The memory hunters
If only there could be an invention that bottled up a memory, like scent. And it never faded, and it never got stale. And then, when one wanted it, the bottle could be uncorked, and it would be like living the moment all over again.
– Daphne Du Maurier, Rebecca
Memory is at once our most valuable possession, and our worst enemy. At any stage of our lives, it can be fleeting, unreliable, painful, or absent. Which is why, like Du Maurier’s melancholy narrator, many of us might wish that our memories lived outside of us: we could then choose their fate, saving the best and consigning others to the trash.
For CIFAR Fellows Sheena Josselyn and Paul Frankland, that kind of control is already happening. Within their integrated lab at Toronto’s Peter Gilgan Centre (part of the Hospital for Sick Children), the married neuroscientists have been able to implant and enhance memories in their rodent subjects, as well as remove them. New tools allow them to see groups of neurons that represent single memories. They’ve also been able to eliminate terrifying recollections, and contribute significant insights to the fight against Alzheimer’s disease.
Memory is mysterious territory indeed, and for all their progress, Josselyn and Frankland admit it remains something of a black box. It will be some years before these results can be translated from animals to humans, though that black box is getting more transparent all the time. Josselyn says that before we can treat major memory disorders, “we have to be able to understand how the brain works at the most fundamental level. We won’t be able to go forward with clinical treatments until that happens.”
The study of memory stretches back thousands of years, but it wasn’t until the 20th century that anyone considered looking for a physical trace of it in the brain. Studying psychology as an undergraduate at Queen’s University, Josselyn first heard the name of one such person: Karl Lashley. Starting in the 1920s, the American scientist spent many fruitless years searching for the engram – an elusive neural structure said to house a memory in the brain. Using crude tools that were state-of-the-art at the time (similar to soldering irons), Lashley induced brain lesions in rodents, then watched to see whether they’d forget how to navigate the mazes on which he’d trained them.
In the end Lashley never found his magic engram, though his research revealed much about the brain (such as the fact that memory is stored in various places, not just one.) But his example was to stay with Josselyn who, after receiving her PhD in psychology, spent time doing clinical work with sex offenders in prison. The work was interesting but somewhat rote. Feeling restless, she yearned to make new discoveries. The terrain that appealed to her most was memory.
“It seemed the most fundamental thing,” says the senior fellow in CIFAR’s Azrieli Program in Brain, Mind & Consciousness. Still enchanted by her chosen field over 20 years later, she says, “We are the sum total of our memories – I mean, that’s really who we are. It’s the essence of what makes us human.”
In the modern era, Josselyn felt she might be able to succeed where Lashley had not. “He’s kind of overlooked now, because he didn’t find what he was looking for and his tools were very blunt,” she says. “But he really had some important ideas. So what I’ve tried to do in my career is apply new techniques to answer questions that couldn’t be answered in the old days.”
Today, in the midst of what Josselyn calls a “renaissance period” for neuroscientists, one of these tools is viral vectors. Using this approach, scientists re-engineer viruses to express genes of interest (rather than viral genes.) In this way, neurons deep in the brain can be made to overexpress a transcription factor called CREB that has been linked to memory.
During her post-doctoral work with Mike Davis at Yale and Alcino Silva at UCLA, Josselyn showed that increasing CREB levels in a small portion of neurons deep within the brains of mice enhanced fear memory: an easy form of memory to study, because it’s so strong. Evolution has wired us to avoid danger; children may need repeated practice to remember multiplication tables, but easily recall the experience of touching a hot stove. Emotional memories of this type are stored in the amygdala, an almond-shaped structure located deep within the temporal lobe.
In these experiments, mice were given a small foot shock, paired with a sound tone. “The foot shock isn’t enough to hurt them, just enough to scare them,” Josselyn says. The shock caused them to exhibit a characteristic freezing response. CREB handily intensified the mice’s memory: when the sound tone was played on its own, they instantly froze in fear of the remembered shock.
Another significant find? Those improved memories could now be seen.
“For the first time, we could really look at what those cells were doing,” Josselyn says. “Certain cells seemed to be recruited over others: into this memory engram, and not for those things the mice learned before or after the manipulation that increased CREB.” Lashley’s engram had been found. Confirmation of this came later at Sick Kids, where Josselyn was able to remove a selected fear memory by eliminating the neurons that represented it.
Adding memories. Taking them away. If this type of cerebral modification were made available to humans, would it be a bane or a blessing? The answer is complicated. On the one hand, if certain memories are awful enough to interfere with our quality of life, the case for their excision is clear. On the other, there’s a healthy reason why we naturally remember terrible events more clearly than pleasant ones: to stop us from making the bad choices that may have sparked their creation.
As for the enhancement of memory – well, there can’t be anything wrong with that. Or can there? A popular 2017 paper co-authored by Paul Frankland suggests that the answer isn’t so easy.
Frankland’s office is right next door to Josselyn’s. Like his wife, he holds a PhD in psychology. The pair met while completing their doctorates and seem perfect complements to each other. The Cleveland-born Josselyn is talkative and funny, while Frankland, a native of England, speaks with polite care and reserve. Both have great memories, of course. She remembers every Madonna lyric she’s ever heard; he’s a walking encyclopedia of soccer statistics.
Frankland has long been absorbed by the question of neurogenesis, the term for new nerve cell formation in the brain’s hippocampus (it also takes place in the olfactory bulb). Less than 100 years ago, adult neurogenesis was deemed impossible. Even today, some neuroscientists are skeptical about whether it occurs naturally in adult humans.
Frankland is not among them. Two years ago, he and CIFAR Fellow Blake Richards published a literature review in Neuron, with the Dali-esque title, “The Persistence and Transience of Memory.” It was widely picked up in the popular press, with most articles headlined by some variation on the idea that Forgetting Makes You Smarter.
The review referenced work from Frankland’s own lab, showing that as new neurons are formed in the hippocampus, existing circuits are remodelled. This process overwrites old memories, making them harder to access. But that, says Frankland, isn’t necessarily a bad thing.
We are the sum total of our memories — I mean, that’s really who we are. It’s the essence of what makes us human.
“Certain kinds of forgetting are beneficial,” affirms the fellow in CIFAR’s Child & Brain Development program. “The world’s changing all the time, so you don’t need outdated information from a week, a month or two years ago. It’s no longer useful to you.” Frankland adds that you wouldn’t want to remember events perfectly anyway, because then you’d expect subsequent events to be identical. That would just be confusing.
His co-author Blake Richards offers an example of this problem. “Imagine that you work somewhere where you’ve got an intern named Matt, who then leaves. The next year, an intern named Mike starts. But if your brain stores the name ‘Matt’ permanently, every time you want to speak to Mike you’re going to call him Matt. An easier solution is just to forget Matt’s name – why bother storing it, if you’re no longer working with him? Then you’ll have a higher chance of getting Mike’s name right.”
In other words: a good memory is not necessarily one that’s packed with information, but one that selects for the most useful information. And it can only do that by erasing data that isn’t relevant anymore.
If adults do this regularly, it turns out that children do it even better. A key focus in Frankland’s lab is the question of infantile amnesia, a term coined by Sigmund Freud. Most of us can’t remember anything that happened in our lives before the age of three, and very little before the age of seven. And yet, small children have much better memories than adults for the details of their experiences. Frankland wanted to know why that was.
In this he was inspired by 10-year-old Charlotte, the daughter he shares with Josselyn. One day some seven years ago, the family was at the Bowmanville Zoo. After she was frightened by a loud duck near the pond, Charlotte was relieved to have her concerned father scoop her up and direct her toward a more benevolent animal.
“For months she would tell people this story about the zoo,” Frankland says. “Then one day it was completely gone. She didn’t remember it at all.”
Several years ago, Frankland led a team investigating why this might be. They found that young mice experience neurogenesis in the hippocampal region at a much higher rate that adults do, making them superior at encoding details of their experiences – but also inclined to forget them faster.
Still the question remains: are those vanished memories truly erased, or just hard to retrieve? Frankland’s most recent experiment provides support for the latter theory. He first tagged clusters of neurons representing memories of locations mice appeared to have forgotten. Weeks later, he stimulated those same clusters. As if by magic, the mice once again recognized the forgotten places. Their recovery was incomplete, however, suggesting the experience wasn’t perfectly preserved.
“We know this from our own experience,” Frankland says. “Even really important memories are not as detailed and precise as we imagine. When Sheena and I got married, or when our daughter was born — I have a sense of how those days unfolded, and what my feelings were. But I don’t remember those events with the kind of precision I have when remembering my flu shot this morning.”
Adding memories. Taking them away. If this type of cerebral modification were made available to humans, would it be a bane or a blessing?
Still, when memories are called up repeatedly, the synaptic connections supporting them are strengthened. This may explain individuals like Jill Price, the much written-about “hyperthymestic” who intentionally (and obsessively) remembers most days of her life in perfect detail. But most people aren’t like Price, because they’re too busy living in the present. They would rather create new memories and sacrifice old ones, instead of replaying the past on an endless tape loop.
Much of what we know about memory comes from research on model animal species, such as snails, zebrafish and mice. Can the extraordinary discoveries neuroscientists have made with animals be translated into humans? There is very good reason to think so. The era of genomic sequencing has revealed greater similarities between mice and humans, for example, than had been previously thought. It has also allowed for gene editing, another potential avenue of discovery and cure. Humans share key biological pathways with a wide variety of other living beings.
But there are obviously key structural differences between animal and human brains, too. “How things differ is in the way the brain seems to be wired up,” says Josselyn. “We’ve solved Alzheimer’s disease in mice, but never in people.”
It’s true: in 2017, she and Frankland stimulated neurogenesis in the hippocampi of both old and young mice who had genetic mutations associated with Alzheimer’s, with very promising results. Initially, both groups exhibited the brain plaques typical of the disease, as well as obvious memory difficulties. But after stimulation, memory was improved regardless of age, and plaque was reduced. Can this be done in humans? Not yet. “There’s something that we’re missing,” Josselyn says, “and it’s still kind of tricky to figure out exactly what that is.”
And yet, as Blake Richards points out, the study of a brain that isn’t even alive can still give valuable insights into how memory works.
Richards is a fellow in CIFAR’s Learning in Machines & Brains program. He comes to neuroscience not from psychology, but from artificial intelligence. As a student in the lab of CIFAR Distinguished Fellow Geoffrey Hinton, he became fascinated with the idea of neural nets: machine learning agents that mimic human brain processes.
“The reason I got into memory specifically,” he says, “is that when you try to design an AI agent and get it to interact with the world, you rapidly discover the need for memory. Because without learning and memory there is simply no way to get a computer to do the sort of stuff that you want it to do.”
One hundred billion neurons fill the human brain, connected by 100 trillion synapses. The brain is also powered by a chemical elixir of neurotransmitters, peptides and lipids. Naturally, a robot brain lacks all that, and is much simpler. But it’s still faced with similar functional challenges.
“If you come up with an AI solution to driving a car or managing someone’s personal accounts, it won’t relate to the human in terms of chemistry at all. But in terms of certain mathematical details, it might,” Richards says. Like humans, neural nets can also experience memory problems if their store of irrelevant memories isn’t cleared – a phenomenon known as overfitting.
Neural nets currently lack many of the human brain’s chemical and cellular details, because the presence of such complicating elements could interfere with scientists’ ability to train them using mathematical analysis. Now, however, Richards is conducting research into designing computational models of artificial brains that are more brain-like than ever before. He says this may lead to better drug design for Alzheimer’s and other diseases, as well as a greater general understanding of brain functioning.
“We could perturb components in the model to make predictions about how doing so impacts the system’s memory capabilities,” he says. “You can think of this as being almost akin to what we do with climate models, since climate is another incredibly complex interconnected system.”
Richards, Frankland and Josselyn have all worked together investigating memory — Richards completed a postdoctoral fellowship in Frankland’s lab. Their separate interests dovetail and diverge, but all agree that studying the concept from different angles provides a richer view of the problems it presents. This is where their membership in different CIFAR programs has been helpful.
… when you try to design an AI agent and get it to interact with the world, you rapidly discover the need for memory.
“Coming to Paul and Sheena’s lab was really good for me, because it was a different perspective on memory than I’d ever encountered,” says Richards. “Our ability to get out of a certain pattern of thinking and refocus on something else is aided significantly by exposure to people with different aspects of expertise. That’s what makes the CIFAR programs such a febrile ground for new ideas.”
Together, the three neuroscientists are helping to redefine what we mean by a “good memory”. Instead of one packed with readily retrievable information, it’s one that is imperfect and constantly changing: free of unnecessary flotsam, enriched only by the content that serves it best. And if the sunny day you remember is rainy in someone else’s recollection, that’s alright too.
“Not to get too philosophical, but this just comes back to the point that we are living in the present,” says Richards. “All we can say for sure is what’s happening to us right now. If you can get comfortable with the idea that your past is something that informs the present but may disappear over time, that’s probably best. Because that’s your reality.”
Believe the Impossible: The Future of… Fairness in AI
September 12, 2022
Toronto Star | Can we make artificial intelligence less racist?
June 22, 2022
CIFAR announces plans for second phase of the Pan-Canadian Artificial Intelligence Strategy
June 22, 2022
MOU between CIFAR, the National Research Council of Canada and Canada’s three national AI institutes opens doors to collaboration across sectors
May 25, 2022
CIFAR is a registered charitable organization supported by the governments of Canada, Alberta and Quebec, as well as foundations, individuals, corporations and Canadian and international partner organizations.