How do we fix content recommendation?

4 points by lordchair 1 year ago | 1 comment

I'm about to start a new job working on content recommendation at a FAANG company and I want to dig into ethical approaches to the problem. So far most techniques I have read about address privacy and bias.

Bias is a very tricky topic to handle without introducing my personal inclinations. If I were to follow my internal compass I would end up censoring the personality cult I see on the right but I also acknowledge that the inverse situation is terrible. My conclusion is that any individual's attempt at addressing bias will only introduce their own so I should stay out of the political realm.

Privacy is important but I don't think I'll be able to influence the company to collect less info and I'm not convinced that better privacy leads to better outcomes (cable news polarized us without personal information).

I’m looking for evidence-based strategies or research on how to ethically and effectively manage a recommendation system. Any insights, links to studies, or arguments would be greatly appreciated.

PaulHoule 1 year ago |

Search arXiv to start:

https://arxiv.org/search/cs?query=Recommend+Ethics&searchtyp...

I do not think there is a right answer in that a feed can be designed with many sorts of characteristics and still be seen as an attractive feed to the user. For instance a feed could show you fashionable upcycling, reuse, repair, thrifting, etc. or it could show you stuff you could buy new at Rodeo Drive , Wal-Mart, Temu, etc. If the feed creator cared they could collect train/test data and make a model that predicts that variable and be able to quantify it as a property of the feed that can be optimized for.

Interaction with the feed also creates a record (more so if the user writes something) of the subject’s mental state including both chronic and acute disorders, stress, anxiety, depression as well as psychosis symptoms.

That responsibility is awesome, it is hard to defend “doing nothing” with this information as an ethical policy but any intervention is likely to be ill-targeted, ineffective and ethically questionable in its own way.