I'm a trained physicist (undergrad) in biology (phd). I've always wanted to give a retort "could a physicist understand a microprocessor?". the answer is no, no they couldn't, unless they were trained in computer science. The abstraction of 'state evolving in time according to analytical equation' really breaks down when looking at computational systems, such as a microprocessor, as well as a biological cell.
Interesting. My intuition is the opposite [1]: computers are highly optimized to decouple the different functions and subsystems. That is, even when the state space is huge, only a very narrow set of state transitions is possible. This constraint is what makes experimentation tractable, in contrast to biological systems, where state will bleed across the whole system because, at some point, that was a fit adaptation.
I think that view is historically validated in that, in wartime, one side would steal the other's encryption hardware (e.g. the Enigma machine) and be able to piece together a meaningful description of what it's doing, without access to the original designers, and even get to the point where they could pose what encryption function they're trying to reverse.
Of course, there's a key distinction there -- those were largely mechanical systems, while the question here is about tiny microprocessors. So I'd agree there's a sort of pre-requisite that physicists be aware of the existence of some key primitives that the system operates on (here, current, voltage, electricity, etc. along with the ability to see small enough parts). If that's what you meant by being "trained in computer science" then I'd agree, though that's a non-standard use of the term.[2]
In any case, there's more to the physicist's toolbox than closed-form analytical equations.
[2] Per the Dijstrka quote ("Computer science is no more about computers than astronomy is about telescopes"), compsci is about the limits of computational processes, not electronic hardware per se.
> Per the Dijstrka quote ("Computer science is no more about computers than astronomy is about telescopes")
Off topic, as the origin and precision of the quote doesn't affect your point, but: AFAICT the closest Dijkstra actually said to this was in
https://dl.acm.org/doi/10.5555/25596.25598 :
> As a result, primarily in the U.S., the topic became prematurely known as "computer science"---which actually is like referring to surgery as "knife science"---and it was firmly implanted in people's minds that computing science is about machines and their peripheral equipment.
I'm a physicist. When I was in school, a year of electronics was typically required for an undergrad physics degree. The capstone project at my college was building a 4-bit minicomputer from gates.
There were two reasons for this tradition. First, physicists are often called upon to act as multi-disciplinary problem solvers. Second, experimental physics is almost exclusively electronic and computerized. Things like data acquisition boards for personal computers were only beginning to be commercialized, and a lot of us rolled our own gear.
I always thought the spirit of the paper was "we have a set of techniques to probe the brain and try to understand functionality (e.g. calcium/voltage indicators, optogenetics, laser ablation); could we have understood how a black-box microprocessor works by using EE equivalents of these techniques?" It's less about applying equations and more about, say, activating one transistor at a time and seeing what happens.
This paper is so frustrating! They chose a microprocessor-based system, where a single set of components does all sorts of things, switching function millions times per second. Of course you cannot analyze it using biological methods, no organisms work like that!
A much better thing to analyze would be something FPGA based, or even old CPU-less games like PONG. Look at the labeled schematics -- 6th image at [0]. Each element has a 2-letter label like G2 or B3, those are actual chip position.. so if you are looking at ball/paddle collision circuit, it is B2 or G3. So for example damaging 2nd chip in 2nd row (B2) will make ball/paddle collision detection not work, 5th chip in first row (A5) will prevent vertical ball movement and so on..
This would give a totally different conclusion for the paper, which I think is much more realistic.
> where a single set of components does all sorts of things, switching function millions times per second. Of course you cannot analyze it using biological methods, no organisms work like that!
uh, what do you think the brain does ..? does it have a magic special place cell for every place you’ve ever remembered being? (no)
The brain has specific area for each function - one part of the brain is responsible for sense of smell, another for vision, etc.. See Wikipedia [0] for basic introduction to this.
Compare this to CPU, where a single register bank, ALU, memory controller are used for everything - player movement, score keeping, sound, disk access, keyboard input handling.
it gets substantially more complicated than this very rapidly, i’m sorry to say.
And the debate between anatomists and physiologists rages as fiercely as ever.
I don’t deny the existence of anatomically distinct brain regions — my research area is in the auditory cortex — but i promise you it’s not nearly as simple as “one area one function”
As somebody who understands radios (Amateur radio is a hobby, and I've scratch built receivers) I had to stop reading. It's a fascinating, but somewhat frustrating read. I do think the overall approach is really interesting though and aptly describes the overlap in knowledge that often happens in otherwise disparate disciplines.
This is an interesting point, and I hope it is actionable. Of course we need to understand the biological components and systems in detail.
Unfortunately biology varies wildly between individuals, while the radio is constructed to a specific design, from parts standardized to three significant figures or so. A wiring diagram for a radio is sufficient, but cells are influenced by a very dynamic environment. We won't diagram my gut biome to that level of detail, yet it influences My biology.
Actually, lots of circuits can be built with components specified to within 20%, perhaps with some key components specified more tightly. Good designs are robust against component variation.
Agreed, got my degree in BME and it was probably evenly split between Bio, EE and Mech E. Went on to work in industrial automation, which a lot of people look at me with a questionable stare, but actually makes perfect sense.
I remember going to a career fair senior year right after doing a physio lab where we directly studied feedback systems. The company I ended up working for was demoing their software, which was control graphics. Looked a lot like the diagrams we were using in labs.
Yes, one was designed while the other evolved. That makes understanding their operation qualitatively different (e.g. one is almost linear sequence of nearly ideal stages with minimal degrees of freedom, while the other are completely non-linear chemical systems with hundreds or thousands of degrees of freedom). Even reverse engineering a GPT3 instance is mind boggling and the number of independent variables is tiny in comparison.
> Yet, we know with near certainty that an engineer, or even a trained repairman could fix the radio. What makes the differ- ence? I think it is the languages that these two groups use (Figure 3). Biologists summarize their results with the help of all- too-well recognizable diagrams, in which a favorite protein is placed in the middle and connected to everything else with two- way arrows. Even if a diagram makes overall sense (Figure 3A), it is usually useless for a quantitative analysis, which limits its predictive or investigative value to a very narrow range. The lan- guage used by biologists for verbal communications is not bet- ter and is not unlike that used by stock market analysts. Both are vague (e.g., “a balance between pro- and antiapoptotic Bcl-2 proteins appears to control the cell viability, and seems to correlate in the long term with the ability to form tumors”) and avoid clear predictions.
This thought-provoking essay nevertheless makes an apples-and-oranges comparison.
Let's try another experiment. Present a radio repair technician with car (ICE) that cranks but won't start. Give the technician, who has zero previous experience, no documentation or training and just watch what happens.
I'm going to guess that the technician will follow something that looks a lot like the approach used by the biologists in the essay. Disassembly. Manipulation of components in isolation. Hypothesis of function. Hypothesis validation. Rinse and repeat.
The reason the comparison doesn't work is not some variation of vitalism (the author brings up as one objection). The reason the radio technician can solve one problem (broken radio) and not the other (cranking no start) is not language or vitalism or anything else.
It's the simple fact that when faced with a broken radio, the technician uses a model that works. It reflects reality and stands up to the harshest scrutiny. The technician doesn't get that model through induction, it's handed down by those who created the system. And to be fair, the manufacturer helps by building its product to code, rather than some hazy spectrum of possible codes.
The author tries to make a case for "formal" approaches by which it seems like he's talking about mathematical approaches. But I think this misses the point.
Trying to understand a defect in an alien system poses a conundrum: science teaches us to break the system down into parts or "systems", but there are many ways to do that. Even if we hit on the right fracture points we easily lose the connections that made the whole work.
A classic radio receiver is reasonably easy to fix, because it's a pipeline. The usual tools are a signal injector and a signal tracer. The signal injector allows you to put in a test signal at various points in the circuit, and the signal tracer lets you listen to what's coming out. You start at the front and back ends, and work towards the part that isn't working.
its spiritual sequel: “Can a neuroscientist understand a microprocessor?” https://journals.plos.org/ploscompbiol/article?id=10.1371/jo...