Research & Science
Peer-reviewed work, preprints, technical methods, and empirical findings.
-
Research & ScienceRead story
When AI does the work, what skill remains?
A model of AI-assisted work suggests that easy reliance may leave new workers with less practice and experienced workers with fading skills. It is a theory, not a workplace trial.
arXiv cs.HC · Nov 5, 2026 -
Research & ScienceRead story
An AI score can stay steady while its decisions move
A study of AI labels for UK public questions found that repeated runs changed some labels, even when overall accuracy looked similar. A single score missed that shift.
arXiv cs.CY · Nov 5, 2026 -
Research & ScienceRead story
AI transparency depends on who needs the answer
A study asks what different people need to check an AI system. A label alone may not give users the source, control, or human route they need.
arXiv · Nov 3, 2026 -
Research & ScienceRead story
Shorter AI alerts worked faster in one timed test
In a timed study with 114 people, shorter alerts helped participants spot more problems and respond sooner. They were nearly as reliable as longer alerts. The test does not show that short warnings are safe for new operators or in emergencies.
arXiv cs.HC · Nov 4, 2026 -
Research & ScienceRead story
Change the AI judge, change the ranking
Researchers used five AI models to judge 4,000 chatbot matchups. The judges often disagreed with a single human vote, and rankings changed by judge. One better pair worked mainly because it left out a poor judge. This does not validate AI grading in schools.
arXiv cs.CL, cs.CY cross-list · Nov 4, 2026 -
Research & ScienceRead story
AI tests may miss the harder part of the job
A preprint reviews AI tests and maps the work a team must do around a task. Fewer than a quarter of prominent tests claimed to cover any team-level demands. That count describes tests, not how often AI fails at work.
arXiv · Nov 2, 2026 -
Research & ScienceRead story
An AI visit-prep tool helped in a study, not a diagnosis
In a randomized study, adults used an AI tool before medical visits. Clinicians judged more completed sessions to be well prepared with the structured tool than with a basic version of the same model. Patients also reported feeling more ready. The two measures used different groups of responses.
arXiv cs.HC · Nov 2, 2026 -
Research & ScienceRead story
Working with AI may beat working alone, but not the best solo option
A research synthesis reports that people working with AI beat people working alone on average. Yet those teams did worse than the stronger of a person or AI working alone. Its latest stated run combines 1,955 measured results from 420 papers; the digest did not appraise each study.
arXiv · Oct 30, 2026 -
Research & ScienceRead story
A private reflection step helped group discussion in one study
A preprint reports a randomized U.S. study of 1,173 people in 391 groups. AI-guided private reflection before discussion improved measured discussion quality, especially when people disagreed. The finding concerns the tested group sessions, not school or workplace results.
arXiv · Oct 27, 2026 -
Research & ScienceRead story
Critical thinking may depend on what people change after AI
A preprint studied 891 college students in the United States and Türkiye. It linked critical-thinking measures with how students coordinated work with AI and reworked its output. These are modeled links, not proof that a new exercise would raise critical-thinking scores.
arXiv · Oct 28, 2026 -
Research & ScienceRead story
When an AI has two answers, it may need to say neither
Van and Tran test what happens when a source disagrees with what an AI already knows. Their method can choose either answer or hold back. Its test gains partly depend on knowing how often the source is right, a number real users may not have.
arXiv cs.AI · Oct 26, 2026 -
Research & ScienceRead story
A community bot may see people who never chose it
Ampomah and Le study user reviews of third-party bots on Discord. Their research abstract points to privacy concerns for members who did not install a bot but may still have data visible to it. Reviews do not prove what any one bot did.
arXiv · Oct 27, 2026