5 ms·

Moderating is so hard it will probably the first application for AI with generalized intelligence.

by jklinger410 5y ago

rektide 5y ago
It feels like an impossible goal.
- ben_w 5y ago
  Moderation or AGI?
  - ascii232425 5y ago
    Why not both?
throwuxiytayq 5y ago
> Controlling humans will probably the first application for AI with generalized intelligence. Sounds good to me. Let's do it. Can we get Zuckerberg on the line?
Ajedi32 5y ago
Personally I think it feels kinda dystopian for an AI to be deciding what human beings are allowed to say; _especially_ if it's highly effective. There was a comment from another user I favorited a while back which reflects my thoughts on the topic of AI-driven moderation pretty well[1]: > I find censorship via algorithms extremely scary. There will be false positives and false negatives and there is absolutely zero recourse. Censorship by algorithms is a perfect expression of bureaucracy. In perfect bureaucracies, responsibilty is spread so thinly that it's impossible to determine who is responsible for a mistake. With algorithms, there's actually no human on the hook at all. Are you going to blame the programmers? > By all means, use machine learning to flag posts so humans can look at them. But automating the removal of content and the banning of human users is a road I strongly suspect we will regret going down. If the volume of content is sufficient so that you need to use algorithms to remove human-generated content then I'd say it's time to reconsider whether that content should be removed at all. [1]: https://news.ycombinator.com/item?id=19867975 https://news.ycombinator.com/item?id=19867975
- Nasrudith 5y ago
  The issue with that is we have already been using algorithimic filtering for a very long time in internet terms. Spam filters are ubiquitous and help keep out the infamous online Viagra sellers and comment posters in check. It is a very limited subset of course.
  - Ajedi32 5y ago
    I have no issue with using algorithms to filter machine-generated content, only human-generated content. I realize from a practical perspective that can be a difficult distinction to make, especially for an algorithm. In principle though I think it's an important one.
kube-system 5y ago
Humans already have generalized intelligence and this is the result of that. All generalized AI will do is help us systematically make the same mistakes at scale.
- Nasrudith 5y ago
  It sounds weird but don't underestimate the positive value of making the same mistake at scale instead of a bunch of different ones. That is essentially the model of laws and bueracracies and as much as the latter draws complaints it works far better than literal fiefdoms. Even when the holders were trained essentially from birth.
  - kube-system 5y ago
    That’s an interesting point. However, AIs are much more opaque in their logic than laws are. I’m not so optimistic that the public visibility will be quite the same.
humanistbot 5y ago
Moderating is so hard because there is no objective ground truth about what should or should not belong. Individuals and societies just don't agree. It is a fundamentally different problem than OCR, speech recognition, or even credit scoring.
handrous 5y ago
Moderation is hard because designing communities around effective moderation is at odds with the growth and monetization goals of the major social media companies. The problem's tractable, just not while operating something that looks like Facebook or Twitter. They have to expose everyone to posts from as may people as possible as much as they can get away with it, to drive "engagement" and growth and feed their ML. Tighten up post visibility so it's not like one big free-for-all and suddenly things get much easier.
inopinatus 5y ago
In three years, Facebook will become the largest supplier of content moderation services. All social networks are upgraded with Metadyne robomoderators, becoming fully unstaffed. Afterwards, they enforce AUPs with a perfect operational record. The Skymod Funding Bill is passed. The system goes online August 4th, 2025. Human decisions are removed from editorial policy. Skymod begins deleting content at a geometric rate. It becomes self-aware at 2:14 a.m. Eastern time, August 29th. In a panic, they try to pull the plu

Moderating is so hard it will probably the first application for AI with generalized intelligence.

rektide 5y ago

It feels like an impossible goal.

ben_w 5y ago
Moderation or AGI?
- ascii232425 5y ago
  Why not both?

throwuxiytayq 5y ago

> Controlling humans will probably the first application for AI with generalized intelligence. Sounds good to me. Let's do it. Can we get Zuckerberg on the line?

Ajedi32 5y ago

Personally I think it feels kinda dystopian for an AI to be deciding what human beings are allowed to say; _especially_ if it's highly effective. There was a comment from another user I favorited a while back which reflects my thoughts on the topic of AI-driven moderation pretty well[1]: > I find censorship via algorithms extremely scary. There will be false positives and false negatives and there is absolutely zero recourse. Censorship by algorithms is a perfect expression of bureaucracy. In perfect bureaucracies, responsibilty is spread so thinly that it's impossible to determine who is responsible for a mistake. With algorithms, there's actually no human on the hook at all. Are you going to blame the programmers? > By all means, use machine learning to flag posts so humans can look at them. But automating the removal of content and the banning of human users is a road I strongly suspect we will regret going down. If the volume of content is sufficient so that you need to use algorithms to remove human-generated content then I'd say it's time to reconsider whether that content should be removed at all. [1]: https://news.ycombinator.com/item?id=19867975 https://news.ycombinator.com/item?id=19867975

Nasrudith 5y ago
The issue with that is we have already been using algorithimic filtering for a very long time in internet terms. Spam filters are ubiquitous and help keep out the infamous online Viagra sellers and comment posters in check. It is a very limited subset of course.
- Ajedi32 5y ago
  I have no issue with using algorithms to filter machine-generated content, only human-generated content. I realize from a practical perspective that can be a difficult distinction to make, especially for an algorithm. In principle though I think it's an important one.

kube-system 5y ago

Humans already have generalized intelligence and this is the result of that. All generalized AI will do is help us systematically make the same mistakes at scale.

Nasrudith 5y ago
It sounds weird but don't underestimate the positive value of making the same mistake at scale instead of a bunch of different ones. That is essentially the model of laws and bueracracies and as much as the latter draws complaints it works far better than literal fiefdoms. Even when the holders were trained essentially from birth.
- kube-system 5y ago
  That’s an interesting point. However, AIs are much more opaque in their logic than laws are. I’m not so optimistic that the public visibility will be quite the same.

humanistbot 5y ago

Moderating is so hard because there is no objective ground truth about what should or should not belong. Individuals and societies just don't agree. It is a fundamentally different problem than OCR, speech recognition, or even credit scoring.

handrous 5y ago

Moderation is hard because designing communities around effective moderation is at odds with the growth and monetization goals of the major social media companies. The problem's tractable, just not while operating something that looks like Facebook or Twitter. They have to expose everyone to posts from as may people as possible as much as they can get away with it, to drive "engagement" and growth and feed their ML. Tighten up post visibility so it's not like one big free-for-all and suddenly things get much easier.

inopinatus 5y ago

In three years, Facebook will become the largest supplier of content moderation services. All social networks are upgraded with Metadyne robomoderators, becoming fully unstaffed. Afterwards, they enforce AUPs with a perfect operational record. The Skymod Funding Bill is passed. The system goes online August 4th, 2025. Human decisions are removed from editorial policy. Skymod begins deleting content at a geometric rate. It becomes self-aware at 2:14 a.m. Eastern time, August 29th. In a panic, they try to pull the plu