iiinova — where your digital self keeps evolving

Julian Skirzynski @jskirzynski.bsky.social · 06/01/2026

I recently gave a 15-min talk at #NeurIPS2025 on why "interpretable" AI doesn't automatically lead to better human decisions, and discussed my research on human-AI collaboration. Watch here: www.youtube.com/watch?v=JTuU...

youtube.com

Do We Make Better Decisions with AI? Human Bias & Interpretability

YouTube video by SAIL Media

000

Reposted by Julian Skirzynski

Petter Törnberg @pettertornberg.com · 07/11/2025

LLMs are now widely used in social science as stand-ins for humans—assuming they can produce realistic, human-like text But... can they? We don’t actually know. In our new study, we develop a Computational Turing Test. And our findings are striking: LLMs may be far less human-like than we think.🧵

arxiv.org

Computational Turing Test Reveals Systematic Differences Between Human and AI Language

Large language models (LLMs) are increasingly used in the social sciences to simulate human behavior, based on the assumption that they can generate realistic, human-like text. Yet this assumption rem...

14329131

Reposted by Julian Skirzynski

Dr. Damien P. Williams, dread portent down from a mountain cave @wolven.blacksky.app · 05/10/2025

Preliminary results show that the current framework of "AI" makes ppl less likely to help or seek help from other humans, or to seek to soothe conflict, and that people actively prefer that framework to any others, literally serving to make them more dependent on it.

arxiv.org

Sycophantic AI Decreases Prosocial Intentions and Promotes Dependence

Both the general public and academic communities have raised concerns about sycophancy, the phenomenon of artificial intelligence (AI) excessively agreeing with or flattering users. Yet, beyond isolat...

15455215

Reposted by Julian Skirzynski

Ezequiel Lopez-Lopez @eloplop.bsky.social · 28/07/2025

New research out!🚨 In our new paper, we discuss how generative AI (GenAI) tools like ChatGPT can mediate confirmation bias in health information seeking. As people turn to these tools for health-related queries, new risks emerge. 🧵👇 nyaspubs.onlinelibrary.wiley.com/doi/10.1111/...

nyaspubs.onlinelibrary.wiley.com

NYAS Publications

Generative artificial intelligence (GenAI) applications, such as ChatGPT, are transforming how individuals access health information, offering conversational and highly personalized interactions. Whi...

1159

Julian Skirzynski @jskirzynski.bsky.social · 24/06/2025

We’ll be presenting ‪@facct on 06.24 at 10:45 AM during the Evaluating Explainable AI session! Come chat with us. We would love to discuss implications for AI policy, better auditing methods, and next steps for algorithmic fairness research. #AIFairness #XAI

010

Julian Skirzynski @jskirzynski.bsky.social · 24/06/2025

But if they are indeed used to dispute discrimination claims, we can expect multiple failed cases due to insufficient evidence and many undetected discriminatory decisions. Current explanation-based auditing is, therefore, fundamentally flawed, and we need additional safeguards.

110

Julian Skirzynski @jskirzynski.bsky.social · 24/06/2025

Despite their unreliability, explanations are suggested as anti-discrimination measures by a number of regulations. GDPR ✓ Digital Services Act ✓ Algorithmic Accountability Act ✓ GDPD (Brazil) ✓

110

Julian Skirzynski @jskirzynski.bsky.social · 24/06/2025

So why do explanations fail? 1️⃣ They target individuals, while discrimination operates on groups 2️⃣ Users’ causal models are flawed 3️⃣ Users overestimate proxy strength and treat its presence in the explanation as discrimination 4️⃣ Feature-outcome relationships bias user claims

110

Julian Skirzynski @jskirzynski.bsky.social · 24/06/2025

BADLY. When participants flag discrimination, they are correct ~50% of the time, miss 55% of the discriminatory predictions and keep a 30% FPR. Additional knowledge (protected attributes, proxy strength) improves the detection to roughly 60% without affecting other measures.

120

Julian Skirzynski @jskirzynski.bsky.social · 24/06/2025

Our setup lets us assign each robot a ground-truth discrimination outcome, which lets us evaluate how well each participant could do under different information regimes. So, how did they do?

110

Julian Skirzynski @jskirzynski.bsky.social · 24/06/2025

We recruited participants, anchored their beliefs on discrimination, trained them to use explanations, and tested to make sure they got it right. We then saw how well they could flag unfair predictions based on counterfactual explanations and feature attribution scores.

110

Julian Skirzynski @jskirzynski.bsky.social · 24/06/2025

Participants audit a model to predict if robots sent to Mars will break down. Some are built by “Company X.” Others by “Company S.” Our model predicts failure based on robot body parts. It can discriminate against Company X by predicting that robots without an antenna fail.

110

Julian Skirzynski @jskirzynski.bsky.social · 24/06/2025

We cannot tell if explanations work or not due to these reasons. To tackle this challenge, we introduce a synthetic task where we: - Teach users how to use explanations - Control their beliefs - Adapt the world to fit their beliefs - Control the explanation content

110

Julian Skirzynski @jskirzynski.bsky.social · 24/06/2025

Users may fail to detect discrimination through explanations due to: - Proxies not being revealed by explanations - Issues with interpreting explanations - Wrong assumptions about proxy strength - Unknown protected class - Incorrect causal beliefs

120

Julian Skirzynski @jskirzynski.bsky.social · 24/06/2025

Imagine a model that predicts loan approval based on credit history and salary. Would a rejected female applicant get approved if she somehow applied as a man? If yes, her prediction was discriminatory. Fairness requires predictions to stay the same regardless of the protected class.

110

Julian Skirzynski @jskirzynski.bsky.social · 24/06/2025

Right to explanation laws assume explanations help people detect algorithmic discrimination. But is there any evidence for that? In our latest work w/ David Danks @berkustun, we show explanations fail to help people, even under optimal conditions. PDF shorturl.at/yaRua

180

Reposted by Julian Skirzynski

Harry Cheon @scheon.com · 24/04/2025

Denied a loan, an interview, or an insurance claim by machine learning models? You may be entitled to a list of reasons. In our latest w @anniewernerfelt.bsky.social @berkustun.bsky.social @friedler.net, we show how existing explanation frameworks fail and present an alternative for recourse

1157

Julian Skirzynski @jskirzynski.bsky.social · 08/12/2024

Actually, I've added you some time ago already so you're good :)

010

Julian Skirzynski @jskirzynski.bsky.social · 08/12/2024

Let's have bioinformatics represented then :) Regarding the clubs, I have not heard of any, might be just a coincidence :D

110

Julian Skirzynski @jskirzynski.bsky.social · 02/12/2024

Interesting stuff, welcome to the hood!

010

Julian Skirzynski @jskirzynski.bsky.social · 01/12/2024

You were in even before the request 8) Cheers

110

Reposted by Julian Skirzynski

Rose 🌹 @rose.bsky.team · 26/11/2024

You wanted starter packs to be searchable. Our engineers are busy keeping us online, so in the meantime, an independent developer built a new searchable library of starter packs. This is the beauty of building in the open 🦋

20939491289