“`html
At any moment, the majority of individuals are chasing various objectives: responding to emails, doing laundry, deciding how to allocate retirement savings, assisting a child with their math studies. When faced with such diversity, how do we determine which objectives to focus on at any particular moment? We clearly can’t wash clothes if we’re on an airplane, but if we strip away these specific scenarios and analyze perseverance towards objectives in general, what elements drive us to persist with a particular goal, and what influences us to shift to an alternative goal, even if just for a short time?
Caltech graduate student Sneha Aenugu and John O’Doherty, the Fletcher Jones Professor of Decision Neuroscience, explored this question through an online gaming experiment. It was previously established by social scientists that individuals often excessively persist in pursuing a long-term goal even when it might be more advantageous to shift to a different goal. However, Aenugu and O’Doherty aimed to measure this inclination and examine how it changes with unpredictable circumstantial shifts.
“There are numerous activities we could engage in at any moment. How do we determine, ‘Alright, this is what I want to focus on right now’?” Aenugu, a third-year grad student in social and decision neuroscience, inquires. “Games serve as an excellent medium for this because during gameplay, you must choose which objective to follow at any specific instant. Additionally, when selecting which objectives to pursue, you need to be attuned to a fluctuating environment. What objectives are your competitors aiming for? Are they obstructing your achievement, or is the environment as a whole impeding you at this time? Do you transition to a different objective and later return to the previous one when conditions are more favorable, or do you stick with your original goal?”
The game developed by Aenugu and O’Doherty to assess human goal-persistence behavior required players to gather cards from three categories: cat, hat, and car. Each category comprised two varieties of cards (for instance, the “car” category has two types of cards: one a key and the other a luggage piece), and the game rewarded points when players assembled seven cards in a category (or, in a subsequent version of the game, four, six, or eight cards depending on that category). Gameplay consisted of distinct segments of online play; in each segment, the likelihood of acquiring a card from a specific category varied. For example, in one segment, there might be an 80 percent chance of obtaining a cat card, while in another, the likelihood might drop to 40 percent or less. In certain versions of the game, players were informed that the odds would shift across different segments and that one category would appear more often than the others, though they were not advised on the extent of these changes. In other iterations, players received explicit information regarding the odds (80, 75, 70, 60, 55, or 50 percent) but were not told which categories would actually dominate.
The findings? As anticipated, players continued to gather cards in a category they were already focusing on, even when this was not the best option. However, the degree of over-persistence among players varied significantly. “Some individuals excel at delaying immediate gratification and anticipating future rewards, while others struggle,” Aenugu notes. “Since people often favor instant rewards, they might prefer to remain with a particular category because it’s closer to fulfillment—that’s when they earn points in the game—even if that category is not favorable in a given segment of gameplay,” Aenugu articulates.
This tendency to over-persist is characteristic of a retrospective method to decision-making in gameplay; players reflect on their progress to decide their next steps. To contrast the decisions from these players with optimal gameplay, Aenugu, whose foundation is in engineering and computer science, developed two additional players, both algorithms, that operated prospectively—using immediate outcomes and future forecasts for their decisions. One algorithm was designed to adapt to the perceived odds in any segment and select cards from categories appearing to perform well, irrespective of past performance or if they were from a category nearing completion. The second algorithm also operated prospectively but exhibited a preference for completing categories when possible, rather than solely selecting cards from categories that seemed to have the best performance. This behavior is termed discounting.
Human players demonstrated greater persistence than either of the computer-generated agents, suggesting that some additional factor beyond discounting influences players. Aenugu employs a metaphor from classical physics—momentum—to elucidate the players’ tendency to over-persist. Momentum, Aenugu describes, “is a byproduct of progress itself and the rate of that progress. We have mathematically demonstrated that momentum, understood in this manner, offers a good approximation of the time it takes for players to achieve their goal.”
Aenugu observes that calculating goal-persistence through momentum does not optimize performance as well as the prospective model. “However, conducting the same calculations based solely on momentum—that is, on current progress and the rate of that progress—is almost as effective as a prospective model, and this computation is very inexpensive to execute. This means it requires considerable cognitive resources to depend on prospective reasoning. Besides, the environment is unpredictable. You can’t foresee when things will change. Thus, relying too heavily on a flawless prospective model is not practical. Occasionally, a simpler model is all you truly require.”
“It’s not as if we are fated to over-persist,” Aenugu emphasizes. “If we acknowledge our tendency to persist excessively, and if we possess more information about the environment where we are making decisions, we may be able to adjust our strategies for better effectiveness. Even in our study, providing players with guidance regarding the odds of a predominant category in any specific segment altered their behavior.”
The researchers intend to explore how these findings on over-persistence, particularly its variability among different individuals, could influence computational psychiatry: integrating neuroscientific approaches into clinical psychiatric practice. “We believe that understanding the nuances in how individuals select their goals can provide insight into specific conditions such as depression, anxiety, ADHD, or OCD,” O’Doherty remarks.
The paper, titled “Building momentum: A computational account of persistence toward long-term goals,” was published in the February 2025 edition of PLOS Computational Biology.
“`