What does this task measure? This question sounds trivial, but especially in psychology and neuroscience, it is not always answered easily. Even if the research hypothesis is seemingly simple, such as: “Does reward influence long-term memory processes?”. In our case, answering it took several years and three large-scale online experiments that changed what we thought we knew about a task we had been using for years.

Measuring the influence of rewards on learning and memory seems easy enough. Just offer participants money, if they can correctly remember some information – a picture, a word or something else – and see whether the reward boosts their memory. This type of “Motivated Learning Task” has appeared under various different names and versions in the literature. In an initial learning session, participants memorize pictures for example and are told they will receive a reward if they later recognize them – for some pictures, the reward is high, for others, it is low. That is, participants have to identify the old pictures among entirely new ones. Typically, researchers report improved memory for highly rewarded pictures. In their seminal paper, Adcock et al. (2006) ran the “Motivated Learning Task” while performing brain scans (MRI) and found that brain activity in reward-sensitive areas of the brain increased for high-value pictures during learning and influenced activity in memory-related areas. This increased activity was indicative of how well the high-reward pictures were remembered. One could conclude that rewards can prioritize specific memories for long-term storage, i.e., highly rewarded pictures are remembered and the low-reward pictures forgotten.

However, in this study and in many other similar studies, memory strength was considered to be the proportion of old pictures correctly identified as old the memory test (the hit rate). For many years, it has been well known in the memory field that accurately measuring memory strength requires also considering errors, i.e., new pictures mistakenly identified as old (called false alarms). Without considering errors, participants who want to use a decision strategy that optimizes hit rate could always answer “old” (leading to a 100% hit rate that does not reflect high memory strength). This is especially crucial when investigating rewards, since decision strategies could be optimized to get more money. The problem with the “Motivated Learning Task” is that we do not know whether rewards also influence the errors participants make, because new pictures are not tied to a reward. This reasoning was the basis of work by Bowen and colleagues (2020) that connected rewards to categories (e.g., landscape pictures vs. interior pictures) rather than to individual pictures. Other than previous tasks, this connected new pictures to a reward, allowing them to compare hit rates and errors. They showed that rewards influence decision strategies (not memory strength): participants become much more likely to say “old”, when they know they will receive a high reward for a hit.

However, in its original form, the task would require participants to precisely remember the reward category a given picture falls into to make a strategic decision based on the reward. In the task Bowen and colleagues (2020) used, that is not necessary, because the picture’s category tells them what the reward is. Therefore, we decided to approach the problem differently, without eliminating the need for reward memory: We by simply asked participants which reward they expected to receive each time they indicated a picture was old. (In our task, participants were pirates trying to win treasures.) We assumed that this would tell us whether participants’ decision strategies where different when they expected higher rewards. Unfortunately, human behavior is often more complex than initially expected and the data indicated that the pattern for error rates was not comparable to the hit rates, so we could not analyze the data in the way we had planned. What initially seemed like a failure, provided the most interesting insight of this set of experiments, but more on that later. We changed to a more direct approach to investigate influences on participants’ decisions by showing reward information before each picture also at test. We assigned new pictures a random reward, so an error rate for high and low rewards could be calculated individually. Unsurprisingly, we found that rewards shown during test influenced participants’ decisions – the higher the reward, the more likely they said a picture was old, so high rewards lead to a higher hit rate and a higher error rate. Nonetheless, we also found higher memory strength for high-reward pictures. This means that although rewards adjust participants’ decision strategies,  they also benefit long-term memory formation.

In another group, we shuffled the reward labels shown for old pictures during the memory test to further understand what decision strategies do to memory performance. This meant that the reward information participants saw during the memory test conflicted with the rewards they remembered from the learning session most of the time. Sure enough, participants that mainly relied on the reward information provided at test instead of the learning session tended to have lower overall memory performance.

So, how does reward information influence our behavior? Although we could show that under certain circumstances, rewards influence memory strength, all in all, reward-based decision strategies during testing dominate what people actually chose to do. This is where our data from the first experiment provided a compelling new perspective. When no reward is shown during the memory test, participants can only use reward-related strategies if they remember which reward belongs to a picture (stimulus-reward associations), and not just the picture itself. In fact, our data showed that participants who remembered the rewards well were also better at correctly identifying pictures as old.  Relying on this more complex memory trace allows for much more flexible behavior than just remembering pictures that had a higher reward and forgetting those that had a lower reward. A more natural example comes from experiments with scrub jays that show caching behavior, i.e., they hide food to retrieve it later. When caching their favorite food, mealworms, as well as less preferred peanuts, they retrieved the highly rewarding mealworms when tested immediately after learning (Clayton, Yu, & Dickinson, 2001). However, they did not forget about the peanuts: When the birds were only allowed to retrieve the food after weeks and the mealworms when the mealworms had decayed, the scrub jays went right for the peanuts. Because they memorized the locations and the value of both types of cached food, they were able to flexibly adapt their response strategy to the circumstances and retrieve the optimal food in each scenario.

Taken together, the “Motivated Learning Task” remains useful to study the relationship between rewards and memory, because it allows us to understand how remembering the relationship between rewards and other information can guide memory-based decisions. It is less useful, when interested only in the influence of rewards on memory strength.

Related Papers

Adcock, R. A., Thangavel, A., Whitfield-Gabrieli, S., Knutson, B. & Gabrieli, J. D. E. Reward-motivated learning: mesolimbic activation precedes memory formation. Neuron 50, 507–517 (2006).

Bowen, H. J. & Kensinger, E. A. Cash or credit? compensation in psychology studies: motivation matters. Collabra Psychol. 3, 12 (2017).

Clayton, N. S., Yu, K. S. & Dickinson, A. Scrub jays (Aphelocoma coerulescens) form integrated memories of the multiple features of caching episodes. J. Exp. Psychol. Anim. Behav. Process. 27, 17–29 (2001).