You know how at a party, you meet someone new, but immediately afterward, you cannot recall their name and only have a vague impression of how they look? You are not alone if this has happened to you. Human memory is as powerful and faithful as it is fragile and fallible. For example, we may remember vividly the face of a childhood friend but only have a blurred mental image of someone we just walked past on the street — which makes it easy for us to mistake a stranger for someone else.

What makes a difference here depends on the precision of our immediate or short-term memory. Clear short-term memory of a person we just encountered may help us avoid mistakes, whereas imprecise memory can lead to confusion (e.g., did I just walk past my neighbor and forget to say hello?). As we interact with an ever-changing environment, the precision of short-term memory has some important behavioral implications. However, surprisingly little is known about how the human brain enables this core cognitive function.

In a recent study, Kareem Zaghloul, Weiwei Zhang, and I decided to tackle this problem by tracking people’s behavior and brain activity during a short-term memory task. In this task, participants briefly saw colored squares flashed on a computer screen, and after 1-2 seconds one square reappeared, except this time without any color in it. The participants had a simple task: to recall the color of one particular square as precisely as possible from a continuous color spectrum.

To track the neural response associated with the remembered color, we leverage the rare opportunity to directly record neural signals from the human brain as intractable epilepsy patients were being monitored for their seizures with implanted electrodes. These patients participated in our study voluntarily and performed this task during a seizure-free recording session. To make it simpler, on each trial we directly told these participants which color square would be tested, reducing the memory load to effectively just one item (see Figure 1a).

With this minimal level of cognitive load, we assume that if participants are unable to recall the color, it would be driven by their imprecise memory instead of a failure in remembering just one simple color square. Our data confirm this assumption. Participants’ recall response mostly centers around the test color with some variability (see Figure 1b). By analyzing the millisecond-resolution neural data, our study further uncovers a novel neural correlate of short-term memory precision.

We find that a region in the middle of the brain — the medial temporal lobe (see Figure 1c) — retains information about the exact color that participants remembered during the brief delay period (see Figure 1d). Although wide-spread brain regions also contain this item-specific information — a finding replicating the existing literature, the memory information within the medial temporal lobe occurs before that in the distributed neocortical structures during the delay period and directly predicts participants’ subsequent recall precision.

This specific temporal relationship places the medial temporal lobe in a special place to support short-term memory precision. Mounting evidence has suggested that circuitry properties of the medial temporal lobe scaffold a neural computation to reduce interference among similar memories remembered over a longer time scale such as minutes, hours, days, or even longer. However, many researchers remain hesitant to generalize this neural computation to short-term memory. Just like RAM (i.e., random access memory) and hard drives constitute separate physical entities of a computer, the traditional view believes that the human brain may also contain different pathways to support temporary and permanent information storage in short-term and long-term memories, respectively.

To clarify this issue further, we also examine whether the medial temporal lobe is necessary for short-term memory precision. In a subset of participants, we delivered a weak electrical current to perturb the medial temporal lobe during the brief delay period on a random half of the trials as they performed a similar short-term memory task. The stimulation was so mild that the participants were unaware of it. Yet, this stimulation led to a reduction of 26% of their short-term memory precision on average (see Figure 2 for an example). Complementing this finding, participants’ short-term memory precision was reduced by 43% on average when we compared their task performance before and after resecting parts of the medial temporal lobe as a treatment of epilepsy.

Our data highlight the critical role of the medial temporal lobe in short-term memory. In contrast to the traditional view, these results suggest that the neural computations and substrates that support the quality of long-term memory may also be exploited in the service of the quality of short-term memory. As these data point to a shared constraint that governs how well we can remember across timescales, our findings may be helpful in characterizing subtle memory deficits in aging and clinical conditions. Moreover, these data suggest that we may target the medial temporal lobe to mitigate imprecise memory – an important question for future investigation.