iiinova — where your digital self keeps evolving

Reposted by Talha Özüdoğru

Christoph Strauch @cstrauch.bsky.social · 02/10/2026

How much of a risk are LLM-generated bots for online behavioral experiments? We present results of an entirely natural language prompt-generated agent, capable of absolving any of the 26 experiments we tested it on. Most tasks are passed with data that can pass as human osf.io/preprints/ps... (1/)

59344

Reposted by Talha Özüdoğru

Ata Karagoz @atabk.bsky.social · 24/09/2026

I just had Opus 5.5 one shot a Human-Like mousetracking driver by looking at my code on cognition. Right now it's choosing left or right randomly. It's still not perfect but this took it literally a simple prompt and 1.5 mins to code a script. It can then be randomized and used across bot accounts.

3134

Talha Özüdoğru @talha-ozudogru.bsky.social · 30/03/2026

I agree; but this does not mean it is economically infeasible. Compensation rates in the EU or the US are high enough to make this time investment worthwhile elsewhere in the world, where minimum wages are much lower. Even small groups of online task workers could collaborate in this kind of fraud.

011

Reposted by Talha Özüdoğru

Richard Huskey @richardhuskey.bsky.social · 27/03/2026

1/n New preprint with Ziyu Zhao, @dougaparry.bsky.social, & @jacobtfisher.online Can an AI bot complete a live online reaction-time task & produce data that passes as human? We built an autonomous bot to take the Attention Network Test (ANT) in real time Preprint: doi.org/10.31234/osf...

Screenshot of a manuscript title page. Title: “An AI agent can complete the Attention Network Test with human-like behavioral signatures: Implications for the bot-or-not debate.” Authors: Richard Huskey, Ziyu Zhao, Douglas A. Parry, and Jacob T. Fisher, with university affiliations listed below. The abstract says an autonomous AI agent completed the Attention Network Test in real time and produced mostly human-like behavioral data. Across seven code revisions, the bot achieved attention network scores within published human norms, 95.8% accuracy, and reaction-time patterns showing positive skew and trial-to-trial autocorrelation. Compared with 796 human participants, the bot fell within the human range on several measures but showed elevated autocorrelation and a bimodal reaction-time distribution due to intermittent detection failures. The paper argues this makes simple bot-vs-human detection harder in online reaction-time studies.

22112

Talha Özüdoğru @talha-ozudogru.bsky.social · 18/03/2026

Nice to see Prolific taking agentic AI detection seriously. It seems detection only works when Qualtrics + Prolific are used together. Any plans to include platforms like Gorilla? Our recent commentary shows behavioral studies are also vulnerable to agentic AI, not just surveys osf.io/3cztr/overview

osf.io

OSF

130

Reposted by Talha Özüdoğru

Christoph Strauch @cstrauch.bsky.social · 18/03/2026

Learnings for us (besides not to rely on shitty bot-detectors): - building bots that solve specific tasks is relatively easy with a base prompt - human-like performance is harder, but not impossible (Figure: Stroop data) - text heavy, slow-paced experiments more vulnerable Caution is appropriate.

Responses on an online Stroop task. A bot achieved average RT and accuracy that is in line with recent papers on the Stroop task.

101

Reposted by Talha Özüdoğru

Christoph Strauch @cstrauch.bsky.social · 18/03/2026

We recently warned of bots in online behavioral research. @achetverikov.bsky.social showed there is no evidence for that in our @joinprolific.bsky.social data - but that doesn't mean we're safe. Agentic AI can do behavioral tasks through prompting alone. Reply & videos: osf.io/3cztr/overview

32112