The Commonplace
Home Dashboard Papers Evidence Syntheses Digests 🎲

Papers (1134)

Filter: Sort:
By 2025–early 2026, agentic and multimodal AI moved from demonstration to broad early adoption, with agentic coding tools beginning to automate end-to-end software workflows; U.S. policymakers have accelerated standards and procurement oversight but have not passed a single unified AI Act.
n/a commentary openalex
Emerging AI Trends · Maikel Leon, Robert Plant · May 06, 2026
The AI boom reshapes, but does not uniformly strengthen, global market linkages: overall connectedness fell modestly after ChatGPT's launch, with AI stocks losing their initial role as net shock transmitters while the S&P 500 stayed the primary source of spillovers.
medium correlational openalex
Artificial Intelligence and Financial Market Connectedness: Evidence from AI-Related Equities, Cryptocurrencies, and Global Assets · Shigeyuki Hamori · May 06, 2026
Decomposing AI recommendations into individually verifiable claims sharply raises clinician trust—trust rates jump from 27% to 66% and the effect is very large (d = 0.94). Traditional transparency tools produce only modest, dose‑response improvements.
high rct arxiv
Atomic Fact-Checking Increases Clinician Trust in Large Language Model Recommendations for Oncology Decision Support: A Randomized Controlled Trial · Lisa C. Adams, Linus Marx, Erik Thiele Orberg, Keno Bressem, Sebastian Ziegelmay... · May 05, 2026
Optimizing models for agent-facing interpretability substantially boosts autonomous data‑science agents: evolved regressors both improve prediction quality and make models more 'simulatable' by LLMs, increasing ADS performance on the BLADE benchmark (Copilot CLI, Claude Code, Codex) by as much as 73%. Gains are demonstrated on tabular tasks using an LLM-graded interpretability metric and may not directly translate to human interpretability or broader real-world workflows.
medium descriptive arxiv
Agentic-imodels: Evolving agentic interpretability tools via autoresearch · Chandan Singh, Yan Shuo Tan, Weijia Xu, Zelalem Gero, Weiwei Yang, Michel Galley... · May 05, 2026
U.S. workforce retraining largely restores wages but rarely shifts workers away from automation-prone jobs; employer-led apprenticeships show the biggest payoff. The WIOA system appears to support short-term wage recovery for vulnerable workers but is not structured to deliver large-scale cross-industry transitions away from routine, automation-exposed work.
medium correlational arxiv
Did US Worker Retraining Reduce Participant Automation Exposure? · Julian Jacobs, Jordan Canedy · May 05, 2026
A large new benchmark of real-world file workspaces finds AI agents far from human-level performance: the best system scores 68.7% against a human 80.7%, with an average across agents of just 47.4%, exposing persistent weaknesses in cross-file reasoning and workspace-level decision-making.
n/a descriptive arxiv
Workspace-Bench 1.0: Benchmarking AI Agents on Workspace Tasks with Large-Scale File Dependencies · Zirui Tang, Xuanhe Zhou, Yumou Liu, Linchun Li, Weizheng Wang, Hongzhang Huang, ... · May 05, 2026
Adoption of advanced, agent-like AI is linked to measurable productivity and growth gains but disproportionately benefits richer countries; developed economies capture markedly larger growth effects than emerging markets, risking reinforcement of existing global inequalities.
medium correlational semantic_scholar
The Economic Value of Agentic AI: A Comparative Analysis of Its Impact on Growth and Business Productivity in Developed and Emerging Economies · Abayomi Titilola Olutimehin, Oluwadayo Mafolasere Olaniyi, Adegbenga Ismaila Ala... · Fetched May 05, 2026
Leading AI-data firms frame human expertise as a low-cost, extractable input to power models and pitch institutional ‘liberation’ to unlock talent for AI — a narrative that could reshape white-collar labour markets and the value of expertise.
low descriptive arxiv
Cheap Expertise: Mapping and Challenging Industry Perspectives in the Expert Data Gig Economy · Robert Wolfe, Aayushi Dangol · May 05, 2026
China’s pilot water-resource tax raised grain output by improving water quality and spurring green innovation, with the biggest gains outside core grain belts and where digital-economy development and environmental regulation are stronger.
medium quasi_experimental openalex
Can water resource tax reform increase grain yield?—Evidence from China · Yashu Qin, Shiyao Yuan, Yue Wang, Luwei Wang · May 05, 2026
Automation reshapes jobs by eroding tasks, not eliminating occupations: routine automation tightens mobility within manual clusters, while AI-driven cognitive automation forges new bridges across diverse occupations, increasing skill overlap and potential labor mobility.
low theoretical openalex
Contrasting pathways of automation: routine task substitution vs. AI complementarity · Soohyoung Lee, Dawoon Jeong, Jeong‐Dong Lee · May 05, 2026
China’s low-carbon city program unexpectedly accelerated firms’ shift to smart manufacturing, with stronger effects in resource-light cities and among financially constrained traditional manufacturers; rising local human capital and better resource allocation account for much of the adoption increase.
medium quasi_experimental openalex
Environmental policy synergies: unintended benefits for smart manufacturing in China’s low-carbon cities · Jiaomei Tang, Wen Gao, Yayun Ren · May 05, 2026
AI is set to transform politics and political science: a few corporations will wield outsized power, opaque and biased models will strain democratic institutions, and scholars must urgently develop standards for evaluation, governance, and teaching.
n/a review_meta openalex
Introduction: Artificial Intelligence, Politics, and Political Science · Nathaniel Persily, Joshua A. Tucker · May 05, 2026
AI can materially boost logistics efficiency through better forecasting, routing and inventory decisions, but real-world gains are often limited by high upfront costs, data-security and governance challenges, and organizational resistance to change.
low review_meta openalex
Evaluating the Role of Artificial Intelligence in Optimizing International Logistics and Distribution Networks · Diar Rachmat · May 05, 2026
SMEs in one of the EU's poorest regions say lack of skills, money and social networks — not just technology — blocks AI adoption; without targeted investment in digital literacy and civic intermediation, AI risks entrenching rather than alleviating local employment inequalities.
low descriptive openalex
Artificial Intelligence, Social Capital, and Sustainable Employment in Peripheral SMEs: A Biocultural Reading from Eastern Macedonia and Thrace, Greece · Eugenia P. Bitsani, Αντώνιος Κώστας, Vasileios Kapilidis, Theophilos Gerasimidis... · May 05, 2026
AI commoditizes standardized cognitive work and inflates premiums for verifiable human presence; policymakers should treat human-provenance verification as critical labour infrastructure, not a luxury label.
n/a theoretical arxiv
Human-Provenance Verification should be Treated as Labor Infrastructure in AI-Saturated Markets · Erin McGurk, David Khachaturov · May 04, 2026
Hyperscale AI training sites threaten to break the electric grid's 'load diversity' assumption, forcing data‑center and power industries to co‑develop infrastructure, control protocols, and market designs. Without deliberate coordination, cultural and economic misalignment between the sectors risks higher costs and reduced reliability as AI scales.
n/a commentary arxiv
From Barrier to Bridge: The Case for AI Data Center/Power Grid Co-Design · Noman Bashir, Rob Sherwood, Le Xie, Minlan Yu · May 04, 2026
A hybrid rules-plus-learning system lets managers tune how work is split between humans and AI, converting autonomous AI actions into supervised collaborations as governance tightens. In benchmarked software and manufacturing scenarios, stronger governance sometimes raises output and lowers fatigue, and moderate governance wins out as the learner accumulates experience.
medium quasi_experimental arxiv
HAAS: A Policy-Aware Framework for Adaptive Task Allocation Between Humans and Artificial Intelligence Systems · Vicente Pelechanoa, Antoni Mestre, Manoli Albert, Miriam Gil · May 04, 2026
Letting users mark constraints as hard rules or soft preferences — and verifying each with a matching technique — makes LLM-generated plans more reliable and usable; in lab studies U-Define improved task success, perceived usefulness, and satisfaction compared with prior approaches.
medium quasi_experimental arxiv
U-Define: Designing User Workflows for Hard and Soft Constraints in LLM-Based Planning · Christine P Lee, Xinyu Jessica Wang, Aws Albarghouthi, David Porfirio, Bilge Mut... · May 04, 2026
More-capable language models produce bloated, tightly coupled code that accrues architectural debt, and code volume reliably predicts this structural decay; functional correctness and detailed prompting do not stop the rot, implying AI-assisted software engineering must add architectural foresight to avoid mounting maintenance costs.
medium descriptive arxiv
AI-Generated Smells: An Analysis of Code and Architecture in LLM and Agent-Driven Development · Yuecai Zhu, Nikolaos Tsantalis, Peter C. Rigby · May 04, 2026
An AI tutor that anticipates teamwork breakdowns improves pair-programming outcomes: in a 26-dyad lab study, forecast-driven prompts raised debugging success and cut task time while enhancing joint attention and mental effort.
medium quasi_experimental arxiv
ProPACT: A Proactive AI-Driven Adaptive Collaborative Tutor for Pair Programming · Anahita Golrang, Kshitij Sharma, olga viberg · May 04, 2026
New causal decision frameworks let economists model bounded rationality and variable discounting inside composable agent models; SCDPs are provably more expressive than POMDPs and suit policy and mechanism simulations for digital infrastructures.
n/a theoretical arxiv
The Design and Composition of Structural Causal Decision Processes · Sebastian Benthall, Alan Lujan · May 04, 2026
A new RL-based index finds reinforcement-learning systems most likely to learn operational and control tasks — lifting feasibility for roles like power plant operators, railroad conductors and cargo-handling supervisors — while creative and interpersonal professions score high on existing exposure indices but low on RL learnability, suggesting different policy priorities than current AI-exposure metrics indicate.
medium descriptive arxiv
What Jobs Can AI Learn? Measuring Exposure by Reinforcement Learning · Philip Moreira Tomei, Bouke Klein Teeselink · May 04, 2026
Cheap AI judgment upends science's monopoly on authoritative judgment; the key bottleneck is institutions' capacity to integrate and certify AI outputs—not computational speed.
n/a theoretical arxiv
AI-Augmented Science and the New Institutional Scarcities · Lauri Lovén · May 04, 2026
AI's riches are accruing to the rich: North America, Europe and parts of Asia–Pacific lead in connectivity, education and AI deployment while large swaths of Africa and South Asia lack the infrastructure and skills to share the benefits; without concerted investment in inclusion and governance, the global AI divide—and its economic and social costs—will deepen.
low descriptive semantic_scholar
GLOBAL DISPROPORTIONS IN THE IMPLEMENTATION AND USE OF ARTIFICIAL INTELLIGENCE · O. Horobets · Fetched May 04, 2026
Next‑generation software‑engineering agents need triadic training data — synchronized human–human context, human–AI interactions, and long‑horizon multi‑engineer work — not bigger GitHub scrapes or solo-agent traces; the authors recommend expert stimulated‑recall recordings and simulated cross‑functional teams plus mechanical verification, corpus characterization, probe experiments, and pre‑registered blind evaluation as minimum quality standards.
n/a commentary arxiv
The Conversations Beneath the Code: Triadic Data for Long-Horizon Software Engineering Agents · Yelin Kim · May 04, 2026