Re:How Shadow Banning Can Silently Shift Opinion Online
by asah on Tuesday May 21, 2024 @12:07PM
(#40430288)
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.
[ Parent
| on HN ]
Re:How Shadow Banning Can Silently Shift Opinion Online
by vasco on Tuesday May 21, 2024 @12:59PM
(#40430953)
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.
Re:How Shadow Banning Can Silently Shift Opinion Online
by janalsncm on Tuesday May 21, 2024 @02:08PM
(#40431751)
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.
Re:How Shadow Banning Can Silently Shift Opinion Online
by asah on Tuesday May 21, 2024 @07:31PM
(#40435530)
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.