve, the cost of making predictions about people is dropping rapidly, so this tempting possibility is becoming less and less far-fetched. This revolution in prediction is making waves across

 industries, paving the way for businesses to transform how they operate. The authors of Prediction Machines: The Simple Economics of Artificial Intelligence illustrate this with a compelling visual metaphor, comparing advances in AI to turning a dial that enhances the precision of predictions. As the dial is turned up, AI accuracy will sharpen our ability to foresee human behavior. 

At a certain level of predictive accuracy, we're not just looking at businesses getting incremental benefits; we're looking at them undergoing complete transformations. In the book, Prediction Machines, the authors use the example of Amazon’s predictive algorithm getting so good that they shift from a shop-then-ship model to a ship-then-shop model, where goods are sent even before the consumer realizes they want them. This level of anticipatory service could redefine the customer experience as we know it.

However, the potential for AI to continue improving and unlocking these transformative changes is limited by how well AI is trained to understand people. This understanding is fundamentally tied to the data AI has access to. Herein lies a crucial distinction: "the map is not the territory." The data that companies have readily available (the map) may offer insights into consumer behaviors, but it falls short of fully capturing the complexity of human psychology (the territory) that drives these behaviors.

At BEworks, we help companies create better machine learning algorithms by applying our team’s expertise in human psychology to better predictions. Our team is unique among behavioral science firms because each of our consultants holds a PhD in a specialized area of psychology and has years of experience applying science to improve business outcomes. We empower our clients to ensure their ML and AI models are leveraging the right data to predict outcomes that matter. By identifying the most informative psychological measures, we help clients train their ML models more effectively.

Here are some examples of BEworks projects where we’ve helped clients train better ML and AI models:

Predicting an Individual's Well-Being: Partnering with a digital well-being company, we conducted a study to identify the unique dimensions of human well-being – and there are 14 of them, ranging from physical health to job satisfaction. Armed with this "ground truth" of well-being, we were able to design a study to train an ML model to predict well-being from passive smartphone data, empowering individuals and healthcare professionals with better insights into user well-being.

Predicting the Likelihood to Seek Financial Advice: Alongside a globally-recognized fund manager, we created a model to predict investors' likelihood of seeking advice. Instead of relying on demographic data alone, which only accounted for about 4% of total variance in advice-seeking, we identified 12 new cognitive factors such as overconfidence in one's own abilities and trust in advisors that significantly increased the predictive ability of the model. The new model accounted for over 55% of total variance in advice-seeking behavior. Eventually, this model will be used to help de-bias investors and help them reach their financial goals sooner.

Predicting COVID-19 Vaccination Likelihood: In collaboration with Delvinia and 19toZero, we conducted a study and found that four cognitive factors, including belief in conspiracy theories and vaccine risk concerns, were excellent predictors of vaccination intentions. This psychological model dramatically outperformed demographic-based predictions, highlighting the importance of measuring how people think, not just who they are.

We have always known that understanding how people think is key to predicting behavior. By refining our measurement tools—through psychologically-informed questionnaires and interviews—we can then train machines to predict human psychological processes from more readily available data, like online activity or smartphone usage. Of course, this must only be done in service of providing customers with better experiences and helping them to make better decisions, and companies will need to comply with regulatory and ethical constraints around predicting human emotions, especially in the workplace.

We are excited by the possibility of AI-enabled transformations of businesses, especially as they will allow for higher quality preventative healthcare, superior customer service, and better decision-making at both the individual and organizational level. By equipping companies with knowledge of the “territory” of the human mind – not just the map of readily available data – we hope that better AI predictions will benefit both businesses and society.