I spent most of college trying to get a degree that the college did not offer.
The problem announced itself early. I have always liked building things that work (both the design side and the hacking-it-together-until-it-holds side), and building things is a poor respecter of departmental boundaries. But the university was organized as if the natural sciences were separate subjects that happened to share a quad. Physics in one building, biology in another, chemistry down the hall, and computer science off somewhere else entirely, treated as a service department. And that's just the sciences. Every other department was isolated to a degree that made the sciences look positively collaborative.
The silos are an administrative fact. They are not a fact about the world.
The common threads are the whole point
Consider what electrodynamics does for understanding the brain. Not as an analogy. Literally. The signal you record off a cortical surface is a volume-conducted field, and if you don't understand how fields propagate through an inhomogeneous conductor you will misread your own data. That isn't physics being applied to neuroscience. That is the physics that the neuroscience is made of.
Or take the visual pathway. The retina and early visual cortex perform what are, in effect, analog convolutions: center-surround receptive fields computing spatial derivatives across the input, edge detection falling out of the arithmetic before anything you'd call cognition has happened. Differential calculus, running in wetware, at the front of the pipeline, so that everything downstream receives structure instead of raw luminance.
A student who takes vision science without calculus learns this as a list of facts about cell types. A student who takes both sees the same object from two directions and understands why it has to be that way. The second student is not smarter. They were just standing in a place the curriculum didn't have a name for.
Making the case
So I made a nuisance of myself. I got involved in every science department I could reach, talked my way into labs that were restricted to majors, and pleaded with department chairs to approve a proposed "Integrated Science" major: coursework drawn across disciplines, organized around problems rather than around buildings.
The chairs were not hostile. They were structurally unable to say yes. A major belongs to a department, and a major that belongs to all of them belongs to none of them, and there is no line on the budget for that.
Alo Basu, who ran neuroscience, was my confidant through the whole process and eventually found the way through: rather than create a new major, absorb the idea into an existing one. Upper-level coursework distributed across computer science, mathematics, biology, and physics could all count, provided each of them fed the lab's actual research in electrophysiological recording and analysis of evoked potentials. Which they did, because that's the point. Recording evoked potentials is a physics problem and a biology problem and a signal processing problem simultaneously. I didn't have to argue for the integration. The experiment did it for me.
I got what I wanted. It just had a different name on the diploma.
Why computer science ended up taking most of my time
Here is the thing I did not expect.
The more I reasoned through the stack (gluons to atoms to neurotransmitters to cells to tissues to organisms to whole biological populations), the more obvious it became that I was never going to hold it. Not through more effort, not through better notes. The pattern-matching throughput required to keep even a few of those levels coupled in working memory exceeds what my brain does, and no amount of wanting it changes that. There is a real ceiling and I was standing under it.
Computation is the only tool I know that reliably holds a structure larger than the head that built it. Not by being clever. By being external. If knowledge can be codified properly, a machine can reason over more of it at once than I can, and it can do so without the context of the argument decaying every time I go get coffee.
That "if" is carrying enormous weight, and it's most of what I've worked on since. Codifying knowledge properly is the hard part. Everyone wants to skip to the reasoning.
Two books that set the direction
Two things shaped how I think about agentic systems more than anything else I've read.
Fluid Concepts and Creative Analogies is about what a general system actually is, approached from both directions: top-down and bottom-up definitions of "intelligence" and what breaks in each. The claim that stuck with me is that human capability rests on recognizing fluid, abstract patterns and reaching for strategic analogies through sparse, parallel activation, not on brute-force computation at massive scale. Those are different mechanisms that can produce similar-looking outputs on the problems you happen to test.
Which is why ELIZA matters. In the 1960s it appeared intelligent to people who should have known better, and it was doing string matching the entire time, fully traceable, start to finish. The lesson isn't that ELIZA was a fraud. It's that fluency at the surface tells you nearly nothing about the mechanism underneath, and that our instinct for detecting understanding is not calibrated for systems built to produce the appearance of it.
Phantoms in the Brain does something adjacent from the clinical side: a computational read on what consciousness might be in the context of an entity's intelligence and knowledge determinism, arrived at by looking closely at what happens when specific pieces break. Lesions are experiments you would never be permitted to run, and the deficits are far stranger and far more specific than a unified-self model predicts.
Together they leave you with a working stance: be suspicious of surface competence, insist on traceable mechanism, and take seriously that the thing doing the reasoning has an architecture, and the architecture shows.
Where this goes
I've kept walking down the computational path (learning how these systems actually operate, at the level of the mechanism rather than the demo) with a specific destination in mind. The point was never to leave neuroscience for software. It was to build enough capability with computational systems to translate it back into neurotech, and get a higher-throughput handle on universal knowledge than a single brain can support.
That's a long project and I'm somewhere in the middle of it. But the shape of it hasn't changed since I was arguing with department chairs about a major that didn't exist. The silos are still an administrative fact. The common threads are still where the advances are. And I still think the interesting work is in the places the curriculum doesn't have a name for.