I thought the universe was like a watch, and God was a clockmaker? Or is the brain a computer? Or is it a finely-tuned machine?
I am an electrical engineer, so I am familiar with the things that he’s trying to compare the brain to. I see the brain as more like an FPGA rather than a computer or a data communication network.
An FPGA, or field-programmable gate array, is a chip that contains many thousands of little logical units, and a bunch of wires and switches.
These building blocks can be connected together to form any sort of complex circuit that the designer can express in a programming language. Once the chip is configured to implement that circuit, then it becomes a logical machine that does that particular task very efficiently. It can later be reprogrammed to implement another circuit very efficiently, as seems to be what happens in our brains over time.
This is another analogy that is by no means 100% accurate, but it’s a lot closer than the ones that we see presented by non-engineers. This is likely due to the fact that most non-engineers have no idea what an FPGA is.
Thanks for posting the excerpt. I’m going to have to buy this book even though it will frustrate me greatly. I’ve written up my basic reaction; hopefully its not too preachy.
To start with the final sentence – proposing “recasting neural computation as neural routing” is pretty close what I’ve written in research proposals. Examining biophysical mechanisms of dynamic routing in biological neural networks is one of my key interests. I agree completely that neural routing deserves more attention and understanding it is critical for understanding neural computation.
The author does a good job of describing the ubiquity of the computer metaphor in neuroscience and points out several of it key flaws, but reading neuroscience from people who’ve never studied neurobiology gets so very frustrating. This whole excerpt seems written by someone that has only been exposed to the “brain is a computer” end of neuroscience and despite seeing some of its major flaws still insists on describing things purely in those terms. We should be replacing the “brain as computer” metaphor, not expanding it.
“The same should be true for brains: if we could work out the basic principles of message passing, we could understand the role of individual neural computations. For decades, neuroscientists have been measuring diodes and transistors and ignoring the larger system of message passing.”
This should be the tagline of the book, because it highlights the premise and how it completely misses the nature of the problem. There are no diodes or transistors in the brain. Because it is not a digital computer. This is the single biggest problem of the computer metaphor for the brain, it fundamentally ignores over a century of research on neurons just because the neurophysiology doesn’t map onto a digital circuit. This has become so prevalent for so long that large branches of neuroscience get wholly ignored in most psychology, medical, and cognitive science teachings of the brain. Understanding the dynamic, analog processing within neurons is not possible if those neurons are treated as binary digital nodes, and the same is true for understanding how neurons route information within the brain.
“Most importantly, it must be possible to send messages selectively in the brain without changing the structure of the cellular network of neurons. All kinds of tasks involve sending messages to one place sometimes and to another place at other times. This seems obvious when stated directly, but it is rarely acknowledged.”
This is a really important point. Most artificial neural networks do presume that network nodes send the same signal to all downstream targets. This is not at all the case in real neural networks, and that is an enormous flaw of ANNs that is often ignored and I’m glad he pointed out. But neurophysiologists do acknowledge this, though, and have been studying it for over a century. Just because computational modelers ignore basic neurobiology research doesn’t mean it doesn’t exist.
“Neuroscientists have extensively studied the decision-making computations occurring in neurons. … But it is not known how the computed decision is routed to the selected output neurons. This question has not really even been asked.”
Of course its been asked. Again, just not by that branch of neuroscience because the question requires an understanding of the physical organ to meaningfully ask
A really basic intro to neural routing:
First key is anatomical connectivity; neurons connect to specific targets determining which cells they can transmit anything to. Second, neurons communicate through synapses – a form of dynamic, nonlinear chemical communication that is wholly different from transmission in digital electronics. Individual neurons can have thousands of different synaptic terminals connecting them to from 1 to thousand of target cells. The dynamics differ between synapses from the same neuron. These differences in dynamics then act as selective filters, so different aspects of the neuron’s activity (or signal) are routed to different postsynaptic targets. Neurons also make a mix of direct and indirect connections to the same target allowing interneurons to further refine which signals are transmitted to which target. The routing is dynamically regulated at many spatial and temporal scales, including intrinsic activity-dependent regulation within the pre- and post-synaptic cell and interneurons, and broad state-dependent signals that can modulate the synapses of a whole network.
The slow, painful task of learning the details of each of these require empirical questions and experimental research. For a couple reasons, the majority of computational neuroscience does its best to completely ignore them. I don’t see how any aspect of neural routing is better understood by trying to map the processes to a computer or a computer network. To be blunt, I see it as more of what I’ve seen since I was in college – people desperately trying to convince themselves that not understanding how neurons work won’t keep them from understanding how a network of billions of neurons works.
If this book gets more people examining how neural signals are dynamically routed within actual biological nervous systems, that would be fantastic. I hope it will and sorry for the screed
Hey Scientist, thanks for the detailed post.
For a pop-sci take on neuroscience in the form of self-help books, what do you think of the stuff Jeffrey M. Schwartz has put out? Seems a little more nuanced than is typical? Any other authors in the popular format (long or short form) who you think do a reasonable job?