https://bsky.app/profile/samerp2p.bsky.social/post/3m5jwqgv7ak2g

#AIES #AIethics #AI #AIES2025

And that's it! There were many more great talks, but sadly I could only attend a limited amount of them. Please check their papers, whose abstracts communicate their work much better than me

Plenty of speakers, way more than I'd expect, were showing a slide blaming Google for aiding the Palestinian genocide, providing a QR link to the UN report, and highlighting Google is a sponsor of the conference.
That was brave of these mostly young researchers, and they have my admiration

I'm happy to highlight our
Pedagogy Panel:
"How (and to Whom) Do We Teach AI Ethics?"
which I shared with brilliant people: @shannonvallor.bsky.social and Julienne LaChance, chaired by @emanuelleburton.bsky.social

We tackled practical challenges on educating the conference topics to the new gen

"How LLM Counselors Violate Ethical Standards in Mental Health Practice: A Practitioner-Informed Framework"
by Zainab Iftikhar et al

A brilliant talk about the malpractices of current LLMs acting as therapists, highlighting the AI is constantly lying when saying "it cares" (since obv it can't)

"No Thoughts Just AI: Biased LLM Hiring Recommendations Alter Human Decision Making and Limit Human Autonomy"
by @kyrawilson.bsky.social et al

It shows how people trust AI hiring proposals blindly, even when the AI's being racist. This highlights the importance to tackle algorithmic bias

"AI Self-preferencing in Algorithmic Hiring: Empirical Evidence and Insights"
by @jiannanxu.bsky.social et al

A surprising bias:
- If a job candidate building their CV uses the same LLM as the evaluator, there is a strong bias to be selected
- And LLMs prefer AI-generated CVs over manual ones

"Govern With, Not For: Understanding the Stuttering Community’s Preferences and Goals for Speech AI Data Governance in the US and China"
by Jingjin Li et al

On an often invisibilized marginalized community: the stuttering community

"Disability Across Cultures: A Human-Centered Audit of Ableism in Western and Indic LLMs"
by @mahika.bsky.social et al

was fascinating, exploring how different LLMs categorize ableist comments in different degrees, depending on the language they were written (English vs Hindi).

I enjoyed a lot:
"Documenting Patterns of Exoticism of Marginalized Populations within Text-to-Image Generators"
by Sourojit Ghosh et al

Unsurprisingly, AIs show stereotypical imagery when trying to represent Global South cultures, to the point of absurdity

Work by our PhD student David Davó, post-doc @silviasemenzin.bsky.social, supervised by professor Javier Arroyo and me
from our institutions @ucm.es and Berkman Klein Center @harvard.edu
Article open licensed with CC-BY

We extract a “DAO fragility schema” and recommendations for the next generation: progressive governance by default, anti-plutocracy voting, gas-free voting, non-transferable time-decaying reputation, health dashboards, and redundant open-source UIs.
Builders: please check out the checklist.

Why did DAOstack fade despite early hype? Fee spikes, UX friction, front-end fragility, thin onboarding, and incentive misalignment.
Our interviews surfaced frictions: onboarding, cost/complexity of voting, informal power imbalances via off-platform coordination.

Attention is considered the scarce resource. Their Holographic Consensus governance model tried to solve it via prediction-market “boosting.” It filtered well (≈97% precision) but failed to sustain independent stakers. In fact, 83% of the stakes came from insiders

DAOstack, like other DAO platforms, promised scalable cooperation and democratic governance.
Reality check: most DAOs are tiny, short-lived, and oligarchic-leaning as they grow. In fact, far from democratic:
• ~65% members never voted
• ~8% avg voter turnout/proposal

We analyzed DAOstack's 92 organizations in terms of size, lifespan, activity, power concentration, and governance model effectiveness. And interviewed core users to better understand their experience.

DAOs aren't just code, but also communities. Technical design is political.
From “WordPress for DAOs” to deprecation: DAOstack’s life cycle offers a rare, data-rich case study. Ideal to see the social implications of techno-determinist views common in #Web3, and theoretical economic incentives.