Just to offer a different PoV: this would benefit progress in the same direction, but IMO we need new math for life/natural intelligence to get us out of copying ML into neuroAI, e.g. claiming an engineering tool as THE way the brain works etc. Breakthrough Theoretical advances need broader search. - ThreadSky

neuroai.bsky.social • 102 days ago

Just to offer a different PoV: this would benefit progress in the same direction, but IMO we need new math for life/natural intelligence to get us out of copying ML into neuroAI, e.g. claiming an engineering tool as THE way the brain works etc. Breakthrough Theoretical advances need broader search.

Reposted from Patrick Mineault

Paul Middlebrooks asked what we need to prepare the next gen of scientists. @tyrellturing.bsky.social said that we need to first teach students hard skills earlier in their careers: math, physics, engineering, and then have them do biology experiments, instead of the opposite order. I agree 2/

Comments

danieljbutler.bsky.social•90 days ago

I get strong Ramon y Cajal vibes from this. Observed anatomy really closely, no math / physics background, came up with key hypotheses for the flow of information. There's something to be said for getting obsessed with observation before getting caught up in formalisms

neuroai.bsky.social•90 days ago

Yes! And to create such amazing art for science in the process of hand-drawing neurons.

whitetigerwarrior.bsky.social•99 days ago

Math seems to be the hardest part of any of the sciences. I was very fortunate to have had great teachers. However, I have noticed that others who can grasp some of the concepts so easily struggle with the math. My teachers made math fun.

charan-neuro.bsky.social•100 days ago

I agree. Also, it's stupid to call math, physics, etc. "hard". They are just concrete skills to learn and knowledge to memorize, and nowhere near sufficient to understand the link between behavior and neural activity. Engineers solve problems, real scientists identify the right problems.

aidanhorner.bsky.social•100 days ago

Lots of different people with different backgrounds done in different orders FTW!

neuroai.bsky.social•100 days ago

I like this a lot! And either way, I still think the dominance of physics and machine learning perspectives on how we view, abstract, and model living organisms and natural intelligence needs rethinking.

colinhales.bsky.social•100 days ago

That "new math" is right in front of everyone and is contacted directly by every brain cognition-related measurement ever made by neuroscience: the math of the actual signalling physics of the brain: electromagnetism. EM fields.

What does it take to get this to be acknowledged & discussed?

neuroai.bsky.social•100 days ago

I think the first step a carefully written theoretical paper with clear exposition, equations, simulations and peer review by experts. Because otherwise it’s not possible to evaluate. Also it’s possible that it’s not the ultimate answer and other scientists disagree.

colinhales.bsky.social•100 days ago

Its origins predate computers! The literature is huge. The standard model of particle physics confirms it.The brain is literally EM field from inside atoms up. Job done.

Me: a PhD, a book, multiple papers.

You want a missing link? EM. To dismiss it is, to be blunt, negligent.

I can help. Let me.

dlevenstein.bsky.social•101 days ago

I’d love to see neurotheory better integrated into intro undergrad classes, so students see it as a core part of the discipline rather than a specialized side quest. I think you’d end up with a next-gen of theorists ready to build new math rather than import tools from their “home” disciplines.

dlevenstein.bsky.social•101 days ago

Ie this needs to be addressed, at the undergrad level.

https://bsky.app/profile/gershbrain.bsky.social/post/3laqsbm2hzc2s

neuroai.bsky.social•102 days ago

…this may mean having students do different kinds of bio experiments across species + learn different math, before getting primed w physicist or ML intuitions about abstraction of life/natural intelligence. Assuming we’re totally on the right track w <100 year old theoretical tools is IMO misguided.

neuroai.bsky.social•102 days ago

…It boils down to whether we want to recruit people for incremental progress on a path we’re comfortable with, or if we need/want breakthroughs & paradigm shifts. I think we need the latter, & that requires learning from biology first, more than we have, rather than starting & staying in model land.

jcbyts.bsky.social•82 days ago

I think this will naturally emerge if we have more working scientist careers instead of just PIs and trainees. You have to learn all the existing frameworks before you can really innovate, and there is no career path that gives people the time to do that right now.

neuroai.bsky.social•82 days ago

Fair point. IMO many of us are making that career path by pursuing different disciplines in a long term open ended way. You’re right that it’s not common and there are costs. But I’d rather follow the path I care about even if it’s more difficult.

jcbyts.bsky.social•82 days ago

Completely agree. It is very hard to do that in academic research these days. At least it feels that way to me. I keep hoping someone will figure out how to restructure things, because “team science” needs more experts with time to do research

wiringthebrain.bsky.social•102 days ago

👏👍⬆️

kamdh.bsky.social•101 days ago

I sort of agree with you. The main thing I see leading to increased creativity isn't the set of tools per se but changing the structures of science (funding, expectations, etc) to get people to think deeply. Current forces are all poised against this, cf Slow Science movement

shahabbakht.bsky.social•101 days ago

I named my lab "snail" to also kinda reflect the Slow Science spirit, but now I have to worry that the snail might get crushed along the way. Moving to a less crowded path may help though.

kamdh.bsky.social•101 days ago

Awesome. It requires constant resistance to the pressures all around you. Did you read "The Slow Professor" yet? Working through it now. Fuerza

neuroai.bsky.social•101 days ago

I haven't! Your praise is nudging me to read it asap :D

neuroai.bsky.social•101 days ago

Oh That too! But I actually am talking about the currently dominant tools/metaphors and how the structure of funding and hiring and reviews enforces a very specific 1970s-physicist friendly or 1990s-CS friendly view of life and intelligence- if that makes sense!

kamdh.bsky.social•101 days ago

I am so indoctrinated in that worldview that it's hard to see beyond it 😵

Agreed that you *have* to learn the domain, biology, before building any models.

neuroai.bsky.social•101 days ago

But also learn math! Just not see biology through a physics or ML lens!

saiprasanna.in•92 days ago

This discussion reminds me of this article
https://tomasp.net/blog/2018/alien-lambda-calculus/

tyrellturing.bsky.social•101 days ago

It's an interesting perspective, but I don't actually agree.

1) I think new mathematical frameworks will only be generated by people who understand the existing ones. So, you're not giong to get the novel paradigm you're hoping for by not training new students in existing models.

tyrellturing.bsky.social•101 days ago

2) TBH, I'm sceptical of the claim that we need some kind of grand paradigm shift in the mathematics. I actually think we're making excellent progress, and have a "progressive research program" à la Lakotos working well, as articulated by Adrien and Tim in this paper:

https://www.nature.com/articles/s41583-023-00705-w

neuroai.bsky.social•101 days ago

Thanks Blake, but I'm not calling for "some kind of grand paradigm shift in the math" as you call it, but for a math for living things.
Also as I said your suggestion is good for incrementally advancing your "progressive research program". I'm saying it might not be the only/best search direction.

tyrellturing.bsky.social•101 days ago

Given that the models we're using (ANNs) were originally developed to model biological systems, not for ML, do our current frameworks not qualify as "math for living things"?

What's missing, in your opinion?

neuroai.bsky.social•101 days ago

I agree with this point and I don't see it as something that negates my PoV above :) I think the point I'm making is to not physics-wash or ML-wash people's lens into biology before you expose them to biology. Why not explain the math, say hey physics used this and ML used that, WHILE doing biology?

tyrellturing.bsky.social•101 days ago

Oh, yes, sure, I agree! 🙂

To clarify what I was arguing for in the session that @patrickmineault.bsky.social was describing above, I would like to see core biology/neuroscience curricula include fundamental math (multi-varaite calculus, lin algebra, etc.) as a *requirement*, rather than an option.

neuroai.bsky.social•101 days ago

Yeah that’s might point, plus I think people should be doing experiments and encounters with living biological systems while learning the math. While incremental progress matters, I worry the physicist & ML lenses cloud our vision & theories a bit too much for true progress when it’s needed.

macaskillaf.bsky.social•101 days ago

💯

manuelbaltieri.bsky.social•101 days ago

With “new math” do you mean some existing material in a maths book and ready to be used in bio/neuro, or entirely new concepts that aren’t written down yet?

neuroai.bsky.social•101 days ago

Either! But mostly let’s start with the the former, new Mathematical ways to conceptualize compared to the math we’re used to using.

Comments

Posting Rules

Reply