For algorithms, anything that isn’t a straightforward scrutable way of presenting user content is bad, IMO. Algorithms that promote engagement, monetization, and sycophants are bad.

I would say scrutability in itself doesn't automatically make an algorithm good. "Demote everything that doesn't support Trump" is perfectly scrutable but leads to a skewed discussion.

In fact I would say any content boosting algorithm at all leads to skew and what you call sycophancy. That includes upvotes/downvotes that affect what posts users see first. So I would get rid of all that stuff and just show purely chronologically.

I haven't noticed much difference between instances either, though I haven't been on many. I moved from lemmy.world to lemmy.ml because .ml has a bit less censorship (e.g. .ml lets me subscribe to covid@hexbear.net). They are otherwise about the same, as far as I can tell.

[–] Glitchvid@lemmy.world 2 points 22 hours ago (1 children)

I would say scrutability in itself doesn’t automatically make an algorithm good. “Demote everything that doesn’t support Trump” is perfectly scrutable but leads to a skewed discussion.

This is mostly getting into normative vs descriptive philosophy. If it's scrutable that a site/instance is demoting everything non-aligned with a worldview; then on the Fediverse it's users' choice to leave (and part of 'community values').

In fact I would say any content boosting algorithm at all leads to skew and what you call sycophancy. That includes upvotes/downvotes that affect what posts users see first. So I would get rid of all that stuff and just show purely chronologically.

To some degree, yes. New Reddit is particularly bad about this, it actively buries unpopular replies (but it goes further, and doesn't just use upvotes) — Software like Lemmy is better, you can easily set Sort by New or sort by Top as the default. There's also no 'Karma' system that propagates across the site.

Sycophancy is a human trait, so it'll always emerge in social systems; but normatively, our systems should not cater to these negative traits (e.g. Twitter).

[–] solrize@lemmy.ml 1 points 21 hours ago* (last edited 21 hours ago) (1 children)

There are some Lemmy instances without downvoting, but none without upvoting. That affects what gets posted. Also it doesn't matter much what an individual instance does, since a lively community has users from lots of instances contributing. That's the point of federation, I thought.

[–] Glitchvid@lemmy.world 1 points 19 hours ago

Yeah, I mean that's true of any social space though, if you say something agreeable (definitionally) you're going to get agreement. If you view upvoting as consensus building (i.e "I like this" / "I agree") it's just a more concise representation of a reply saying as much.

But that is scrutable.

What becomes a problem is content getting surfaced/buried on non-scrutable metrics (typically engagement) — ragebait isn't anything new, online or in societies. But when algorithms target content that gets engagement, ragebait is naturally surfaced in higher proportions. Often time such platforms completely bury content or make it impossible to find something not explicitly surfaced (YouTube search for example is widely known to be terrible here, FB rabidly buries comments on posts).

WRT communities, there definitely are instances and communities with very different rules, values and expected behaviors. Federation allows communities to pick and choose what other communities they think they'll get along with. This includes banning individual remote users if they don't follow local rules, or defederating entirely if other instances have drastically different values.

The federation model as described does well by my metrics. I can pick an instance that shares my values, participate in communities (in the Lemmy technical sense) that share them as well — and largely avoid or choose not to engage with people from communities (in the instance sense) that I don't share values with. This is extending "freedom of association" to online spaces in a way that large platforms largely cannot and willingly do not enable.