3 ms·

Just out of curiosity, what do people mean when they say "publish the algorithm"? A poster-sized UML diagram? A million lines of spaghetti code saying "if foo

Just out of curiosity, what do people mean when they say "publish the algorithm"? A poster-sized UML diagram? A million lines of spaghetti code saying "if foo > 11.78 then bar += 2.6"? An exhaustive design document exactly specifying the entire ranking system (which absolutely does not exist at any company)? The ranking signals the algorithm operates on?

munksbeer 2y ago
> Just out of curiosity, what do people mean when they say "publish the algorithm"? In whatever form it takes such that a regulator can scrutinise it. If that seems arduous or impacts the social media business model, too bad. Nothing of value will have been lost. I am not typically a person who favours over regulation, even my views on AI are likely too liberal for HN, but on the matter of social media impacting children, I'm all in for regulation.
lupusreal 2y ago
They should have to publish everything required to independently reproduce the recommendations, given a user's inputs. If their code is spaghetti that makes this neigh impossible, that's their problem and they shouldn't be in business.
- gruez 2y ago
  >They should have to publish everything required to independently reproduce the recommendations, given a user's inputs. So they also have to give up all their user data?
  - lupusreal 2y ago
    > given a user's inputs Meaning if I record my own inputs, I should be able to recreate the suggestions I see using the published software. This doesn't mean the inputs of other users are exposed to me, but it does imply that "the algorithm" can only use data that is either publicly exposed or is my own inputs.
    - janalsncm 2y ago
      “The inputs” here is going to include every video, or at least a large subset of every video.
      - brokenmachine 2y ago
        Also how long a user spent watching each individual video, and what order they were watched in, and other info (whether they scrolled the view, etc). All that would be very hard/impossible to recreate in order to test the algorithm.
      - lupusreal 2y ago
        The videos themselves are public, and I think it would be good for all the relevant video metadata to be public as well.
    - gruez 2y ago
      >but it does imply that "the algorithm" can only use data that is either publicly exposed or is my own inputs. That seems pretty restricting given that even the most basic recommendation would try to cluster users based on interests and use that to recommend stuff (eg. Amazon's "customers who viewed this item also bought..." from decades ago).
asah 2y ago
speaking practically, I'd start with open source repo and (somehow) a chain-of-custody proving that this is the (only) code that gets to production.
- vasco 2y ago
  You assume recommendation systems are mostly code. AI models depend mostly on data, not code, and AFAIK we can see activation functions for neurons but nobody actually can explain in->out for the global thing.
  - janalsncm 2y ago
    In other words, you won’t be able to tell the difference between a net that prefers pro-Chinese propaganda and one that prefers pro wrestling. To make things worse, for 99% of people things might look normal but for 1% it may show a large skew. The net can have memorized this.
    - asah 2y ago
      I'm a bit more optimistic: the CI/CD pipeline can include adversarial model(s) which probes for the worst behavior. We already have codebases of sufficient complexity that attacks can be inserted into the software supply chain. Obviously, this becomes an arms race. The usual strategy is to add random checks, friction to contribution and severe-enough punishments for offenders (lifetime RL bans, etc.) Expert opinions welcome - I'm a tourist.

munksbeer 2y ago

> Just out of curiosity, what do people mean when they say "publish the algorithm"? In whatever form it takes such that a regulator can scrutinise it. If that seems arduous or impacts the social media business model, too bad. Nothing of value will have been lost. I am not typically a person who favours over regulation, even my views on AI are likely too liberal for HN, but on the matter of social media impacting children, I'm all in for regulation.

lupusreal 2y ago

They should have to publish everything required to independently reproduce the recommendations, given a user's inputs. If their code is spaghetti that makes this neigh impossible, that's their problem and they shouldn't be in business.

gruez 2y ago
>They should have to publish everything required to independently reproduce the recommendations, given a user's inputs. So they also have to give up all their user data?
- lupusreal 2y ago
  > given a user's inputs Meaning if I record my own inputs, I should be able to recreate the suggestions I see using the published software. This doesn't mean the inputs of other users are exposed to me, but it does imply that "the algorithm" can only use data that is either publicly exposed or is my own inputs.
  - janalsncm 2y ago
    “The inputs” here is going to include every video, or at least a large subset of every video.
    - brokenmachine 2y ago
      Also how long a user spent watching each individual video, and what order they were watched in, and other info (whether they scrolled the view, etc). All that would be very hard/impossible to recreate in order to test the algorithm.
    - lupusreal 2y ago
      The videos themselves are public, and I think it would be good for all the relevant video metadata to be public as well.
  - gruez 2y ago
    >but it does imply that "the algorithm" can only use data that is either publicly exposed or is my own inputs. That seems pretty restricting given that even the most basic recommendation would try to cluster users based on interests and use that to recommend stuff (eg. Amazon's "customers who viewed this item also bought..." from decades ago).

asah 2y ago

speaking practically, I'd start with open source repo and (somehow) a chain-of-custody proving that this is the (only) code that gets to production.

vasco 2y ago
You assume recommendation systems are mostly code. AI models depend mostly on data, not code, and AFAIK we can see activation functions for neurons but nobody actually can explain in->out for the global thing.
- janalsncm 2y ago
  In other words, you won’t be able to tell the difference between a net that prefers pro-Chinese propaganda and one that prefers pro wrestling. To make things worse, for 99% of people things might look normal but for 1% it may show a large skew. The net can have memorized this.
  - asah 2y ago
    I'm a bit more optimistic: the CI/CD pipeline can include adversarial model(s) which probes for the worst behavior. We already have codebases of sufficient complexity that attacks can be inserted into the software supply chain. Obviously, this becomes an arms race. The usual strategy is to add random checks, friction to contribution and severe-enough punishments for offenders (lifetime RL bans, etc.) Expert opinions welcome - I'm a tourist.