Why do we need a new metaphor for the brain?

The metaphor of the brain-as-computer has dominated scientific study of the brain almost since the idea of “neuroscience” was invented. Why should we look to the internet as a new metaphor for studying how brains work?

We think of our brains as computers. This is no less the case for neuroscientists than for other everyone else. The computer metaphor has served us well. The explosion in knowledge about how brains work in recent decades has come in large part because researchers have approached the brain as a computing machine, one that performs mathematical manipulations, possesses memory allocations, and generally processes information in a way similar to a computer.

But there are brain functions that cannot be explained in reference to computing devices. In addition to computing, the brain is also a fantastically complex communication system. The thing about communication is that they operate on fundamentally different principles than those that govern computation.

For example, communication across a large network requires flexibility. It also exploits randomness and indeterminacy. And because a mass communication system spans a large area, it requires verification that messages arrive at their intended destinations. The internet has clever solutions to these and other challenges. It is time that we adopt the internet metaphor to guide the next phase of neuroscience research. The internet metaphor can help build better machine intelligence, and it can also help all of us utilize our brains more effectively.

An Internet in Your Head is a new book by computational neuroscientist Daniel Graham that lays out this argument in fine detail, and in a way accessible to anyone with an interest in brains.

How do brains meet the challenge of network communication?

The physical laws of the universe make communication among separated entities unlikely. The central and perhaps most unsettling implication of Einstein’s theory of relativity is that no two locations in space can ever be fully reconciled. They will always be separate, and even light cannot provide instantaneous connection. At galactic scales, light can take thousands of years to travel from star system to star system, ensuring that no reliable connection or exchange of information is possible. At a fundamental level, choosing to interact with someone or something at a distance away from ourselves is problematic in our universe.

Though the brain is much smaller than a galaxy, it still faces the challenge of getting information from one place to another in real time. It must do so in an uncertain environment. Lots of things can go wrong when trying to get a message from here to there, even if the distance traveled is just a few millimeters of brain tissue. Message corruption can come from a host of possible failures in a neuron’s chemical machinery, from spontaneous firing of a nearby neuron, or even from an errant cosmic ray. With the speed of message transmission far slower than that of light, the consequences of message failure could be large, and not easily remedied.

The biggest problem of all is selectivity. How do we let one out of many distant parties know that we would like to communicate with them, and then do so reliably and in real time? A communication system is not worthy of the name if we cannot select at will with whom we will communicate.

It has taken humans thousands of years to solve the challenges of communication. For most of human history, fast, selective, reliable communication has been a distant dream, and a sharp contrast to the clunky, furtive, slow, inflexible and error-prone reality of semaphore, telegraph, post, and other systems.

Yet with today’s internet, we have an extremely efficient and robust solution. The internet is not perfect—and the content of messages we send on it are another matter entirely. But the internet has provided a stupendously efficient solution the problems of fast, selective, and reliable message passing, allowing billions of entities to intercommunicate across every land mass, as well as in the skies and underground. This success comes in the context of components and a global environment that are uncertain even in the best of times. Everything from squirrels to heat waves and solar storms can interrupt signals on the internet. But it succeeds because it was designed to work despite an uncertain environment.

The brain faces the same challenges of massive intercommunication, but this fact has been obscured by thinking of it as a computer. While the internet was engineered to address the challenges of an uncertain world, computers largely don’t need to worry about them. After loading it with algorithms, data, and electrical current, a computer requires no interactions outside its housing. The core of a computer has a small number of strictly regulated streams of information (essentially just external data and memory data), and permits a similarly small number of outputs. There is little selectivity in the process.

How has the brain met the challenges of communicating within itself?

We can rule out the notion that the brain has no overarching communication strategy. The brain has 86 billion neurons to work with—each a tiny bag of highly choreographed chemical interactions, and each just a few hops away from any other.

Together, neurons need to achieve diverse and flexible behavior. To do so, evolution must have shaped the brain’s strategy for intercommunication. The idea of An Internet in Your Head is to examine how has the brain has achieved reliable and selective communication using neurons. Though the brain is not exactly like the internet, I argue that it likely exploits similar communication strategies for similar reasons.


At the beginning of a book, there is usually an Acknowledgments section where the author thanks the many people who helped them in writing the book. I certainly have lots of people to thank who have made An Internet in Your Head possible. But acknowledgments have a special meaning in the infrastructure of the internet. Any time you send information over the Internet, like sending an email, the receiver acknowledges receipt of your data with a tiny return message called an ack.

The internet succeeds in part because of trust among routers, but acks ensure that messages arrive intact. The internet operates according to Ronald Reagan’s maxim for international relations: trust but verify.

If a receiving router doesn’t send back an ack for some message parts in a timely manner, the sender knows to resend them because those parts probably got lost. It’s a spectacularly effective solution, one that allows any sender to pass messages to almost any receiver with very high reliability.

Under the internet metaphor, acks should exist in some form in the brain. Individual parts of the brain send messages to many possible destinations, and do so flexibly. How can a given brain part know that its message was received? I hypothesize that the brain could provide acknowledgments through its many looping connections.

One brain area that is famous for looping connections is the thalamus. This area gets connections originating in all sensory organs (except smell); for example, everything you will ever see travels from the eye to the thalamus, then on to the cerebral cortex.

But the cortex sends many more connections back down to the thalamus. Each part of the thalamus specializes in a particular sensory function (also motor functions), and each one receives a great deal of feedback from the cortex. Information traveling over these neural wires (axons) goes in a loop. The thalamus’ connection architecture seems ideally suited to providing acknowledgments: higher areas in the cortex–ones involved in detecting motion, for example–can talk back to the thalamus to say that relevant visual information coming from the eye made it to the right place, and was processed successfully (or not). It is known that signals can traverse loops from the thalamus to the cortex and back very quickly, with round trips taking as little as 9 milliseconds.

Seeing the looping connections between the thalamus and cortex as supporting reliable message delivery is a change from the computer metaphor explanation of this architecture. Under the computer metaphor, the thalamus serves to “adjust the weights” of signals as they move around the brain, making some stronger and some weaker. But no one has explained why loops are necessary to do this. The long axons required to build a loop are difficult to build during development. They are also metabolically expensive. We can only fit so much wiring in our heads, so the brain sharply economizes on long axons. Building loops with longer axons also costs precious milliseconds of signal travel time. If the thalamus’ job is to adjust weights, why can’t it do this locally, on the first pass?

One reason the thalamus hasn’t been considered in terms of communication engineering–and why the possibility of acks being passed in corticothalamic networks hasn’t been raised before–is that the thalamus’ job has traditionally been seen as computation. In other words, the signals sent over thalamic loops have been assumed to be undergoing computational operations, like making signals bigger or smaller. You would only notice that thalamic signals were something like an ack if you thought of those signals as messages, and if you saw the system as a whole as a communication network. This is the change in perspective afforded by the internet metaphor for the brain.