By 
The old idea about the first couple of years of life is that babies don't remember anything, so those years don't really count for much. If experiences are being registered in the brain early on, including ones nobody can recall later, that assumption gets shakier, and those experiences could still matter for how a child learns and feels.
A study published in Science in March 2025 reports that activity in the hippocampus while a baby first saw a photograph was linked to whether that baby later looked longer at it. A team at Yale ran it, scanning awake babies while they looked at pictures.
The hippocampus is a seahorse-shaped structure deep in each temporal lobe, and it is central to forming memories of specific events. Our inability to recall those events from the first few years of life has a name, infantile amnesia.
The prevailing explanation has been that the infant hippocampus isn't up to the job yet, because it keeps maturing into adolescence, the authors write. Rodent work conflicts with that explanation, they note, since the infant rodent hippocampus does form memories.
That leaves the rest of us with a question about our own first years: are those memories gone, or just locked away?
Each picture came back later beside a new one, and a baby looking longer at the old picture counts as a familiarity preference, an indirect sign of memory.
The measure of interest was brain activity during encoding, meaning the moment a picture first goes in. Hippocampal activity was significantly greater while a baby saw a picture they would later look at longer, compared with pictures they later passed over, the researchers report.
Taken as a group, the babies showed no reliable preference for the old pictures. Across the full sample they looked longer at the familiar image on 51% of test trials, against the 50% expected by chance, which is not a reliable difference, according to the study. The same was true in both age groups.
The effect showed up in the back part of the hippocampus rather than the front, though the difference between the two parts was not itself significant, the authors write.
It was driven by the older infants, the 12 to 24 month group, with nothing detectable in the 4 to 9 month olds. In those older infants the effect was clearer at shorter gaps between seeing and testing, so whatever was picked up looked fleeting.
Tristan Yates, a Yale researcher and the study's first author, worked with senior author Nicholas Turk-Browne, a Yale psychologist who directs the university's Wu Tsai Institute. They used fMRI, or functional magnetic resonance imaging, which tracks changes in blood oxygen as a proxy for brain activity.
Getting usable scans from an awake baby is the hard part. The team managed it with 26 infants aged between 4.2 and 24.9 months, half of them under a year old.
Each baby saw photographs of faces, objects and scenes, every picture new and each shown for two seconds. About a minute later on average, with a few other images in between, that picture reappeared beside a fresh one from the same category for a four-second test.
Turk-Browne set out the comparison the study rests on. "We measured brain activity when a baby successfully formed a memory of a picture, as indicated by later familiarity with that picture in a visual paired comparison," he said in an email. "There was a difference in activity in the hippocampus during successful vs. unsuccessful encoding, suggesting that this brain region is able to encode memories."
Babies with better overall memories, he added, showed stronger evidence of hippocampal encoding.
The analysis that produced the hippocampal result works inside each baby, comparing pictures that individual later looked at longer with ones it didn't. Turk-Browne gave a plain reason for doing it that way: "People tend to remember some things and forget others."
Paul Frankland, who runs a memory lab at the SickKids Research Institute in Toronto and the University of Toronto, had no part in the Yale work. "What the Yates paper shows is that the rudimentary machinery for encoding episodic memories is in place by 12 months," he said in an email.
The way he put it, "the ability to *make* memories is emerging. Just not the ability to keep them around." The problem, on his account, is downstream: "Something happens after encoding that makes them gradually inaccessible."
The Yale authors make a version of the same argument, writing in their abstract that mechanisms after encoding may be more responsible for infantile amnesia.
Frankland's lab has studied what that something might be, in mice. New neurons keep being added to a part of the hippocampus called the dentate gyrus, a process called neurogenesis.
In a 2014 study in Science, Akers and colleagues report that increasing neurogenesis in adult mice after a memory had formed was enough to cause forgetting.
In infant mice, where neurogenesis runs high and memories disappear fast, reducing it after learning cut the forgetting down, the same paper reports. Guinea pigs and degus make most of these neurons before birth, and their infants didn't show the same forgetting, until the researchers pushed neurogenesis up after learning.
Frankland's explanation is that "ongoing remodeling of hippocampal circuits through the addition and integration of new neurons likely overwrites existing memories."
A 2018 study in Current Biology from the same lab, also in mice, tagged the neurons active while infant mice learned to fear a particular place. Switching those tagged neurons back on with light in adulthood, up to three months after training, brought the apparently lost memory back, Guskjolen and colleagues report.
The authors take that as evidence the memory had become inaccessible rather than erased. None of this has been shown in a person.
Turk-Browne has two unpublished projects aimed at the human version. One is a case study of two children, aged 3 and 9, shown head-mounted camera video from their own infancy against video from the other child's, and his team is preparing to submit it imminently. "The takeaway is that we see consistent evidence that the 3-year-old's brain remembers their own videos, whereas the 9-year-old's doesn't," he said in an email.
The other follows 7 families who filmed at home through infancy, with the babies scanned every three months while watching their own footage and another child's. Data collection is finished and submission is expected by the end of the year. Neither has been peer reviewed.
Looking time is an indirect stand-in for memory. The test came about a minute after each picture, so it says nothing about whether anything survives a month or a year.
Twenty-six babies is a small sample, and the group never showed a familiarity preference overall, so the brain effect rests on within-baby comparisons.
The authors write that their design "may reflect early stages of encoding that might be partial or incomplete", and they say directly that they are not claiming all familiarity preferences depend on hippocampal encoding.
The flat result in the younger babies may partly be task difficulty, since the two-second exposure was fixed for every age. The neurogenesis and reactivation results come from mice, and the home-video work is unpublished, with one part of it involving two children.
The finding is worth taking seriously, and it rests on 26 babies, so another few dozen could move it either way.
The scans Turk-Browne's team has already collected, one set every three months, are still being analysed and not yet published. They are set up to test whether anything encoded in infancy is still detectable months later.
The larger gap is time. The delay in this task was about a minute, while infantile amnesia covers the first few years of life.
Sources
- T. S. Yates and colleagues, Hippocampal encoding of memories in human infants, Science, 2025.
- K. G. Akers and colleagues, Hippocampal neurogenesis regulates forgetting during adulthood and infancy, Science, 2014.
- A. Guskjolen and colleagues, Recovery of "lost" infant memories in mice, Current Biology, 2018.
- Emails from Nicholas Turk-Browne and Paul Frankland, October 2026.
