Deep-Linking: The Metadata of Our Minds

Metadata is the hidden layer that organizes information and makes it easy to find. For example, a library catalog card gives details about a book– its title, author, subject, and location. Similarly, our memories are not isolated; they possess metadata in the form of connections to other memories. These connections are deeply linked, creating pathways that help us locate, recall, and make sense of what we know.

When you recall a memory, it often brings others along with it. Thinking of an actor might remind you of a movie they starred in, a particular scene, or even the emotion you felt watching it. These connected memories act like metadata for the original thought–contextual cues that enrich its meaning and make it easier to retrieve. This deep linking is not just a feature of how memory works; it's the very reason we can organize, strengthen, and use what we know.

Let’s explore our current understanding of how human memory works. This excursion will help us understand the essential role of deep linking and metadata in our ability to learn, comprehend, manage, and recall knowledge. We’ll begin by explaining the fundamental storage unit of a single memory and then review its entire lifecycle.

NOTE: Neuroscience has made tremendous progress in understanding how memory works, but this is a continuously evolving and fast-moving field. Many questions remain unanswered.

It’s well-established that our memories are not stored within a single neuron. Rather, memories are stored in the synaptic connections of a population of neurons. If you are familiar with the basics of large language models, like OpenAI’s ChatGPT and Anthropic’s Claude, you can think of these synaptic connections of a group of neurons, with the varying “strengths” of each synapse as analogous to a large language model’s parameters and parameter weights. In fact, AI neural networks were conceived of and designed to mimic neuroanatomy.

The concept of these neuronal groups as the storage depots of our memories has been around for over a century and is embodied in the neurobiological construct termed an engram. There are two other memory storage unit terms: neuronal ensembles and neuronal complexes.

Neuronal ensembles are populations of neurons residing in a local brain region that share synaptic connections and act in concert to perform a particular function, be it memory storage, sensory, or motor activity. In reality, a function such as the storage of an individual memory involves neuronal ensembles from multiple regions of the brain that are synaptically wired together in what neuroscientists call a neuronal complex. A favorite axiom in neuroscience is that “neurons that wire together fire together.” This “firing together” is the basis for storing and recalling a memory.

Engrammic synapses have different strengths and polarities. Generally speaking, each neuron releases only a single type of neurotransmitter. About 50-60% of the neurons in our brain are excitatory neurons. The axons of these neurons release glutamate at their synapses. Another 30-40% of our neurons are inhibitory and release Gamma-Aminobutyric Acid (GABA) at their synapses. Our engrams store memories in the synapses of the excitatory neurons. But If an engram were composed solely of excitatory neurons, it would store both a lot of signal and a lot of noise. The presence of inhibitory neurons in the engram silences much of the noise, increasing the signal-to-noise ratio and improving the memory quality. When sufficient excitatory axons fire in an engram, the result is reactivation and retrieval of the stored memory.

So, here’s our most fundamental understanding of how memory works: when we learn something or create a new memory, a physical change in our brain occurs, programming an engram into existence. Later, when that specific engram is reactivated and fires, we recall that memory.

I would be remiss if I didn’t mention at this point in the story of memory storage that recent research suggests that in some cases, epigenetic mechanisms—biological processes that modify gene activity without changing the DNA sequence—play a role in how memories are stored and passed between members, and even generations, of certain animal species. However, it's difficult to conceptualize a non-synaptic physiological process that allows for the specific and rapid retrieval of memories in humans. This alternative, non-synaptic storage of memories, likely plays a minor role in how we remember what we learn.

Now, let’s dig deeper into the entire lifecycle of memories.

Encoding is the initial step in creating memory, your brain’s representation of both external and internal experiences. These initial memory traces result in a labile representation within your brain tissue. We describe them as labile because they are fragile memories that may quickly vanish if further strengthening steps are not taken. These initial engram neurons are located in the hippocampus, a seahorse-shaped structure in the medial or inner aspect of the temporal lobe. There is one hippocampus in each of your cerebral hemispheres.

The hippocampus serves many functions. It plays a central role in memory formation, spatial navigation (your internal GPS), and emotional processing. When you have some extrinsic, sensory experience, learn something new, or are about to create a new memory of a novel thought or emotion you’re experiencing, a competition of sorts happens in your hippocampus. The competition is won by the most highly excitable group of neurons that can coordinate to encode that memory. Other factors beyond intrinsic excitability are likely also at play, including the density of pre-existing synaptic connections of the “winning” group of neurons.

This immediate, initial encoding is not a process where the synapses within the engram are instantaneously made more robust or numerous. Those synaptic changes take time. The instantaneous encoding begins with the transcription of DNA, resulting in the de novo manufacture of proteins within the engrammic neurons. These proteins will subsequently facilitate new synaptic connections and alterations in the strengths of existing synapses. Again, if you know something about neural networks, think of these synaptic changes as the alterations in the weights of the parameters in LLMs as they are trained with data.

In summary, the initial encoding is based on the intrinsic excitability of a group of neurons, already wired together to some degree,  that can act in a coordinated fashion and win the race to encode a weak version of the memory by manufacturing some proteins that can subsequently enable physical changes in the wiring that we call synaptic plasticity.

Consolidation is the fortification of the initial memory trace into a more durable memory through physical changes in the brain tissue. It comes in two varieties: synaptic consolidation and system consolidation. Now that we’re discussing the next steps in a memory’s life cycle, it’s more proper to use the terms neuronal ensemble and neuronal complexes since these more accurately reflect the dynamic processes that neurobiologists study and go beyond the concept of an engram as the fundamental construct of a storage unit of a single memory.

Once proteins have been manufactured during encoding, the winning neuronal ensemble may undergo synaptic plasticity. Its existing synapses are modified to become physically larger and more potent manufacturers and emitters of their neurotransmitters. In addition to changes in the existing synapses, new synapses may develop, further hardening the memory. These new synapses arise from developing dendritic spines. Synaptic consolidation is a process that takes hours and may progress over days, weeks, months, and even years.

If the initial memory trace encoded in a coordinated group of neurons were never reactivated, synaptic plasticity would be unlikely to develop, and that labile storage unit functionality would be lost. Fortunately, we have an innate, automatic mechanism for reactivating these memory neuronal ensembles.

During states of quiet wakefulness and, even more so, during deep sleep, our hippocampi replay the memory, stimulating additional de novo protein synthesis and its accompanying synaptic plasticity effects. Research suggests that this replay of memories can happen at 20X the speed of the initial encoding, and replay can happen in both forward and backward directions! It’s hard to imagine what that  ​​might be like.

It’s fascinating that some research suggests that the memories we replay during sleep are frequently not just the most recent and newest. Even older memories, perhaps needing consolidation or reconsolidation, are given replays. This behavior is analogous to spaced-repetition flashcard apps that deliver retrieval practice of the knowledge you are getting closest to forgetting.

So far,  we’ve primarily discussed dynamic changes in the initial hippocampal neuronal ensemble. However, another subsequent type of consolidation may develop over time: system consolidation. Here, other neuronal groups not part of the original encoding get recruited to store that memory. These include neuronal ensembles in the medial prefrontal cortex and other regions throughout the cerebral cortex that were active during the initial encoding but didn’t “win the competition.”  These ensembles may be brought into participation in the storage of that memory. Many researchers believe that the hippocampus may ultimately serve as an index, akin to a library’s card catalog, that points to the storage location for a given memory in cortical regions that were active during the initial encoding.

Some memories get duplicated and separately stored in the hippocampus, the medial prefrontal cortex, and other areas of the cerebral cortex. What is the benefit of storing the same memory separately in these different, unconnected ensembles? Could this serve as a backup system, like a mental Dropbox?

It’s remarkable that the experiential quality of memories stored in the hippocampus differs from those stored elsewhere in the brain. Hippocampal memories are more detailed. These detailed memories reflect the nature of the hippocampus, which functions as our internal GPS and is crucial for episodic and autobiographical memory. It holds more detailed recollections of places we've been and the events in our lives. In contrast, memories in the cerebral cortex are usually less detailed and more gistlike. This is why so much of our semantic memory, the memory of “what we know” and have learned, is more like a snapshot summary than a fully detailed blow-by-blow movie of what we recall.

How do we successfully recall a memory? The simplest explanation is that memory retrieval happens when we recreate some of the cues and context present when the memory was first formed. This process reactivates the engram—a network of brain cells that store the memory—bringing it from storage into working memory, where we can think and reason about it. Interestingly, even a partial recreation of the original state and context can be enough to trigger memory recall; a perfect recreation isn’t necessary.

However, memory retrieval is more than just reactivation—it’s also a dynamic process that changes the memory itself. When you recall a memory, you reactivate the network of brain cells that store it, but you’re also re-encoding or rewriting that memory with new information. This new data could include your current thoughts, feelings, surroundings, and environmental cues. These added details act as 'metadata' and can take two forms: external factors, such as what you're seeing and hearing, and internal factors, like your emotions and other thoughts during the recall.

This process is closely related to how we learn new things. When we think about new knowledge, we often connect it to what we already know. Learning becomes easier when we have a foundation of related knowledge. It’s much harder to absorb new information if it doesn’t tie into what we already understand. The more we know about a topic, the easier it is to incorporate new learnings.

As you recall a memory and re-encode it with new information, some of the neurons involved in recalling the memory may also be participating in engrams that are encoding your current experiences, thoughts, and emotions. This creates new links between the original memory and the present moment. As a result, the next time you try to recall that memory, it may be easier to retrieve, as the 'metadata' and additional connections formed during the previous recall provide new pathways to access it. Each time you remember something, you add more layers of metadata and strengthen the connections, making future retrieval more efficient.

This is why the saying, "Every time you remember something, you change it," holds true—we effectively rewrite our memories each time we recall them. This process also helps explain why eyewitness testimony can sometimes be unreliable in court—repeated recall can gradually alter the original memory.

We recall our memories by reactivating the engrams that store them. We trigger engram reactivation by recreating a subset of the context and cues present during the initial encoding and/or subsequent re-encodings. These contexts and cues are the deep-linked metadata of the memory. With each retrieval, we strengthen the engramic storage unit through synaptic neuroplasticity and create additional layers of metadata that point toward the engram.

Imagine for a second that you learned a concept in physics class last week that surprised you.  Days later, I asked you, “What was that concept that Ms.Smith told you last week that surprised you and shook a pre-existing belief?” My question recreated some of the elements of the context that were present when you encoded the memory of that experience. These cues of person, place, time, and emotion will likely help you to recall the concept. Those details in the question I asked you are parts of the metadata associated with the memory. When we remember what we have learned, we may have leveraged metadata from semantic and episodic memories to help us locate the target knowledge as we search our memory.

Our knowledge often comes from experiences arriving in our minds through our senses: hearing a lecture, reading a book, watching and listening to a video. However, we also gain new knowledge via intrinsic activities, such as thinking and reasoning about some knowledge we are trying to comprehend and relate to our pre-existing knowledge. This deep-linking of our thinking creates an internal web of knowledge, like the Internet and Wikipedia. A piece of knowledge that is not linked to any other engrams might be impossible to locate and retrieve from our 100 trillion neurons and their combined quadrillion synapses.

We have extrinsic and intrinsic experiences that become context and metadata to our semantic memories of our knowledge. Any of the available metadata for memory may serve as a pathway to reactivating the relevant neuronal ensembles and recalling that memory. For instance, you were able to remember that physics concept because of the cues about your teacher’s name and your emotion of surprise. Powerful emotions can add robust metadata to memories, especially episodic memories.

If I asked you to think of the word pizza, you might a moment later remember the excitement and thrill of the first date you had with the love of your life when you ate at a favorite pizza parlor. You might then recall other details of that event, even sensory details such as sight, taste, and smell. After I said pizza, your favorite toppings popped into consciousness. And these additional recollections triggered remembrances that veered off in other directions.

You likely were already aware that our memory functions by association. Memories are linked to other memories. But have you ever considered the value of purposely building as much metadata and links between your memories as possible, pointing to the concepts, facts, terms, formulas, and knowledge you want to remember? The more metadata exists for a given memory, the easier it is to recall.

Does this cascade of linked memories sound like a curse? When it comes to learning and remembering what you’ve learned, it’s a feature, not a bug.

What is the neuroanatomical basis of our associative memory? The engrams and neuronal ensembles that store our memories are composed of thousands of neurons, and each engrammic neuron of a memory can have thousands of synaptic connections to other neurons. Neurons that are participants in one engram are likely also participants in many other engrams via their synaptic connections. Our memories are woven together in a web of overlapping connections. If there is enough “wiring” from one engramic memory to another different engramic memory, reactivating the first memory may reactivate the second.

Metadata and deep-linking of memories do more than help you recall what you’ve learned. The deep linking of memories enables you to comprehend and make sense of new knowledge as you organize it, connect it to other knowledge you possess, and create mental models of the world. The more metadata you have attached to your knowledge, the smarter you’ll be. We are continually linking our thinking and weaving a complex mental web of knowledge and its metadata. The more robust and complex this web is, our understanding of the world will be richer. Thinking is linking. Every time you learn something new, you think about what you already know in that domain of knowledge and adjacent domains.

Linking our knowledge is a path towards greater creativity. All of us are creative beings. The richer, broader, and deeper your mental web of knowledge, the more capable you’ll be of coming up with novel perspectives and ideas. Creativity is often a remix of seemingly unrelated ideas that may not have been previously blended into a new concoction. One of life’s greatest joys is learning and using our knowledge to exercise our creativity.

Why do we sometimes permanently forget a memory? Is permanent forgetting a feature or a bug? The reality is that sometimes it is a feature and other times a bug. Some scientists theorize that memories exist and persist because they serve as predictions of the future. However, when memory is no longer useful, it may benefit our mental health and future memory capability to lose that memory. In this scenario, it sounds like a feature. But don’t think for a second of this as a case where we have limited storage capacity and need to eject a marble from our brain to make room for a new one.

Permanent forgetting can result from the degradation of the memory storage unit or from our loss of the ability to reactivate the engram and recall the memory even though the storage unit is intact. Disease, trauma, and tissue damage can partially or completely destroy memory storage units. Barring those pathologies, synaptic remodeling is likely the primary cause of memory erasure or loss of the ability to reactivate and retrieve them.

We continue to develop new neurons in the hippocampus, even into our adult years. Neuronal ensemble synaptic remodeling occurs in the hippocampus as this lifelong neurogenesis produces new neurons that insinuate into the “wiring” of existing hippocampal engrams. With sufficient synaptic changes, this neurogenesis may result in the loss of memory storage or simply the loss of our ability to reactivate and retrieve that stored memory.

Engrammic synapses are also subjected to the influence of non-neuronal brain cells. Microglia can upregulate synaptic activity, prune synapses, and promote forgetting. Astrocytes can also eliminate synapses and cause forgetting.

Laboratory animal research suggests that permanent forgetting, in the absence of disease, trauma, and tissue damage, may more often be a retrieval deficit than memory storage erasure. When synapses are lost, the engram may still have enough capacity to store the memory but insufficient capacity to support retrieval. Some scientists have postulated that the synaptic potency required for storage and retrieval are wholly different capabilities.

The best way to remember what you’ve learned is to practice retrieving it from memory. Getting knowledge out of our heads is as important as getting in! Hundreds of research papers support the incredible effectiveness of retrieval practice. It is the best tool for building robust recallability of what you’ve learned. The most established reason retrieval practice is so powerful is that when we recall a memory, we reactivate the neuronal ensembles that are storing it, stimulating protein synthesis and leading to synaptic re-consolidation. The end result is stronger and more potent synapses, and new dendritic spines lead to more synapses. But, there is another reason retrieval practice is so effective for building recallability, and it’s often overlooked. The new layers of metadata and linking of memories you add with each retrieval create many more pathways to locate that memory when you need it.

Once you appreciate the benefits of retrieval practice, metacognition, and linking your knowledge, you might be interested in learning about software tools that help you by supporting this process in ways that enable you to become the smartest version of yourself.

SmarterHumans.ai is a tool that helps you organize, comprehend, and recall what you’ve learned. SmarterHumans takes a holistic approach, enabling you to wire together and link your learnings from web pages, books (Kindle), videos on YouTube, Vimeo, Coursera, and Udemy, files such as PDFs, Word, PowerPoint, images, and your notes.

SmarterHumans is also a powerful note-taking app. You can bidirectionally link text in one note to text in other notes, creating a knowledge graph, your personal Wikipedia. You can trigger AI-generated flashcards of key concepts, facts and formulas in your notes. With SmarterHumans’ flashcards, If you ever struggle to recall the answer, with a click, the source will open at the exact relevant location where you can refresh your memory in the context where you learned it. The source where you learned it can be a PDF, Word document, PowerPoint, image, Kindle highlight, time code in a video on YouTube, Coursera, or Udemy, or almost any public web page.

SmarterHumans also provides training wheels to help you develop your metacognition and ability to focus as you learn. Metacognition is our ability to think about our thinking, monitor and reflect upon the quality of our thinking, assess if we are focusing and paying attention, and plan how to remediate gaps in our knowledge. Metacognition is a hallmark of elite learners.

We invite you to take advantage of our free accounts and discover if SmarterHumans.ai is the platform that will help you reach your full potential as a student and lifelong learner.

Once you appreciate the benefits of retrieval practice, metacognition, and linking your knowledge, you might be interested in learning about software tools that help you by supporting this process in ways that enable you to become the smartest version of yourself.

Also, be sure to check out Dr. Barbara Oakley’s Books