4 ms·

> Forming these coils, loops, and sheets is what proteins generally do, but ‘why?’ doesn’t enter into it. How do we know the model hasn’t figured out some of t

> Forming these coils, loops, and sheets is what proteins generally do, but ‘why?’ doesn’t enter into it. How do we know the model hasn’t figured out some of the ‘whys’ somewhere in there?

bigdict 4y ago
Because it learns a conditional distribution. It doesn't work on figuring out why the distribution is the way it is.
lrem 4y ago
Your question fundamentally falls into the area of unanswerable philosophy akin to "do insects feel pain?" But there's a reasonable intuition suggesting that the answer to your question is "no". What we're looking at is a non-linear regression model reproducing the function (which according to the article isn't really a function, but that's above both my and the model's knowledge) from a gene sequence to a 3d structure. It is heavily meta-optimised, so the "why's" would only be in the model, if reproducing the process of folding the protein was the cheapest way to guess the structure (). Intuitively it introduces at least one extra dimension, so should be way more expensive than finding analogues among known sub-aspects of the function. Hence, I would expect none of the "why's" to be in there. Sadly, if any insight for the "why's" was there after all, we don't have a method to extract it anyway. Disclaimer: I work in Google, far away from DeepMind, have no internal knowledge on this.
- salty_biscuits 4y ago
  "Sadly, if any insight for the "why's" was there after all, we don't have a method to extract it anyway." This has been my central frustration with working in ML. People always expect a "why" to exist, and by why I mean a cogent narrative explanation to complex phenomena. Maybe there is no "why" like this for a bunch of physical phenomena, maybe it is just a bunch of low level intricate stuff interacting in complex ways. There might be an emergent model that you can get a useful predictive model for with an ML model, then people get mad because the prediction doesn't solve the real meta problem that they were expecting to solve via the sub problem (e.g. solve folding then get mad because folding itself turns out not to be super useful because we don't know which protein to target, solve image classification then get mad because that doesn't make it easy to make a self driving car, etc, etc). "More is different" is definitely an idea in physics that needs to propagate into other fields to temper our expectations.
sgt101 4y ago
The "why" is a bit of an odd question anyway - the structure is as as the structure is. It's like asking why "red hears a galaxy", just words.
- freemint 4y ago
  Well, no. A bunch of mechanism have models of lower complexity that have almost exactly the same predictive power but a completely different structures. Those higher order structures are the "why". Why did does a cube on a inclined plane start to slide? You act like the correct answer is "because the subatomic particles and space time in the light cone of the experiment made it that way" when one should expect "because the sin of the incline angle times mass times local gravity became bigger then the static friction between a cos(incline angle) times the original cube weight and the surface at no incline" which is a lot simpler.
  - visarga 4y ago
    When the process being studied has more moving parts than humans can grasp in working memory there comes a limit after which we can't understand it anymore. Maybe the nice clean abstract concepts we expect are not there. It could be perfectly correct and yet complicated like front-end spaghetti code.
    - tsimionescu 4y ago
      That is possible, yes. But that doesn't make the question of "why?" meaningless, as the GGP was suggesting.
      - visarga 4y ago
        I see neural nets as extensions of human thought - they explore the fuzzy depths where we can't grasp directly, like a microscope or a telescope enhance vision. They are enhanced correlation engines.
        4y ago
pelorat 4y ago
It's probably not even possible for a human to understand the "why"

bigdict 4y ago

Because it learns a conditional distribution. It doesn't work on figuring out why the distribution is the way it is.

lrem 4y ago

Your question fundamentally falls into the area of unanswerable philosophy akin to "do insects feel pain?" But there's a reasonable intuition suggesting that the answer to your question is "no". What we're looking at is a non-linear regression model reproducing the function (which according to the article isn't really a function, but that's above both my and the model's knowledge) from a gene sequence to a 3d structure. It is heavily meta-optimised, so the "why's" would only be in the model, if reproducing the process of folding the protein was the cheapest way to guess the structure (). Intuitively it introduces at least one extra dimension, so should be way more expensive than finding analogues among known sub-aspects of the function. Hence, I would expect none of the "why's" to be in there. Sadly, if any insight for the "why's" was there after all, we don't have a method to extract it anyway. Disclaimer: I work in Google, far away from DeepMind, have no internal knowledge on this.

salty_biscuits 4y ago
"Sadly, if any insight for the "why's" was there after all, we don't have a method to extract it anyway." This has been my central frustration with working in ML. People always expect a "why" to exist, and by why I mean a cogent narrative explanation to complex phenomena. Maybe there is no "why" like this for a bunch of physical phenomena, maybe it is just a bunch of low level intricate stuff interacting in complex ways. There might be an emergent model that you can get a useful predictive model for with an ML model, then people get mad because the prediction doesn't solve the real meta problem that they were expecting to solve via the sub problem (e.g. solve folding then get mad because folding itself turns out not to be super useful because we don't know which protein to target, solve image classification then get mad because that doesn't make it easy to make a self driving car, etc, etc). "More is different" is definitely an idea in physics that needs to propagate into other fields to temper our expectations.

sgt101 4y ago

The "why" is a bit of an odd question anyway - the structure is as as the structure is. It's like asking why "red hears a galaxy", just words.

freemint 4y ago
Well, no. A bunch of mechanism have models of lower complexity that have almost exactly the same predictive power but a completely different structures. Those higher order structures are the "why". Why did does a cube on a inclined plane start to slide? You act like the correct answer is "because the subatomic particles and space time in the light cone of the experiment made it that way" when one should expect "because the sin of the incline angle times mass times local gravity became bigger then the static friction between a cos(incline angle) times the original cube weight and the surface at no incline" which is a lot simpler.
- visarga 4y ago
  When the process being studied has more moving parts than humans can grasp in working memory there comes a limit after which we can't understand it anymore. Maybe the nice clean abstract concepts we expect are not there. It could be perfectly correct and yet complicated like front-end spaghetti code.
  - tsimionescu 4y ago
    That is possible, yes. But that doesn't make the question of "why?" meaningless, as the GGP was suggesting.
    - visarga 4y ago
      I see neural nets as extensions of human thought - they explore the fuzzy depths where we can't grasp directly, like a microscope or a telescope enhance vision. They are enhanced correlation engines.
      - 4y ago

pelorat 4y ago

It's probably not even possible for a human to understand the "why"