New studies tagged cognitive science and learning, October 2026
By Matthew Rhoads, Ed.D.
Evidence levels in this roundup are assigned by our published, automated rules, except where a grade note says a person changed it. How we grade
This roundup covers items with the Cognitive science and learning Field tag, applied by keyword rules when an item discusses that area. This month we pulled every item in The Evidence Signal tracker tagged Cognitive science and learning and selected 10 of its 11 items at Level 2 or higher, plus two Level 1 items and one framework paper.
The tracker currently holds 69 items with this tag: 3 at Level 3: Controlled comparison, 8 at Level 2: Correlational, 17 at Level 1: Descriptive or exploratory, 14 at Technical test, 23 at Commentary, 4 at Not yet classified. See them all on the Cognitive science and learning topic page. One Level 2 item on the tracker, a Tai Chi study with men with methamphetamine use disorder, is left out of this roundup because it measured attentional bias, not learning.
The strongest evidence we track
Generative AI Availability, Grades, and Student Satisfaction at a Large University
arXiv (back catalog), 2026-07-23. Level 3: Controlled comparison. Labels: Learners: Higher-ed students, Preprint.
We then use a difference-in-differences design comparing courses before and after ChatGPT's release, modeling COVID-19 effects as persistent or transient. Effects on grades and satisfaction are small and sensitive to COVID modeling choices, with no detectable differences by students' prior academic preparation.
Quoted from the abstract at arXiv (back catalog).
AI Assistance Reduces Persistence and Hurts Independent Performance
arXiv (back catalog), 2026-04-06. Level 3: Controlled comparison. Labels: Learners: Adult learners, Preprint, Short exposure.
Here, through a series of randomized controlled trials on human-AI interactions (N=1,222), we provide causal evidence for two key consequences of AI assistance: reduced persistence and impairment of unassisted performance. Across a variety of tasks, including mathematical reasoning and reading comprehension, we find that although AI assistance improves performance in the short-term, people perform significantly worse without AI and are more likely to give up.
Quoted from the abstract at arXiv (back catalog).
Wakeful Targeted Memory Reactivation during Short Rest Periods Modulates Early Motor Learning
ERIC (peer reviewed), 2026-03-02. Level 3: Controlled comparison. Labels: no further labels.
To examine the motor learning effect of cue structure, we compared motor learning across three groups (TMR[subscript regular], TMR[subscript fast], and TMR[subscript random]). The TMR[subscript fast] group enhanced early learning gains compared with the TMR[subscript regular] group.
Quoted from the abstract at ERIC (peer reviewed).
How assigned AI use before class shapes active student engagement in class
arXiv (back catalog), 2026-10-07. Level 2: Correlational. Labels: Learners: Higher-ed students, Preprint, Preregistered.
Here, we report the results of a preregistered field experiment with 759 MBA students enrolled in ten sections of a course, in which each student was randomly assigned two of ten class sessions to prepare for with a purpose-built voice-based AI discussion partner. After two uses of the AI discussion partner, students made about 31% more voluntary contributions in each later class session. Students who used the AI discussion partner more also reported greater comfort speaking up and greater perceived learning, but not greater focus or motivation.
Quoted from the abstract at arXiv (back catalog).
When AI Tutors Speak: Evidence from a Randomized Field Experiment
arXiv (back catalog), 2026-09-21. Level 2: Correlational. Labels: Learners: Higher-ed students, Preprint, Preregistered, Small or imprecise.
Within the tutored arm, each student's channel alternated weekly between voice and text, so the modality effect is identified within student. Structure mattered: tutored students gained 6.63 points more than ability-matched peers (p=.007), and the gain was concentrated in written reasoning, where the share of answers reaching relational quality rose from 8% to 49% in the tutored arm against 8% to 27% in the holdout. Modality did not matter for learning. The instructor's own final, on file for all 86 randomized students, shows the same direction (2.6 points of 100, with no difference on a pre-treatment midterm).
Quoted from the abstract at arXiv (back catalog).
A Semi-Automated System for Generating Dialogue-Based TTS Lessons Using Large Language Models: An Exploratory Study of Educational Potential
arXiv (back catalog), 2026-07-14. Level 2: Correlational. Labels: Learners: K-12 students, Preprint, Short exposure, Researcher-made measure.
TTS audio did not substantially degrade the learning experience versus instructor voice, supported by TOST equivalence testing. Dialogue TTS was significantly superior to single TTS in comprehension (p=.006, q=.025) and cognitive engagement (p=.019, q=.048); enjoyment was non-significant after FDR correction (q=.081) but reached significance after controlling for prior knowledge (proportional-odds model, OR=1.65, q=.025), and these advantages were not attributable to prior-knowledge imbalance. Conversely, single TTS was superior in audio naturalness (p<.001, q<.001, r=-.238), revealing a trade-off between dialogue's benefits and higher extraneous cognitive load. Dialogue format was preferred by 66.9% of learners as most enjoyable (p<.001). These results reflect a fixed-order design; replication is needed before generalizing them as effects of lesson format.
Quoted from the abstract at arXiv (back catalog).
Beyond Information Retrieval: Generative AI as an Epistemic Arbiter to Enhance Collaborative Problem-Solving
arXiv (back catalog), 2026-06-25. Level 2: Correlational. Labels: Learners: K-12 students, Preprint.
Chi-square analysis showed significant differences in CPS behavior distributions between groups. Compared with the control group, the GAI-supported group demonstrated more social behaviors, particularly engagement and conflict management, but less frequent cognitive behaviors such as task planning and solution reasoning. Lag sequential analysis further revealed distinct interaction patterns: while the control group followed a more conventional transition from listening to planning, the GAI group showed a robust pathway from task planning to conflict management to solution reasoning.
Quoted from the abstract at arXiv (back catalog).
Do Gains from Generative AI-Enabled Adaptive Pretesting Persist? Evidence from a Retention Study
arXiv (back catalog), 2026-06-21. Level 2: Correlational. Labels: Learners: Higher-ed students, Preprint, Small or imprecise. A version also appears in the AIED 2026 Late Breaking Results proceedings (DOI 10.1007/978-3-032-29788-4_49), published online 2026-06-27; our grade has not yet been rechecked against that version.
Undergraduate participants completed an adaptive AI-assisted pretesting session, received instruction, and took a baseline assessment, then were randomly assigned to adaptive spaced retrieval practice, fixed spaced retrieval practice, or learner-directed AI-supported study. Multivariate analyses revealed a significant effect of condition on posttest performance and observed practice effort, with retrieval-based conditions outperforming learner-directed study.
Quoted from the abstract at arXiv (back catalog).
Self-Efficacy and Favorability Shape Learning from Tutoring Systems and Paper Practice
arXiv (back catalog), 2026-06-16. Level 2: Correlational. Labels: Learners: K-12 students (97 ninth-grade mathematics students, per the full paper), Preprint, Small or imprecise. A version also appears in the EC-TEL 2026 proceedings, Mindful TEL: Learning Technologies Shaped with Intention (DOI 10.1007/978-3-032-37979-5_10), published online 2026-10-07; our grade has not yet been rechecked against that version.
Using a counterbalanced within-subject design with matched problem sets, we isolate the effect of practice format while modeling motivational differences. Results indicate that students with lower baseline self-efficacy achieved greater learning gains regardless of practice format. Among students with lower baseline self-efficacy, greater favorability toward the tutor was associated with greater learning gains during tutor practice, whereas the pattern differed in paper-based practice. Intelligent Tutoring System (ITS)-based practice did not significantly improve post-training self-efficacy relative to paper-based methods.
Quoted from the abstract at arXiv (back catalog).
Fostering Dialogic Peer Feedback in Synchronous Online Discussions: The Effect of Argumentation Scaffold and Role Assignment
ERIC (peer reviewed), 2026-06-05. Level 2: Correlational. Labels: no further labels.
To address this issue, this study investigated the effects of argumentation scaffold and role assignment on students' emotion, social metacognition, and discourse patterns in dialogic peer feedback. Results showed that argumentation scaffold enhanced students' positive emotion, social metacognition knowledge, and social metacognition judgment, whereas adding role assignment reduced these benefits.
Quoted from the abstract at ERIC (peer reviewed).
Also worth a look (Level 1, descriptive)
Perceptions, practices and governance of generative AI in university social science teaching in Ecuador
Frontiers in Education, 2026-10-19. Level 1: Descriptive or exploratory. Labels: no further labels.
The results show a moderate-to-high level of AI use (mean = 3.68), explained by perceived usefulness and extrinsic motivation rather than ease of use, and a gap between the demand for regulation (4.00) and perceived regulatory clarity (3.21); only one of the seven universities has a policy formally designated as such, and ethics predominates over professional and disciplinary development.
Quoted from the abstract at Frontiers in Education.
Connectivism as an Innovative Didactic Experience in Preservice Physics Teacher Education
arXiv physics.ed-ph, 2026-10-19. Level 1: Descriptive or exploratory. Labels: Preprint.
The chapter discusses the possibilities of connectivist learning networks for physics teacher education, drawing on the findings of this particular classroom experience.
Quoted from the abstract at arXiv physics.ed-ph.
Frameworks and commentary
What Human-AI Interaction Leaves Behind: Residual Human States - A Trajectory-Based Framework and Multi-Study Analysis
arXiv search: AI + education, 2026-10-13. Commentary (moved from Level 1 on 2026-10-21 after a full-paper check approved by the editor: it introduces a framework and illustrates it with four cases, including published medical-training evidence and a reanalysis of earlier data). Labels: Preprint.
Published medical-training evidence illustrates how one AI interaction history can be associated with both slightly higher observed accuracy on some later unaided cases and transmission of a systematic decision bias on others. A second diagnostic-learning study shows that later unaided behaviour differs across AI-support policies, while also revealing that aggregate accuracy cannot by itself distinguish improved discrimination from a shifted response criterion.
Quoted from the abstract at arXiv search: AI + education.
What this set of studies does and doesn't tell us
- 10 of the tracker's 11 items at Level 2 or higher are listed above. None holds Level 4, which our method awards only after a person reads the full paper.
- 8 of those 10 are listed on arXiv. Two of the 8 also have conference proceedings versions (the self-efficacy and adaptive-pretesting studies), which we have not yet graded.
- These levels are set by our automated rules. Apart from the framework paper above, which was moved to Commentary after a full-paper check approved by the editor, no level here has been re-checked against a full paper. The self-efficacy card uses a sample size taken from its full paper.
- A tag means an item discusses this topic, not that it is good evidence about it. Each study tested its own question with its own learners, so these results should not be added together into one conclusion.
- Level 2 items are either correlational studies or experiments that our rules lowered for a specific reason, such as a small sample, a non-learning outcome, or no reported starting equivalence. Check each item's labels before drawing a conclusion about cause.
How we graded these
Every item links to its source above. Evidence levels describe study design and who took part, and they are not endorsements or official ratings. Read our method.