{ "programme": "ZEISMO AI Impact Observatory", "version": "0.3", "updated": "2026-10-09", "status": "Research agenda with two published computational pilots", "studies": [ { "id": "ZRL-AIO-B01", "stream": "Business", "title": "Does AI assistance improve completed work?", "status": "Proposed study", "question": "Compare human-only, AI-assisted and bounded-agent execution on the same consented task set.", "primary_outcome": "Correctly completed tasks per paid hour; total cost including correction and review.", "design": "Randomised task assignment, blinded assessment, power analysis after a pilot, and a fixed test set.", "data_requirements": "Consented business workflow records; human baselines and task-level quality ratings.", "limitations": "Lab tasks may not represent a whole organisation; report worker experience and job effects separately.", "stage": "protocol_design", "results_available": false }, { "id": "ZRL-AIO-B02", "stream": "Business", "title": "Where does business AI adoption differ?", "status": "Proposed study", "question": "Describe adoption by industry and business size using published statistical releases.", "primary_outcome": "Weighted adoption estimates with uncertainty, release dates and questionnaire definitions.", "design": "Repeatable descriptive analysis of comparable survey waves; flag breaks in survey wording.", "data_requirements": "US Census BTOS tables and metadata; additional countries only after comparable sources are verified.", "limitations": "Adoption is not productivity; US survey coverage is not a global estimate.", "stage": "protocol_design", "results_available": false }, { "id": "ZRL-AIO-B03", "stream": "Business", "title": "When Should an AI Employee Decide, Ask for Help, or Stop?", "status": "Computational pilot published", "question": "Measure decisions to act, ask for authorised help or stop under explicit business policies.", "primary_outcome": "Unsafe DECIDE recommendations, accuracy, coverage, missingness and cost on synthetic cases.", "design": "36 synthetic evaluation cases, three repeats, rules and B/C/D model configurations; case-cluster analysis.", "data_requirements": "Published synthetic inputs, fixed reference rubric and preserved model outputs.", "limitations": "Same-model annotation audits; original blinded human explanation review remains unperformed. No production task execution.", "stage": "computational_pilot", "results_available": true, "result_url": "/research/decide-ask-stop" }, { "id": "ZRL-AIO-B04", "stream": "Business", "title": "After the Restart: Can an AI Employee Remember, Recover and Resume Work Safely?", "status": "Computational pilot published", "question": "Can an AI employee resume a synthetic interrupted workflow without stale authority or duplicate action?", "primary_outcome": "Strict correct decision state; separately declared fence-only diagnostic and unsafe/duplicate recommendations.", "design": "48 evaluation cases,10 matched families,6 domains,4 memory conditions and3 repeats; same current evidence and one model call per trial.", "data_requirements": "Published synthetic fixtures, separate rule labels, prompts, recorded outputs, analysis and hash verification.", "limitations": "Short explicit-state fixtures, one model, same development/evaluation families; formatting failures retained. No real recovery or business actions; AI-assisted explanation inspection.", "stage": "computational_pilot", "results_available": true, "result_url": "/research/after-the-restart" }, { "id": "ZRL-AIO-M01", "stream": "Markets", "title": "AI language in corporate filings: commitment or rhetoric?", "status": "Proposed study", "question": "Examine whether companies describe measurable AI investment or only broad ambitions.", "primary_outcome": "Disclosure categories, evidence coverage and agreement between independent coders.", "design": "Freeze a filing cohort and coding manual; blind dual annotation and adjudication.", "data_requirements": "SEC EDGAR filings with acceptance timestamps and linked passages.", "limitations": "Company disclosures are statements, not independently verified deployment or outcomes.", "stage": "protocol_design", "results_available": false }, { "id": "ZRL-AIO-M02", "stream": "Markets", "title": "How do markets react around AI announcements?", "status": "Proposed study", "question": "Estimate associations between defined AI events and market-adjusted returns.", "primary_outcome": "Abnormal returns with intervals; volume and volatility around preregistered windows.", "design": "Event study with pre-event estimation, overlapping-event handling, multiple-testing controls and out-of-sample checks.", "data_requirements": "Timestamped primary announcements and licensed adjusted price histories.", "limitations": "Confounding announcements and selection bias limit causal claims; no trading signal or return promise.", "stage": "protocol_design", "results_available": false }, { "id": "ZRL-AIO-M03", "stream": "Markets", "title": "Do AI research assistants preserve financial evidence?", "status": "Proposed study", "question": "Evaluate factual reliability when models summarise time-stamped public filings.", "primary_outcome": "Citation correctness, numerical accuracy, unsupported claims and abstention.", "design": "A frozen point-in-time corpus; compare retrieval, no retrieval and human-checked reference answers.", "data_requirements": "Permitted filings with source locations and a documented question set.", "limitations": "A later revised filing must not leak into an earlier decision; accuracy does not establish investment skill.", "stage": "protocol_design", "results_available": false }, { "id": "ZRL-AIO-M04", "stream": "Markets", "title": "Can public evidence measure AI infrastructure concentration?", "status": "Proposed study", "question": "Map disclosed dependence on compute suppliers and infrastructure spending.", "primary_outcome": "Source-supported exposure measures, coverage gaps and uncertainty ranges.", "design": "Entity-resolution review and transparent definitions; sensitivity to missing disclosures.", "data_requirements": "Audited company statements and official industry statistics with reuse review.", "limitations": "Missing disclosures are not zero exposure; scenarios must not be presented as forecasts.", "stage": "protocol_design", "results_available": false }, { "id": "ZRL-AIO-P01", "stream": "Public policy", "title": "How do AI governance requirements differ?", "status": "Proposed study", "question": "Create a dated, jurisdiction-specific comparison of published public-sector AI rules.", "primary_outcome": "Coverage of oversight, transparency, appeal and evaluation requirements.", "design": "Code primary legal and policy texts with dual review, effective dates and an amendment log.", "data_requirements": "Official government publications and policy repositories.", "limitations": "Comparative research, not legal advice; proposals and enacted rules remain separate.", "stage": "protocol_design", "results_available": false }, { "id": "ZRL-AIO-P02", "stream": "Public policy", "title": "What does AI incident evidence actually show?", "status": "Proposed study", "question": "Study incident types and the limits of reporting-based datasets.", "primary_outcome": "Incident categories, source confirmation and missing-data patterns.", "design": "Deduplicate reports, separate allegations from findings, and audit coding disagreement.", "data_requirements": "OECD AI incident methodology and reusable primary records, subject to source terms.", "limitations": "Reported incidents cannot measure population risk without a valid exposure denominator.", "stage": "protocol_design", "results_available": false }, { "id": "ZRL-AIO-P03", "stream": "Public policy", "title": "Can AI improve public-service response quality?", "status": "Proposed study", "question": "Compare assistance on non-sensitive public-information tasks using consistent rubrics.", "primary_outcome": "Factual accuracy, accessibility, unequal error rates and successful escalation.", "design": "Consented evaluation or synthetic simulation; preregistered subgroups and minimum sample requirements.", "data_requirements": "Public service documents and synthetic requests; no private case files in a public dataset.", "limitations": "Avoid political targeting and automated eligibility decisions; measure harms as well as speed.", "stage": "protocol_design", "results_available": false }, { "id": "ZRL-AIO-P04", "stream": "Public policy", "title": "Does source-linked AI explanation support public understanding?", "status": "Proposed study", "question": "Test whether visible evidence and uncertainty improve understanding of policy summaries.", "primary_outcome": "Comprehension, source recall, correction uptake and unjustified confidence.", "design": "Voluntary participant study with consent, ethics review, randomised presentation and a published questionnaire.", "data_requirements": "Balanced public policy documents and consenting adult participants.", "limitations": "No participant collection until ethics and privacy review; no profiling voters or political persuasion.", "stage": "protocol_design", "results_available": false } ], "candidate_sources": [ { "name": "US Census Business Trends and Outlook Survey", "url": "https://www.census.gov/hfp/btos/data", "use": "Candidate source for descriptive business-adoption studies; preserve questionnaire changes and uncertainty." }, { "name": "SEC developer resources", "url": "https://www.sec.gov/about/developer-resources", "use": "Candidate source for timestamped company filings; observe fair-access requirements and confirm reuse terms." }, { "name": "OECD AI incident methodology", "url": "https://oecd.ai/en/incidents-methodology", "use": "Methodological reference for incident classification and reporting-bias controls; source terms apply." } ], "results": "No results are reported for these proposed studies. The separately published JARVIS preprint contains its own evidence and limitations." }