2 ms·

While I agree that it seems very problematic that these ingredients are so accessible, I also have to challenge this: > They had decades to iron out these kink

While I agree that it seems very problematic that these ingredients are so accessible, I also have to challenge this: > They had decades to iron out these kinks. Aren’t the developments in algorithms that lead to these kinds of unexpected outcomes a bit more recent? This seems similar to the content moderation problems the social media giants are having. The sheer volume of products in Amazon’s marketplace is such that these kinds of things seem bound to happen when applying recommendation engines. This is not to imply that such recommendations are acceptable, but my primarily contention is that this doesn’t seem like a problem that has been well solved anywhere yet.

indymike 4y ago
> Aren’t the developments in algorithms that lead to these kinds of unexpected outcomes a bit more recent? No. Unexpected outcomes are pretty normal for search engines (and really, any complex computer system) since... forever, and really, the solution, regardless of input is is really easy: on input if (keyword in prohibited_phrases) disable recommendations, and on output if (keyword in recommendation_results) disable recommendations to user. I do believe this was solved sometime prior to E. F. Codd's 1970 "Relational Completeness of Data Base Sublanguages".
- haswell 4y ago
  I was referring to the product pairings, not the search result autocompletion (which does seem like a pretty glaring problem). Cutting off a search early is easy. Deciding that “even though these are commonly purchased together, this pairing shouldn’t be shown” seems like a fundamentally different problem.
  - indymike 4y ago
    Dropping results is easy to do, too. Figuring out what shouldn't be paired? That's a curation problem.
    - haswell 4y ago
      Agreed, and this is why I compared it to the social media curation problem. The thing I was replying to was the sentiment that this should have been solved decades ago. Curation/moderation is never solved, only managed.

indymike 4y ago

> Aren’t the developments in algorithms that lead to these kinds of unexpected outcomes a bit more recent? No. Unexpected outcomes are pretty normal for search engines (and really, any complex computer system) since... forever, and really, the solution, regardless of input is is really easy: on input if (keyword in prohibited_phrases) disable recommendations, and on output if (keyword in recommendation_results) disable recommendations to user. I do believe this was solved sometime prior to E. F. Codd's 1970 "Relational Completeness of Data Base Sublanguages".

haswell 4y ago
I was referring to the product pairings, not the search result autocompletion (which does seem like a pretty glaring problem). Cutting off a search early is easy. Deciding that “even though these are commonly purchased together, this pairing shouldn’t be shown” seems like a fundamentally different problem.
- indymike 4y ago
  Dropping results is easy to do, too. Figuring out what shouldn't be paired? That's a curation problem.
  - haswell 4y ago
    Agreed, and this is why I compared it to the social media curation problem. The thing I was replying to was the sentiment that this should have been solved decades ago. Curation/moderation is never solved, only managed.