Challenging assumptions about infant memory, a new Yale University-led study shows that infants as young as 12 months old can encode memories. The findings suggest that infantile amnesia — the inability to remember our first few years of life — is more likely caused by memory retrieval failures rather than an inability to form memories in the first place.
Thank you for reading this post, don't forget to subscribe!
Yates et al. investigated the mechanistic basis of this infantile amnesia by scanning the brains of awake infants with functional magnetic resonance imaging while they performed a subsequent memory task. Image credit: Kang Heungbo.
Despite infancy being a period of rapid learning, memories from this time do not persist into later childhood or adulthood.
In general, humans cannot recall events from the first three years of life — a phenomenon known as infantile amnesia.
Why grown humans have a years-long blind spot in their episodic memory for the period of infancy remains a puzzle.
One theory suggests this occurs because the hippocampus, a brain region crucial for episodic memory, is not fully developed during infancy.
However, research in rodents challenges this idea, showing that memory traces, or engrams, are formed in the infant hippocampus but become inaccessible over time.
In humans, infants demonstrate memory through behaviors such as conditioned responses, imitation, and recognition of familiar stimuli.
However, whether these abilities rely on the hippocampus or other brain structures remains unclear.
“The hallmark of these types of memories, which we call episodic memories, is that you can describe them to others, but that’s off the table when you’re dealing with pre-verbal infants,” said Yale University’s Professor Nick Turk-Browne, senior author of the study.
For the study, the researchers wanted to identify a robust way to test infants’ episodic memories.
The scientists used an approach in which they showed infants aged four months to two years an image of a new face, object, or scene.
Later, after the infants had seen several other images, they showed them a previously seen image next to a new one.
“When babies have seen something just once before, we expect them to look at it more when they see it again,” Professor Turk-Browne said.
“So in this task, if an infant stares at the previously seen image more than the new one next to it, that can be interpreted as the baby recognizing it as familiar.”
The authors, which over the past decade has pioneered methods for conducting functional magnetic resonance imaging (fMRI) with awake infants (which has historically been difficult because of infants’ short attention spans and inability to stay still or follow directions), measured activity in the infants’ hippocampus while they viewed the images.
Specifically, they assessed whether hippocampal activity was related to the strength of an infant’s memories.
They found that the greater the activity in the hippocampus when an infant was looking at a new image, the longer the infant looked at it when it reappeared later.
And the posterior part of the hippocampus (the portion closer to the back of the head) where encoding activity was strongest is the same area that’s most associated with episodic memory in adults.
These findings were true across the whole sample of 26 infants, but they were strongest among those older than 12 months (half of the sample group).
“This age effect is leading to a more complete theory of how the hippocampus develops to support learning and memory,” Professor Turk-Browne said.
Previously, the team found that the hippocampus of infants as young as three months old displayed a different type of memory called statistical learning.
While episodic memory deals with specific events, like, say, sharing a Thai meal with out-of-town visitors last night, statistical learning is about extracting patterns across events, such as what restaurants look like, in which neighborhoods certain cuisines are found, or the typical cadence of being seated and served.
These two types of memories use different neuronal pathways in the hippocampus.
And in past animal studies, researchers have shown that the statistical learning pathway, which is found in the more anterior part of the hippocampus, develops earlier than that of episodic memory.
Therefore, the authors suspected that episodic memory may appear later in infancy, around one year or older.
This developmental progression makes sense when thinking about the needs of infants.
“Statistical learning is about extracting the structure in the world around us,” Professor Turk-Browne said.
“This is critical for the development of language, vision, concepts, and more. So it’s understandable why statistical learning may come into play earlier than episodic memory.”
Even still, the new study shows that episodic memories can be encoded by the hippocampus earlier than previously thought, long before the earliest memories we can report as adults. So, what happens to these memories?
“There are a few possibilities,” Professor Turk-Browne said.
“One is that the memories may not be converted into long-term storage and thus simply don’t last long.”
“Another is that the memories are still there long after encoding and we just can’t access them.”
“And we suspect it may be the latter.”
“In ongoing work, we are testing whether infants, toddlers, and children can remember home videos taken from their perspective as (younger) babies, with tentative pilot results showing that these memories might persist until preschool age before fading.”
“The new findings provide an important connection.”
“The work in humans is remarkably compatible with recent animal evidence that infantile amnesia is a retrieval problem.”
“We’re working to track the durability of hippocampal memories across childhood and even beginning to entertain the radical, almost sci-fi possibility that they may endure in some form into adulthood, despite being inaccessible.”
The study was published in the journal Science.
_____
Tristan S. Yates et al. 2025. Hippocampal encoding of memories in human infants. Science 387 (6740): 1316-1320; doi: 10.1126/science.adt7570
