iiinova — where your digital self keeps evolving

“This highlights a core tension..of social media: frictions between users’ stated preferences and their behaviors..are exploited by engagement-maximizing algorithms to create runaway feedback loops of increasing value misalignment” Thanks @404ink.bsky.social @mjgault.bsky.social for the discussion!

161

Important caveat: We ran our study in October 2024 & platforms change over time. See the below tweet re X's tweak in July 2026. Beyond any particular design, our paper raises the question, if we can measure the values the algorithm amplifies, whose values should they amplify? x.com/nikitabier/s...

020

This effect could create a positive feedback loop and is especially strong for Democrats, who disagreed more with replied content, and in turn whose feeds featured more content that clashed with their values.

130

But why is this happening? It seems that the algorithm places more weight on replies — and we are often more likely to reply to content we disagree with, which the algorithm (mis)takes as a positive signal. It treats those rare replies as a stronger signal than the flood of value-aligned likes.

(This is despite the fact that the inventory of posts from following accounts reflects users’ self-stated values, and the misalignment is stronger for Democrats than for Republicans).

When contrasting these values to the values that the users actually hold, we find a disconnect: there's a negative correlation (misalignment) between users’ explicit values and the values in content the algorithm is more likely to amplify

We find that the FYP algorithm is amplifying conservation values such as rule conformity, societal security and tradition, while demoting self-transcendent values such as caring , dependability, as well as universal concern and preservation of nature.

We analyze N=715 participants' feeds and measure the values expressed in the posts (using the method from a recent ICWSM paper to understand which values are amplified (vs demoted) by X’s in-production FYP algorithm. bsky.app/profile/ziv-...

Do our social media algorithms correctly reflect our values? Our new article published today in @pnas.org shows that the answer is often not, and that the content that gets promoted into their ranked feeds is often actively counter to our values.

23012

We believe this opens the door to new ways to measure the values that underlay online discourse. If you are at ICWSM 2026 in LA come say hi! Thanks to co-authors @farnazj.bsky.social @tiziano.bsky.social Isabel Gallegos @dorazhao.bsky.social @jugander.bsky.social @mbernst.bsky.social

000

In evaluation, we find that people agree with labels from off-the-shelf LLMs less than a random other person! But fine-tuning and then applying our personalization method yields a 66% relative improvement in agreement compared to human-human agreement rates, leading to SOTA performance.

110

We collect 32K ground truth value expression annotations from 1K people on 5K representative social media posts using Schwarz Values. We then construct a personalization architecture for predicting value expressions by learning from a small number of informative calibration annotations per user.

110

Measuring values is key for aligning systems with people’s values, but value expression is fundamentally subjective, leading to divergent labels from different people. We find meaningful disagreement in value annotations across raters. See for yourself, do you think the example contains humility?

100