Social Science Research Assistants

19-4061.00

Median wage $61,990/yr30,640 employed (US)Rank #72 of 923 scored · top 8% by substitution

Assist social scientists in laboratory, survey, and other social science research. May help prepare findings for publication and assist in laboratory analysis, quality control, or data management.

Sub-scores

0–100 · band = confidence interval from rater disagreement

Substitution53

Exposure49

Augmentation75

Substitution — the headline: capability discounted by cost, barriers and adoption.

Exposure — technical capability alone, regardless of whether anyone deploys it.

Augmentation — how much AI assists without replacing. High here + moderate substitution = a changing job, not a disappearing one.

Tasks on the substitution scale

22 rated tasks, binned by substitution score.

Position among all scored occupations

Distribution of 923 occupation scores; the marker is this occupation.

Tasks with substitution ≥ 70

41%

Run 1.0.0-draft.1 · computed 2026-08-05 · rater panel: claude-sonnet-5, claude-haiku-4-5-20251001 · intervals span rater disagreement.

Why this score

The five weighted dimensions of the composite, averaged across this occupation's tasks (importance-weighted, panel mean). Exact weights and formulas: /api/v1/methodology.

Task automatabilityw 35%50

panel mean rating 3.0/5 → substitution pressure 50/100

Technical feasibility todayw 20%49

panel mean rating 3.0/5 → substitution pressure 49/100

Cost vs. human wagew 15%60

panel mean rating 3.4/5 → substitution pressure 60/100

Adoption barriersw 20%inverted — strong barriers lower the score60

panel mean rating 2.6/5 (barrier strength) → substitution pressure 60/100

Sector adoption velocityw 10%46

panel mean rating 2.8/5 → substitution pressure 46/100

Task breakdown (22 tasks)

Substitution pressure per task, weighted by O*NET importance in the composite. Expand a task for the full rater audit trail — every rating, every model, every rationale.

Perform data entry and other clerical work as required for project completion.

CI 79–92 · exposure 83 · augmentation 63 · importance 3.9/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	4/5	Research institutions, universities, and administrative organizations are actively adopting RPA and automated data processing systems. Clerical automation is well-established in information-intensive sectors, with widespread production deployment and measurable displacement of routine data-entry roles.
Sector adoption velocity	claude-sonnet-5	4/5	Research and administrative offices in academia and social science settings have broadly adopted spreadsheet automation, forms software, and RPA tools for clerical tasks, though full displacement lags top digital sectors.
Augmentation potential	claude-haiku-4-5-20251001	3/5	AI can assist research assistants by flagging data quality issues, auto-populating forms, and accelerating data validation workflows. However, the task itself is so automatable that augmentation is less salient than replacement—the human's primary value here shifts to exception-handling and oversight rather than enhanced performance on core entry work.
Augmentation potential	claude-sonnet-5	4/5	AI tools substantially speed up data entry, cleaning, and formatting tasks for research assistants who remain responsible for verifying accuracy and handling exceptions.
Task automatability	claude-haiku-4-5-20251001	4/5	Data entry and clerical work are highly structured, rule-based tasks that AI systems can automate at scale. Current OCR, RPA, and form-filling tools can handle large volumes of routine data entry with minimal setup, easily meeting the 50% time-saving threshold for standard clerical operations.
Task automatability	claude-sonnet-5	5/5	Data entry and clerical work is highly structured, repetitive, and text/number based, making it well suited to OCR, form-parsing, and automated data pipelines that meet or exceed the 50% time-saving threshold.
Adoption barriers	claude-haiku-4-5-20251001	2/5	Data entry and clerical work face minimal legal or regulatory barriers to automation. Primary friction comes from organizational inertia, need for oversight to catch errors, and institutional preference for human verification rather than hard requirements that prevent substitution.
Adoption barriers	claude-sonnet-5	1/5	No licensing, liability, or human-contact requirements attach to routine clerical data entry; organizations freely automate it.
Cost vs. human wage	claude-haiku-4-5-20251001	5/5	The cost of cloud-based RPA, OCR, and data entry automation is typically a small fraction of the fully-loaded wage of a research assistant performing manual data entry, often yielding 10x or greater cost savings at scale.
Cost vs. human wage	claude-sonnet-5	5/5	Automated data entry tools cost a small fraction of a per-hour clerical wage once set up, especially at any scale of repeated entry.
Technical feasibility today	claude-haiku-4-5-20251001	4/5	Deployed RPA platforms, document processing services, and form-automation tools reliably perform data entry and clerical work in production environments across research institutions and administrative settings. Error rates on well-structured data are low, though quality assurance oversight remains necessary.
Technical feasibility today	claude-sonnet-5	4/5	Mature products (RPA tools, OCR/document AI, spreadsheet automation, LLM-based data extraction) are deployed widely in production for clerical data entry, though edge cases and messy source data still require human correction.

Conduct internet-based and library research.

CI 77–84 · exposure 75 · augmentation 100 · importance 3.8/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	4/5	Academic institutions, think tanks, and research organizations are rapidly adopting AI-assisted research tools; pilot projects are widespread and production use is growing quickly in information-intensive sectors.
Sector adoption velocity	claude-sonnet-5	3/5	Academic and research support settings show growing but uneven AI adoption—some labs and institutions use AI search tools heavily while others remain cautious about citation reliability and academic integrity norms.
Augmentation potential	claude-haiku-4-5-20251001	5/5	AI research assistants dramatically augment human researchers by instantly retrieving and synthesizing sources, freeing researchers to focus on critical analysis, interpretation, and hypothesis development while remaining in the loop.
Augmentation potential	claude-sonnet-5	5/5	AI dramatically accelerates literature discovery, summarization, and initial synthesis, letting research assistants cover far more ground while still verifying and contextualizing findings.
Task automatability	claude-haiku-4-5-20251001	4/5	Current AI systems can autonomously conduct vast amounts of internet-based research, summarize findings, and retrieve library resources (via APIs and databases) with substantial time savings. However, the task requires judgment about source credibility and synthesis that may still benefit from human oversight, preventing a perfect 5.
Task automatability	claude-sonnet-5	4/5	AI systems can search, retrieve, summarize, and synthesize information from web and academic databases with substantial time savings, though verification of accuracy and source quality still benefits from human oversight.
Adoption barriers	claude-haiku-4-5-20251001	1/5	No legal, licensing, or authorization requirements prevent AI from conducting research; organizations face minimal friction adopting automated research tools beyond user preference for human verification.
Adoption barriers	claude-sonnet-5	1/5	No licensing, legal, or regulatory requirement mandates a human perform internet or library research; it's a low-stakes information-gathering task with minimal institutional friction.
Cost vs. human wage	claude-haiku-4-5-20251001	5/5	AI-based research (inference cost + API calls) is orders of magnitude cheaper than hiring a research assistant to manually search databases, visit libraries, and compile findings over hours.
Cost vs. human wage	claude-sonnet-5	4/5	AI research tools cost a small fraction (subscription/API costs) compared to an assistant's hourly wage for equivalent search and summarization volume, though human review time adds some cost back.
Technical feasibility today	claude-haiku-4-5-20251001	4/5	Deployed products (search APIs, AI research assistants, library discovery tools, LLMs with web access) reliably perform internet research at scale. Some limitations remain in accessing paywalled academic content and verifying sources, but core research retrieval is production-grade.
Technical feasibility today	claude-sonnet-5	4/5	Deployed products (AI search assistants, research tools with citation retrieval like Perplexity, Elicit, and enterprise AI search) reliably perform literature and web research today, though occasional hallucinated citations or missed nuance persist.

Perform descriptive and multivariate statistical analyses of data, using computer software.

CI 75–84 · exposure 80 · augmentation 100 · importance 4.1/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	4/5	Academic research, tech, finance, and professional services sectors are rapidly adopting AI-assisted statistical workflows; production displacement of research assistants on this task is measurable and accelerating in information-intensive organizations.
Sector adoption velocity	claude-sonnet-5	4/5	Academic and social science research settings are increasingly adopting AI coding assistants and data analysis tools, following broader fast adoption trends in research and professional services.
Augmentation potential	claude-haiku-4-5-20251001	5/5	AI dramatically augments research assistants by automating code generation, exploratory analysis, and result summarization, allowing humans to focus on hypothesis design and interpretation while staying in the loop.
Augmentation potential	claude-sonnet-5	5/5	AI dramatically accelerates writing analysis code, generating visualizations, and suggesting appropriate statistical tests, while the researcher retains judgment over interpretation and methodological validity.
Task automatability	claude-haiku-4-5-20251001	4/5	Current AI can reliably execute descriptive and multivariate statistical analyses end-to-end using standard software (Python, R, SQL) with minimal human guidance, easily meeting the 50% time-saving threshold. However, full autonomy requires human judgment on model selection, interpretation, and validation, preventing a perfect 5.
Task automatability	claude-sonnet-5	4/5	AI code-generation tools can write and execute statistical scripts (R, Python, SPSS syntax) for descriptive stats and multivariate models like regression or factor analysis with substantial time savings, though complex model selection and interpretation still need human review.
Adoption barriers	claude-haiku-4-5-20251001	2/5	Few legal or licensing barriers exist for automated statistical analysis itself; oversight requirements are organizational rather than regulatory, and many sectors eagerly adopt AI for this routine computational task without human sign-off requirements.
Adoption barriers	claude-sonnet-5	2/5	No licensure is required to run statistical analyses; the main friction is quality control and ensuring correct methodology, which organizations can manage internally without legal barriers.
Cost vs. human wage	claude-haiku-4-5-20251001	5/5	AI inference cost per analysis is negligible (cents to dollars), while a research assistant's loaded wage for the same output is typically $25–50+ per hour; this easily achieves an order-of-magnitude cost advantage.
Cost vs. human wage	claude-sonnet-5	4/5	Running statistical analyses via AI-assisted coding costs a fraction of an hourly wage for a research assistant, though some oversight and data cleaning cost remains.
Technical feasibility today	claude-haiku-4-5-20251001	5/5	Mature products (ChatGPT with code interpreter, Claude, specialized stats packages with AI integration) demonstrably perform statistical analysis in production; academic researchers and analysts routinely deploy AI for this work at scale with reliable outputs.
Technical feasibility today	claude-sonnet-5	4/5	Tools like ChatGPT with code interpreter, Copilot, and specialized data-analysis agents are deployed and reliably handle standard statistical workflows, though edge cases (messy data, unusual designs) still require analyst intervention.

Design and create special programs for tasks such as statistical analysis and data entry and cleaning.

CI 75–79 · exposure 75 · augmentation 100 · importance 4.5/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	4/5	Academic and research institutions, as well as data-heavy sectors (tech, finance, consulting), are rapidly adopting AI coding and data-cleaning tools; GitHub Copilot, ChatGPT, and cloud-native statistical platforms show strong uptake in research labs and corporate data teams, indicating faster-than-average adoption.
Sector adoption velocity	claude-sonnet-5	4/5	Academic and research settings have rapidly adopted AI coding assistants for data wrangling and statistical scripting, though full pipeline automation in production is less common than in tech/finance sectors.
Augmentation potential	claude-haiku-4-5-20251001	5/5	AI systems powerfully augment human researchers by drafting code templates, suggesting statistical approaches, flagging data quality issues, and accelerating iteration; researchers remain in the loop for validation and interpretation, and productivity gains are substantial and well-documented.
Augmentation potential	claude-sonnet-5	5/5	AI dramatically accelerates writing, debugging, and documenting analysis code and cleaning scripts while the researcher retains oversight over methodology and validity.
Task automatability	claude-haiku-4-5-20251001	4/5	Current AI systems can substantially automate statistical analysis setup, data cleaning pipelines, and data entry workflows; tools like ChatGPT, Claude, and specialized ML platforms can generate reproducible code, handle missing data, and flag anomalies at significant time savings. However, full end-to-end automation requires domain expertise and custom validation that typically needs human oversight, placing this near but not quite at the 5-level threshold.
Task automatability	claude-sonnet-5	4/5	Writing statistical analysis scripts and data cleaning code (e.g., in Python/R) is a well-bounded, text-based task that current LLM coding tools handle well, though custom edge cases still need human review.
Adoption barriers	claude-haiku-4-5-20251001	2/5	Few legal or regulatory barriers exist for automating statistical code generation or data cleaning; research institutions have adopted AI tools widely and publication standards do not mandate human design of these programs. Some IRB or data governance friction may arise with sensitive data, but these are organizational rather than legal barriers.
Adoption barriers	claude-sonnet-5	2/5	No licensing requirement exists for writing analysis scripts, though some organizational oversight and validation of statistical correctness is expected before use in research outputs.
Cost vs. human wage	claude-haiku-4-5-20251001	5/5	API costs for LLM-driven code generation and data cleaning are typically $0.01–$0.10 per task compared to a research assistant's loaded wage ($25–$35/hour for that work), yielding cost-per-output ratios at least 10× cheaper once amortized across many tasks.
Cost vs. human wage	claude-sonnet-5	4/5	AI coding assistance costs a small subscription fee versus the hourly cost of a research assistant, offering substantial savings for routine script writing and data cleaning.
Technical feasibility today	claude-haiku-4-5-20251001	4/5	Deployed products (e.g., OpenAI Codex, GitHub Copilot, SAS Viya, Alteryx) reliably generate statistical code, automate data cleaning rules, and assist with data entry validation in production environments. Error rates remain material on novel or ambiguous datasets, and integration with research-specific workflows varies, so this falls short of fully mature, error-free production deployment across all contexts.
Technical feasibility today	claude-sonnet-5	4/5	Production tools like GitHub Copilot, ChatGPT Code Interpreter, and specialized data-cleaning assistants are widely used today to generate and debug statistical/data pipeline code reliably for common cases.

Prepare tables, graphs, fact sheets, and written reports summarizing research results.

CI 75–79 · exposure 75 · augmentation 100 · importance 4.2/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	4/5	Research institutions, analytics teams, and knowledge work sectors are adopting AI-assisted report generation rapidly. Many research organizations now use LLMs and BI tools to draft summaries and visualizations, demonstrating measurable adoption momentum in information-intensive sectors.
Sector adoption velocity	claude-sonnet-5	4/5	Research and academic/policy analytics sectors have rapidly adopted AI-assisted writing and visualization tools, aligning with fast-adopting information/professional-services patterns.
Augmentation potential	claude-haiku-4-5-20251001	5/5	AI dramatically enhances productivity for research assistants by auto-generating draft tables, visualizations, and summaries that the human can review, edit, and contextualize. This allows assistants to focus on interpretation and quality assurance rather than mechanical formatting, significantly multiplying output per person.
Augmentation potential	claude-sonnet-5	5/5	AI substantially accelerates drafting of reports, generating charts, and summarizing findings, letting researchers focus on interpretation and quality control.
Task automatability	claude-haiku-4-5-20251001	4/5	AI systems can reliably generate tables, graphs, and basic written summaries from structured data with minimal human intervention. However, the task requires domain interpretation and contextual judgment to ensure accuracy and appropriateness of visualizations, which introduces some friction. Current tools can achieve ~60–75% time savings on routine summary generation.
Task automatability	claude-sonnet-5	4/5	Modern AI tools can analyze data, generate charts, and draft narrative summaries from structured inputs with substantial time savings, though final review and framing for accuracy still require human oversight.
Adoption barriers	claude-haiku-4-5-20251001	2/5	There are minimal regulatory or legal barriers to automating report generation and summarization. The main friction is organizational preference for human review and sign-off, and potential quality-control practices, but no mandatory human-performed task requirement exists.
Adoption barriers	claude-sonnet-5	2/5	No licensing or legal requirement mandates human authorship of internal research summaries, though some organizations require human review for accuracy and interpretation before dissemination.
Cost vs. human wage	claude-haiku-4-5-20251001	5/5	API costs for inference and integration are orders of magnitude lower than a loaded human salary for equivalent output. A single API call generating a report costs pennies compared to hours of human labor.
Cost vs. human wage	claude-sonnet-5	4/5	AI-assisted report generation and visualization is dramatically cheaper per unit output than a research assistant's hourly wage, though data preparation and validation overhead reduce the full order-of-magnitude gain.
Technical feasibility today	claude-haiku-4-5-20251001	4/5	Mature products (e.g., ChatGPT, Claude, business intelligence platforms) demonstrate reliable production performance for generating tables, graphs, and fact sheets from research data. Some edge cases around complex statistical interpretation or highly specialized formats may require oversight, but the core task is consistently handled at scale.
Technical feasibility today	claude-sonnet-5	4/5	Deployed tools (e.g., code interpreters, BI dashboard generators, LLM report writers) already produce tables, graphs, and draft reports reliably in many workplaces, though customization for nuanced research fact sheets still needs human editing.

Verify the accuracy and validity of data entered in databases, correcting any errors.

CI 72–79 · exposure 75 · augmentation 100 · importance 4.0/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	4/5	Research institutions, universities, and data-intensive organizations have rapidly adopted automated data-quality tools and validation pipelines; major platforms (REDCap, Qualtrics, enterprise data warehouses) now include built-in AI-assisted validation by default.
Sector adoption velocity	claude-sonnet-5	3/5	Academic and social science research settings are moderate adopters of AI tools compared to finance or tech, with growing but uneven uptake of automated data QA.
Augmentation potential	claude-haiku-4-5-20251001	5/5	AI-powered flagging systems dramatically assist human validators by highlighting suspicious records, suggesting corrections, and prioritizing review effort, allowing researchers to focus on complex contextual judgment while the system handles routine checks.
Augmentation potential	claude-sonnet-5	5/5	AI significantly assists by flagging inconsistencies, duplicate entries, and outliers, letting the human research assistant focus on judgment calls and edge cases.
Task automatability	claude-haiku-4-5-20251001	4/5	Data validation tasks are highly amenable to automation: rule-based checks, pattern matching, and statistical anomaly detection can identify ~80% of common data-entry errors without human intervention, easily meeting the 50% time-saving threshold for structured database verification.
Task automatability	claude-sonnet-5	4/5	AI tools can cross-check entries, flag anomalies, and apply validation rules across large datasets much faster than manual review, meeting the time-saving threshold for most routine validation work.'
Adoption barriers	claude-haiku-4-5-20251001	2/5	Few regulatory or licensing barriers exist for automated data validation itself; the main friction is organizational (preference for human sign-off on research integrity) and integration overhead, but these are surmountable with standard oversight.
Adoption barriers	claude-sonnet-5	2/5	No licensing requirement exists for this task, but institutional data governance policies and the need for accountability on research integrity create moderate friction.
Cost vs. human wage	claude-haiku-4-5-20251001	5/5	Automated data validation (via scripts, ML models, or low-cost APIs) costs pennies per record, while human verification requires $15–25/hour labor; AI is multiple orders of magnitude cheaper at scale.
Cost vs. human wage	claude-sonnet-5	4/5	Automated validation scripts and AI-assisted checks cost a fraction of a research assistant's hourly wage once set up, though initial configuration and oversight add some cost.
Technical feasibility today	claude-haiku-4-5-20251001	4/5	Mature data validation tools, including AI-powered anomaly detection and rule-based systems, are already deployed in production across research institutions and enterprises; however, complex contextual validity checks still require human review, preventing a perfect-5 rating.
Technical feasibility today	claude-sonnet-5	4/5	Data validation software, ETL pipelines, and LLM-based anomaly detection are widely deployed in research and business settings today, though edge cases and domain-specific judgment still require human review.

Track laboratory supplies and expenses such as participant reimbursement.

CI 72–79 · exposure 75 · augmentation 75 · importance 3.8/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	4/5	Research institutions and academic laboratories have rapidly adopted digital inventory and accounting tools over the past decade. Cloud-based expense management and supply-chain automation are now standard in most higher-education and professional research settings.
Sector adoption velocity	claude-sonnet-5	3/5	Academic and research labs adopt administrative software at a moderate pace; larger institutions have adopted automated expense systems while many small labs still use manual spreadsheets.
Augmentation potential	claude-haiku-4-5-20251001	4/5	AI systems significantly enhance research assistants' productivity by automating data entry, flagging discrepancies, and generating expense reports, freeing them to focus on verifying accuracy, handling exceptions, and supporting principal investigators with higher-level analytical tasks.
Augmentation potential	claude-sonnet-5	4/5	AI tools can significantly speed up receipt scanning, categorization, and reconciliation, letting the assistant focus on exceptions and approvals.
Task automatability	claude-haiku-4-5-20251001	4/5	This task involves routine inventory and expense tracking with clearly structured data entry and reconciliation. Current AI systems and accounting software can automate the bulk of supply tracking, expense categorization, and participant reimbursement processing, achieving >50% time savings, though some human oversight for exception handling and approval workflows remains necessary.
Task automatability	claude-sonnet-5	4/5	Tracking supplies and expenses is largely a structured data-entry and reconciliation task that off-the-shelf expense/inventory software and AI-assisted tools can handle with high time savings.To fully close it out still requires occasional human verification for edge cases.
Adoption barriers	claude-haiku-4-5-20251001	2/5	Few regulatory or hard legal barriers exist for automating supply and expense tracking. The main friction comes from organizational preferences to maintain human control over approvals and occasional policy verification, but these are soft rather than binding constraints.
Adoption barriers	claude-sonnet-5	2/5	No licensing requirement, though institutional financial policies and audit trails may require some human sign-off on reimbursements.
Cost vs. human wage	claude-haiku-4-5-20251001	5/5	AI-driven automation of inventory management and expense processing via off-the-shelf software costs a fraction of a full-time research assistant's salary; even with integration and oversight, the cost per transaction is orders of magnitude lower than human labor.
Cost vs. human wage	claude-sonnet-5	4/5	Automated expense/inventory tracking software costs a small fraction of a research assistant's time-equivalent wage for this narrow bookkeeping function.
Technical feasibility today	claude-haiku-4-5-20251001	4/5	Deployed accounting and laboratory management software already perform expense tracking and supply inventory reliably in production settings. Integration with standard tools like Quickbooks, Airtable, and SAP is mature, though some organizations may need minor customization for research-specific workflows.
Technical feasibility today	claude-sonnet-5	4/5	Mature expense-management and inventory-tracking products (e.g., automated receipt capture, spreadsheet automation, procurement software) are widely deployed in labs and institutions today.

Code data in preparation for computer entry.

CI 67–81 · exposure 70 · augmentation 88 · importance 3.9/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	3/5	Academic and research institutions show moderate adoption of automated coding tools, with many pilots underway but slower institutional rollout than in commercial analytics. The information sector and tech-enabled research firms are adopting faster than traditional academia.
Sector adoption velocity	claude-sonnet-5	3/5	Social science and academic research settings are moderate adopters of AI tools—some labs use AI-assisted coding pipelines, but many still rely on manual or semi-manual processes due to funding, training, and methodological conservatism.
Augmentation potential	claude-haiku-4-5-20251001	4/5	AI augmentation is particularly strong here: research assistants can use AI to pre-code large batches, then review and refine outputs, dramatically accelerating the coding workflow while maintaining human judgment over edge cases and scheme validation.
Augmentation potential	claude-sonnet-5	5/5	AI tools substantially speed up initial coding passes, suggest categories, and flag inconsistencies, letting researchers focus on validation and interpretation while remaining in the loop.
Task automatability	claude-haiku-4-5-20251001	4/5	Coding data for computer entry is highly structured work that involves applying predefined categorical rules to raw information. Current AI systems (particularly LLMs and classification models) can handle categorization and data labeling at scale with minimal human oversight, potentially achieving >50% time savings with equal or better quality than manual coding.
Task automatability	claude-sonnet-5	4/5	Coding/categorizing textual or survey data into standardized schemas is a well-structured pattern-recognition task that LLMs handle well, especially with defined codebooks, saving substantial time versus manual coding.4
Adoption barriers	claude-haiku-4-5-20251001	2/5	There are minimal regulatory or legal barriers to automating data coding. The main friction points are organizational preference for human oversight, QA workflows, and ensuring coding scheme documentation—all relatively easy to address within research teams.
Adoption barriers	claude-sonnet-5	1/5	There is no licensing or legal requirement that a human perform data coding; it's a routine clerical/analytical task with no regulatory protection.
Cost vs. human wage	claude-haiku-4-5-20251001	5/5	AI-based data coding costs per unit (inference + platform integration) are typically one to two orders of magnitude lower than research assistant labor ($25–40/hour loaded wage), especially for high-volume datasets where fixed setup costs amortize across thousands of records.
Cost vs. human wage	claude-sonnet-5	4/5	Automated coding via LLM APIs is dramatically cheaper per data point than paying a research assistant hourly, even after factoring in setup and spot-checking costs.
Technical feasibility today	claude-haiku-4-5-20251001	4/5	Multiple deployed products (data labeling platforms, classification APIs, and LLM-based coding tools) reliably perform this task in production for market research, survey analysis, and social science datasets. While accuracy depends on clear coding schemes and training data quality, mature solutions exist with documented performance at scale.
Technical feasibility today	claude-sonnet-5	3/5	Products exist (LLM-based classifiers, qualitative coding tools like Atlas.ti AI, survey platforms) that perform this in production, but accuracy on nuanced or ambiguous categories still requires human review, limiting full reliability.

Prepare, manipulate, and manage extensive databases.

CI 67–75 · exposure 70 · augmentation 88 · importance 3.9/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	4/5	Information and research sectors have rapidly adopted cloud data platforms, automated ETL pipelines, and AI-assisted data preparation tools. Universities and research institutions increasingly deploy these systems; adoption is mainstream in digital-first organizations.
Sector adoption velocity	claude-sonnet-5	3/5	Academic and research settings adopt data tools unevenly; some labs use AI-assisted coding heavily while others lag due to funding, training, and legacy workflows.
Augmentation potential	claude-haiku-4-5-20251001	4/5	AI excels at suggesting schema optimizations, flagging data quality issues, and automating routine transformations while humans validate assumptions and govern policy compliance. This pairing materially amplifies a research assistant's throughput and reduces tedious manual work.
Augmentation potential	claude-sonnet-5	5/5	AI coding assistants dramatically speed up query writing, cleaning scripts, and database restructuring while the assistant retains oversight of accuracy and research validity.
Task automatability	claude-haiku-4-5-20251001	4/5	Database preparation, manipulation, and management is highly structured and rule-based. Current AI systems with SQL generation, ETL tools, and data transformation agents can automate the majority of routine data cleaning, loading, and management tasks, achieving well over 50% time savings on standardized operations, though complex schema design may require human oversight.
Task automatability	claude-sonnet-5	4/5	Data cleaning, transformation, merging, and management can largely be handled by AI-assisted scripting (e.g., pandas/R code generation) and agentic data pipelines, saving substantial time over manual work, though complex domain-specific validation still needs human oversight.
Adoption barriers	claude-haiku-4-5-20251001	2/5	Few legal or regulatory barriers prevent automated database management; institutional friction (legacy systems, training needs) and data governance/compliance checks exist but do not legally require human oversight of the automation itself.
Adoption barriers	claude-sonnet-5	2/5	No licensing requirement for database management, but data privacy/IRB compliance and need for accurate, defensible research data create some oversight friction.
Cost vs. human wage	claude-haiku-4-5-20251001	4/5	Cloud database services and AI-assisted data tools have marginal costs far below research assistant wages when amortized across large datasets. A single instance can process thousands of records per dollar, while a human might cost $25–50/hour for the same work.
Cost vs. human wage	claude-sonnet-5	4/5	AI-assisted coding and automated ETL tools are far cheaper per unit of database management work than paying a research assistant, especially for repetitive structuring and cleaning tasks.
Technical feasibility today	claude-haiku-4-5-20251001	4/5	Production-grade tools (Palantir, Alteryx, cloud-native data platforms) and AI-assisted SQL/Python agents reliably handle database preparation and manipulation at scale in real organizations. Error rates on routine operations are low, though quality assurance and validation typically require human review.
Technical feasibility today	claude-sonnet-5	3/5	Tools like GitHub Copilot, ChatGPT code interpreter, and data pipeline automation products are used in production for database wrangling, but reliability drops for messy, domain-specific research datasets requiring nuanced judgment.

Provide assistance with the preparation of project-related reports, manuscripts, and presentations.

CI 64–75 · exposure 67 · augmentation 100 · importance 4.2/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	4/5	Universities, research institutions, and professional services firms—high-digitization sectors—have rapidly adopted AI writing and presentation tools in research workflows; pilots are common and production use is accelerating, particularly for junior research support roles.
Sector adoption velocity	claude-sonnet-5	3/5	Academic and research institutions are adopting AI writing tools steadily, but usage remains uneven across disciplines and often informal rather than institutionalized.
Augmentation potential	claude-haiku-4-5-20251001	5/5	AI substantially amplifies research assistants' productivity in preparing presentations, structuring manuscripts, literature synthesis, and polishing prose, allowing humans to focus on analytical and interpretive judgment while AI handles formatting, drafting scaffolding, and revision cycles.
Augmentation potential	claude-sonnet-5	5/5	AI substantially speeds up drafting, formatting, and summarizing for reports and presentations while the researcher retains control over content accuracy and interpretation.
Task automatability	claude-haiku-4-5-20251001	3/5	AI can automate significant portions of report and manuscript preparation—literature synthesis, draft structuring, figure/table organization, and editing—but typically requires human researchers to guide framing, validate findings, and ensure accuracy. This likely meets the 50% time-saving threshold for routine components while leaving analysis and synthesis decisions to humans.
Task automatability	claude-sonnet-5	4/5	Drafting reports, manuscripts, and presentations from data/notes is well within current LLM capabilities, especially with structured inputs, though final human review remains needed for accuracy and nuance.9
Adoption barriers	claude-haiku-4-5-20251001	2/5	Minimal hard barriers exist: research assistants are not licensed professionals, publications already accept AI-assisted preparation, and oversight is institutional policy rather than law. Some organizations have guidelines on AI disclosure, but these are governance friction rather than legal prohibition.
Adoption barriers	claude-sonnet-5	2/5	No licensing requirement exists for this task, but academic integrity norms, citation accuracy expectations, and authorship conventions create moderate friction.
Cost vs. human wage	claude-haiku-4-5-20251001	4/5	API costs for AI writing assistance are typically 5–50× cheaper than the fully-loaded hourly rate of a research assistant, amortized over routine preparation tasks like formatting, initial drafting, and copyediting.
Cost vs. human wage	claude-sonnet-5	4/5	AI drafting assistance costs a small fraction of a research assistant's hourly wage, though human editing and fact-checking add some overhead.
Technical feasibility today	claude-haiku-4-5-20251001	4/5	Multiple mature products (Claude, ChatGPT, Grammarly, reference management tools) are deployed in academic and research organizations and demonstrably assist with drafting, editing, and formatting. Error rates on routine preparation tasks are low, though content accuracy remains human-dependent; this is narrower in scope than full manuscript authorship but highly functional.
Technical feasibility today	claude-sonnet-5	4/5	Products like ChatGPT, Claude, and Microsoft Copilot are routinely used in academic and research settings today to draft reports and slide decks, though quality varies with complex domain content.

Edit and submit protocols and other required research documentation.

CI 50–62 · exposure 58 · augmentation 75 · importance 3.9/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	3/5	Universities and research institutions have begun piloting automated compliance tools and document-management systems, but widespread production adoption remains limited. Research is knowledge-work intensive and risk-averse; adoption is faster in some quantitative fields but lagging in others.
Sector adoption velocity	claude-sonnet-5	3/5	Academic and research institutions are adopting AI writing tools steadily but cautiously, with pilots for document drafting more common than fully automated submission workflows.
Augmentation potential	claude-haiku-4-5-20251001	4/5	AI can substantially assist researchers by auto-formatting documents, catching compliance gaps, and suggesting revisions, allowing humans to focus on substantive protocol design. AI as a co-editor meaningfully raises the productivity of protocol preparation while researchers retain final authority and judgment.
Augmentation potential	claude-sonnet-5	4/5	AI substantially speeds up drafting, editing, and formatting of protocols and documentation, letting research assistants focus on review and submission logistics, while a human remains responsible for final accuracy and compliance.
Task automatability	claude-haiku-4-5-20251001	4/5	Current AI systems can handle substantial parts of protocol editing (grammar, formatting, compliance checking) and document submission workflows with minimal human intervention, achieving clear time savings. However, substantive revisions requiring nuanced judgment about research ethics or institutional requirements may still need human review, preventing a full 5.
Task automatability	claude-sonnet-5	3/5	AI can draft and edit protocol text and check formatting/compliance requirements, but final submission involves institutional systems, signatures, and judgment calls that require human oversight, so only partial time savings are achievable end-to-end.
Adoption barriers	claude-haiku-4-5-20251001	3/5	Institutional review boards and funding agencies often require human sign-off and accountability for submission accuracy, creating oversight requirements. However, there is no strict legal prohibition on AI preprocessing or drafting these materials, only organizational friction and prudent verification norms.
Adoption barriers	claude-sonnet-5	3/5	IRB and institutional research protocols often require accountable human sign-off and adherence to regulatory/ethical review standards, creating moderate friction against full automation even though editing itself is unregulated.
Cost vs. human wage	claude-haiku-4-5-20251001	4/5	AI-driven document editing and submission is substantially cheaper than paying a research assistant's fully-loaded wage for the same output, especially at scale. Integration and oversight costs are modest relative to labor savings, though not quite an order of magnitude difference.
Cost vs. human wage	claude-sonnet-5	3/5	AI editing assistance is cheap per use, but human review, compliance verification, and submission through institutional portals still require paid staff time, keeping overall costs roughly comparable to a human-only process when done properly.
Technical feasibility today	claude-haiku-4-5-20251001	3/5	Tools exist for document editing and formatting (LLMs, specialized compliance software), and some institutions use automated submission systems, but reliable end-to-end performance across diverse protocol requirements and institutional variations remains inconsistent. Products work in narrower contexts but lack the maturity for universal deployment.
Technical feasibility today	claude-sonnet-5	3/5	Products like ChatGPT, Grammarly, and specialized research-writing tools reliably assist with editing documentation, but no deployed product autonomously handles full protocol submission workflows including IRB systems and compliance checks.

Track research participants, and perform any necessary follow-up tasks.

CI 29–56 · exposure 38 · augmentation 63 · importance 4.1/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	2/5	Academic and research institutions remain conservative in automating participant-facing tasks due to liability concerns, ethical review processes, and lower digitization than commercial sectors, resulting in slow adoption.
Sector adoption velocity	claude-sonnet-5	2/5	Academic and research support settings are typically slow adopters of AI tools for administrative participant management, with most tracking still done via legacy database/spreadsheet systems.
Augmentation potential	claude-haiku-4-5-20251001	4/5	AI systems excel at assisting with scheduling, reminder generation, data tracking dashboards, and flagging missing follow-ups, substantially improving research assistant productivity while humans retain control over sensitive participant communications and protocol decisions.
Augmentation potential	claude-sonnet-5	3/5	AI can assist by automating reminder emails, flagging non-responders, and drafting follow-up messages, meaningfully easing part of the administrative burden while humans manage actual participant relationships.
Task automatability	claude-haiku-4-5-20251001	3/5	Tracking participants and basic follow-up (sending reminders, scheduling) can be partially automated with CRM-like systems and email automation, but complex participant issues, consent management, and sensitive follow-ups still require human judgment and empathy.
Task automatability	claude-sonnet-5	2/5	Tracking and scheduling can be partly automated via CRM/reminder systems, but locating lost participants, judging appropriate follow-up actions, and handling nuanced participant communications still require human judgment and effort.
Adoption barriers	claude-haiku-4-5-20251001	4/5	Research involving human subjects involves IRB oversight, informed consent protocols, and duty-of-care obligations that typically require human staff involvement and legal accountability, creating regulatory barriers to full automation.
Adoption barriers	claude-sonnet-5	2/5	No licensing requirement, but IRB/ethics protocols around participant contact and privacy impose some procedural friction and require documented human oversight.
Cost vs. human wage	claude-haiku-4-5-20251001	4/5	Basic automation (scheduling, reminders, data entry) is substantially cheaper than human labor; however, oversight and exception-handling add cost that prevents a full order-of-magnitude difference.
Cost vs. human wage	claude-sonnet-5	3/5	Automated reminders/emails are cheap, but the overall task including retention efforts and troubleshooting non-responsive participants still requires human labor, making costs roughly comparable when the full task is considered.
Technical feasibility today	claude-haiku-4-5-20251001	3/5	Tools exist for participant tracking (databases, scheduling software, automated reminders), but systems rarely integrate full participant management with the contextual reasoning needed for sensitive research populations and protocol compliance.
Technical feasibility today	claude-sonnet-5	2/5	CRM and study-management tools exist for participant tracking, but they are largely rule-based database systems rather than AI-driven, and reliable AI-driven follow-up (e.g., outreach, retention efforts) is not standard in deployed research tools.

Provide assistance in the design of survey instruments such as questionnaires.

CI 30–46 · exposure 30 · augmentation 75 · importance 3.5/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	2/5	Academic and research organizations have adopted AI writing tools, but survey instrument design remains methodologically conservative and slower to automate than general writing tasks. Adoption is still in pilot and early-adoption phases; production displacement is limited.
Sector adoption velocity	claude-sonnet-5	3/5	Academic and social science research settings are moderately digitized with growing LLM use for drafting tasks, but rigorous methodological workflows adopt AI more cautiously than fast-moving commercial sectors.
Augmentation potential	claude-haiku-4-5-20251001	4/5	AI substantially assists survey design by rapidly generating question variants, flagging ambiguous phrasing, and offering formatting suggestions. Researchers using AI assistants can iterate and refine instruments faster, with AI handling routine drafting while the expert retains final judgment on validity and appropriateness.
Augmentation potential	claude-sonnet-5	4/5	AI is genuinely useful for brainstorming question wording, checking clarity, translating items, and suggesting response scales, meaningfully speeding up a research assistant's drafting work while a human finalizes design choices.
Task automatability	claude-haiku-4-5-20251001	2/5	AI can help generate survey question templates and check for ambiguity, but designing effective survey instruments requires domain expertise, understanding research objectives, and validation against cognitive and statistical principles that AI cannot reliably perform end-to-end. The task involves substantial human judgment on question order, skip logic, and testing that current AI cannot automate to the 50% time-saving threshold.
Task automatability	claude-sonnet-5	2/5	AI can draft candidate survey questions and suggest structure, but designing valid, unbiased instruments requires domain judgment, iterative pretesting, and stakeholder input that current AI cannot fully replace end-to-end.
Adoption barriers	claude-haiku-4-5-20251001	3/5	Survey design is not legally restricted, but organizational best practices and methodological standards create friction. Institutional review boards and research integrity requirements mean organizations prefer human-certified survey design, and liability for poor instrument design creates reluctance to rely solely on AI output.
Adoption barriers	claude-sonnet-5	2/5	No licensing requirement, but methodological rigor, IRB/ethics review, and PI sign-off on instrument validity create moderate organizational and quality-control friction.
Cost vs. human wage	claude-haiku-4-5-20251001	2/5	AI inference cost is low, but the oversight burden is high because survey design errors are costly; researchers must verify every suggestion for validity and fitness-to-purpose. The total cost of AI-assisted design with mandatory expert review is comparable to or exceeds employing a research assistant directly.
Cost vs. human wage	claude-sonnet-5	3/5	AI drafting assistance is cheap per query, but the overall task still requires substantial human expert review and validation, keeping blended cost roughly comparable to a research assistant's time for quality work.
Technical feasibility today	claude-haiku-4-5-20251001	3/5	Products like ChatGPT and specialized survey platforms offer assistance with question drafting and format suggestions, but deployed systems lack rigorous validation of survey quality, bias detection, and compliance with methodological standards. Error rates in suggested wording and question construction remain material enough that human review is essential.
Technical feasibility today	claude-sonnet-5	2/5	General LLM products can generate draft questionnaire items and flag wording issues, but no deployed product reliably handles full survey instrument design (construct validity, scale selection, bias checks) in production research settings.

Recruit and schedule research participants.

CI 25–46 · exposure 30 · augmentation 63 · importance 4.2/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	2/5	Academic and non-profit research settings (where most social science research occurs) are slow to adopt AI automation, especially for human-subject interactions. Adoption remains limited to pilot scheduling tools in a small fraction of institutions.
Sector adoption velocity	claude-sonnet-5	3/5	Academic and research institutions have adopted scheduling and online recruitment tools moderately, but many labs still rely on manual outreach and coordination.
Augmentation potential	claude-haiku-4-5-20251001	3/5	AI can meaningfully assist by drafting recruitment messages, suggesting outreach strategies, and automating calendar scheduling, helping research assistants work faster. However, the human must retain responsibility for consent, eligibility verification, and relationship management.
Augmentation potential	claude-sonnet-5	4/5	AI-driven scheduling assistants, automated reminders, and recruitment platforms significantly reduce administrative burden and speed up participant pipeline management while humans retain final decision-making.
Task automatability	claude-haiku-4-5-20251001	2/5	While AI can generate recruitment messages and manage scheduling calendars, recruiting requires navigating consent, eligibility screening, and rapport-building that depend on nuanced communication and legal compliance. Current systems cannot reliably handle the full recruitment workflow end-to-end with 50% time savings at equal quality.
Task automatability	claude-sonnet-5	2/5	Scheduling logistics can be automated with calendar tools, but recruitment involves screening, persuasion, and outreach that still typically requires human judgment and relationship management, limiting full end-to-end automation.
Adoption barriers	claude-haiku-4-5-20251001	4/5	Research recruitment is heavily gated by institutional review board (IRB) requirements, informed-consent protocols, and legal liability for proper participant screening and documentation. These regulatory and liability barriers mean that human researchers typically must oversee or sign off on recruitment, preventing full automation.
Adoption barriers	claude-sonnet-5	2/5	No licensing requirement, though IRB/ethics compliance and participant consent processes create some friction requiring human accountability.
Cost vs. human wage	claude-haiku-4-5-20251001	2/5	AI-assisted recruitment (chatbots, scheduling automation) still requires human oversight of consent, screening, and communication quality. The combined cost of AI infrastructure, integration, and required human review remains comparable to or exceeds the cost of a research assistant doing the work directly.
Cost vs. human wage	claude-sonnet-5	3/5	Automated scheduling tools are cheap, but recruitment platforms and screening still involve subscription costs and human review, making overall cost savings moderate rather than dramatic.
Technical feasibility today	claude-haiku-4-5-20251001	2/5	Email outreach and scheduling assistants exist (e.g., calendar AI, email templates), but deployed products struggle with the interpersonal negotiation, eligibility determination, and informed-consent documentation required in research recruitment. No mature production system handles the full task reliably.
Technical feasibility today	claude-sonnet-5	3/5	Scheduling assistants and CRM-like recruitment tools (e.g., Calendly, survey panel platforms like Prolific) are deployed and reliable for logistics, but participant recruitment quality control still often needs human oversight.

Administer standardized tests to research subjects, or interview them to collect research data.

CI 29–37 · exposure 25 · augmentation 63 · importance 4.4/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	2/5	Academic and research institutions adopt new methods slowly; most funded research still uses human RAs for test administration and interviews due to IRB conservatism, protocol specificity, and institutional inertia. Adoption remains in pilot phase rather than widespread displacement.
Sector adoption velocity	claude-sonnet-5	2/5	Academic and social science research settings are generally slow adopters of AI-driven data collection tools compared to fast-moving digital-native sectors.
Augmentation potential	claude-haiku-4-5-20251001	3/5	AI can assist by transcribing interviews, flagging anomalies in responses, auto-coding qualitative data, and generating summary reports, meaningfully raising researcher productivity. However, the human assistant remains essential for the actual administration and interpersonal components.
Augmentation potential	claude-sonnet-5	4/5	AI can meaningfully assist with test scoring, transcription, generating interview guides, and analyzing responses, boosting researcher productivity even though the human remains central to subject interaction.
Task automatability	claude-haiku-4-5-20251001	2/5	Administering standardized tests can be partially automated (reading prompts, recording responses), but the human interaction, rapport-building, and handling of unexpected subject behaviors are difficult for current AI to manage end-to-end with equal quality. Interviewing requires contextual understanding, adaptive questioning, and trust that AI systems today cannot reliably replicate.
Task automatability	claude-sonnet-5	2/5	Interviewing and administering standardized tests to human subjects requires live rapport, adaptive follow-up, and physical or verbal presence that current AI cannot fully replicate end-to-end, though structured survey delivery can be partially automated.'
Adoption barriers	claude-haiku-4-5-20251001	4/5	IRB (Institutional Review Board) approval and ethical requirements for informed consent, subject welfare, and data protection often mandate human presence and judgment. Many research protocols explicitly require a human administrator for liability and regulatory compliance reasons.
Adoption barriers	claude-sonnet-5	3/5	No licensing requirement typically, but IRB protocols, informed consent processes, and subject preference for human interaction create meaningful friction against full automation.
Cost vs. human wage	claude-haiku-4-5-20251001	4/5	AI-administered tests (simple chatbots, automated prompts) have near-zero marginal cost per session compared to the fully-loaded hourly wage of research assistants ($20–$35/hour). Even accounting for setup, integration, and oversight, automation is substantially cheaper once deployed.
Cost vs. human wage	claude-sonnet-5	3/5	Automated survey/chat tools can be cheaper for large-scale structured data collection, but oversight, subject trust issues, and validation needs keep costs roughly comparable to trained human assistants for rigorous research.
Technical feasibility today	claude-haiku-4-5-20251001	2/5	While chatbot systems can conduct basic surveys and some structured interviews exist as deployed products, they perform poorly on nuanced human interaction, cannot reliably adapt to subject distress or confusion, and do not match human interviewers in data quality. Real-world research protocols typically require human administration.
Technical feasibility today	claude-sonnet-5	2/5	Chatbot-based survey tools and voice assistants exist for simple structured interviews, but deployed products rarely handle nuanced, standardized psychometric test administration reliably in research settings.

Perform needs assessments or consult with clients to determine the types of research and information required.

CI 25–38 · exposure 25 · augmentation 63 · importance 3.4/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	2/5	Social science research institutions have been slow to adopt AI for client-facing assessment work; most operations still rely on human research assistants and established consultation protocols. Adoption remains in the pilot stage rather than production deployment.
Sector adoption velocity	claude-sonnet-5	3/5	Social science and research support functions are in professional services/academia, sectors with moderate AI tool adoption for drafting and analysis, but consultative needs-assessment work still lags in automation.
Augmentation potential	claude-haiku-4-5-20251001	3/5	AI can assist by auto-structuring client responses, suggesting relevant data sources, and flagging common information gaps—meaningfully raising productivity of a human research assistant conducting the assessment. However, the human must lead and validate the consultation.
Augmentation potential	claude-sonnet-5	4/5	AI can help draft intake questionnaires, summarize prior research, suggest methodologies, and organize client requirements, meaningfully speeding up the analyst's preparatory work even though the human-client interaction remains central.
Task automatability	claude-haiku-4-5-20251001	2/5	Needs assessment and client consultation require understanding nuanced client contexts, building trust, and making judgments about information gaps—tasks heavily dependent on human interaction and contextual understanding. Current AI can assist with information gathering and structuring but cannot reliably replace the full discovery dialogue.
Task automatability	claude-sonnet-5	2/5	This requires interactive dialogue, contextual judgment, and relationship-building with clients that current AI cannot reliably conduct end-to-end; AI can support parts (e.g., drafting questions) but not replace the consultative process.
Adoption barriers	claude-haiku-4-5-20251001	4/5	Client trust, confidentiality requirements, and professional standards in social science research create substantial friction. Many clients expect human-conducted needs assessments; regulatory and institutional review requirements often mandate human judgment and sign-off on research scope.
Adoption barriers	claude-sonnet-5	2/5	No formal licensing requirement, but client trust, communication nuance, and organizational preference for human interaction create moderate friction against full automation.
Cost vs. human wage	claude-haiku-4-5-20251001	2/5	AI-assisted intake systems are nascent and require significant setup and human oversight to validate client needs accurately. For now, the cost of implementing, fine-tuning, and supervising such systems likely exceeds the hourly wage of a research assistant conducting these tasks directly.
Cost vs. human wage	claude-sonnet-5	2/5	Because AI cannot fully perform this task, using it still requires substantial human oversight and follow-up, so cost savings versus a human research assistant are limited despite low per-query AI costs.
Technical feasibility today	claude-haiku-4-5-20251001	2/5	No mature deployed products perform end-to-end needs assessment and client consultation at scale. While chatbots can conduct structured interviews, they lack the domain expertise, judgment, and relationship-building capability that deployed solutions in this domain would require, and pilots remain limited.
Technical feasibility today	claude-sonnet-5	2/5	No deployed product autonomously conducts needs assessments or client consultations for research design; chatbots can gather requirements in narrow structured contexts but not the open-ended, nuanced consultation described.

Present research findings to groups of people.

CI 25–35 · exposure 25 · augmentation 75 · importance 3.5/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	2/5	Research organizations and academic institutions have not significantly automated research presentations; human presenters remain the norm. Adoption of AI-generated slides is emerging, but end-to-end replacement of the presentation task itself is rare and not a visible production trend.
Sector adoption velocity	claude-sonnet-5	2/5	Research and academic settings adopt AI for drafting and analysis but still overwhelmingly rely on human researchers to present findings in person or via video conferencing.
Augmentation potential	claude-haiku-4-5-20251001	4/5	AI excels at assisting presentation preparation: generating outline drafts, creating visualizations, summarizing data, and offering feedback on clarity. These tools meaningfully boost a researcher's productivity while the human remains the credible voice delivering findings to the audience.
Augmentation potential	claude-sonnet-5	4/5	AI substantially assists with creating slides, summarizing data, anticipating questions, and refining talking points, meaningfully boosting presentation preparation productivity.
Task automatability	claude-haiku-4-5-20251001	2/5	While AI can generate presentation slides and summarize findings, presenting to live audiences requires real-time audience engagement, dynamic adjustment, and credibility—elements that demand human presence and judgment. Current AI cannot reliably handle audience questions, non-verbal communication, or adaptive explanation in a way that meets the equal-quality bar.
Task automatability	claude-sonnet-5	2/5	AI can help draft slides and talking points, but live presentation delivery, audience engagement, and real-time Q&A require human presence and adaptive judgment that current AI cannot fully replicate end-to-end.'
Adoption barriers	claude-haiku-4-5-20251001	4/5	Organizational and stakeholder expectations strongly favor human presentation of research findings, especially in academic and professional contexts where credibility and accountability matter. Audiences typically expect to see and interact with the researcher, creating a cultural and legitimacy barrier to full automation.
Adoption barriers	claude-sonnet-5	2/5	No strict licensing requirement, but norms of academic/research settings strongly favor human presenters for credibility, audience trust, and interactive discussion.
Cost vs. human wage	claude-haiku-4-5-20251001	2/5	Generating slides and speaker notes is cheap at scale, but the marginal cost of a human presenter remains low for presentations that require credibility and live interaction. The all-in cost of AI presentation infrastructure plus human oversight does not yet undercut hiring a research assistant or researcher to present.
Cost vs. human wage	claude-sonnet-5	2/5	Preparing materials with AI is cheap, but the human delivery portion still requires the researcher's time and salary, so overall cost savings versus a human presenter are modest.
Technical feasibility today	claude-haiku-4-5-20251001	2/5	AI systems can produce presentation materials and deliver scripted content via video, but no deployed product reliably performs live group presentations with the nuance, responsiveness, and authority expected in research contexts. Automated delivery systems exist but lack the adaptability and trust-building capacity of human presentation.
Technical feasibility today	claude-sonnet-5	2/5	Products exist for generating slide decks and even AI avatars for presentations, but deployed use for actually presenting research findings to live groups in professional/academic settings is rare and unreliable.

Screen potential subjects to determine their suitability as study participants.

CI 25–34 · exposure 25 · augmentation 63 · importance 4.2/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	2/5	Academic and clinical research settings—the primary sectors performing this task—have lagged in AI adoption for core human subjects procedures. Adoption remains pilot-stage, with most organizations preferring human staff for the gatekeeping and liability-sensitive role of participant screening.
Sector adoption velocity	claude-sonnet-5	2/5	Academic and social science research settings adopt AI tools slowly due to ethics review processes, funding constraints, and small-team structures typical of research assistant roles.
Augmentation potential	claude-haiku-4-5-20251001	3/5	AI can usefully assist by pre-populating eligibility checklists, flagging obvious disqualifications, and organizing participant data, helping coordinators work faster. However, the final eligibility judgment and informed consent verification remain human-centered, limiting the transformative potential of augmentation.
Augmentation potential	claude-sonnet-5	4/5	AI can meaningfully assist by pre-screening applicants against criteria, flagging inconsistencies, and drafting screening questions, while a human remains responsible for final suitability determinations.
Task automatability	claude-haiku-4-5-20251001	2/5	Screening participants requires evaluating complex inclusion/exclusion criteria against individual circumstances, medical histories, and contextual nuance that varies by study design. While AI can flag obvious disqualifications from structured data, the nuanced judgment calls and edge cases that characterize real screening—especially in human subjects research—demand expert oversight and typically fall short of 50% autonomous time savings.
Task automatability	claude-sonnet-5	2/5	Screening requires judgment calls about eligibility criteria, rapport-building, and sensitive follow-up questions that current AI cannot reliably automate end-to-end, though structured intake questionnaires can be partially automated.
Adoption barriers	claude-haiku-4-5-20251001	4/5	Human subjects research is heavily regulated (IRB oversight, informed consent requirements, liability for adverse inclusion decisions). Regulatory bodies and institutional review boards typically require human researchers to verify participant eligibility and confirm informed understanding, creating legal and organizational barriers to full automation.
Adoption barriers	claude-sonnet-5	3/5	IRB/ethics oversight and human-subjects protection rules impose procedural requirements and often require documented human decision-making, though not a hard licensing requirement for AI exclusion.
Cost vs. human wage	claude-haiku-4-5-20251001	2/5	Current AI systems require significant integration (study-specific rule coding or fine-tuning), human oversight of every decision, and regulatory documentation, making the all-in cost per screened participant comparable to or higher than a trained research coordinator.
Cost vs. human wage	claude-sonnet-5	3/5	Automated screening forms and chatbots can reduce cost for high-volume, low-complexity criteria, but complex or sensitive study populations still need human judgment, keeping the average cost ratio near parity.
Technical feasibility today	claude-haiku-4-5-20251001	2/5	No deployed product reliably performs independent participant screening in human subjects research at scale. Existing tools can assist with questionnaire administration and basic data validation, but eligibility determination in real studies requires human verification due to liability, regulatory oversight, and the need to confirm informed understanding—limiting current deployable capability.
Technical feasibility today	claude-sonnet-5	2/5	Some survey/chatbot tools exist for intake screening (e.g., automated eligibility questionnaires), but deployed products for nuanced human-subjects screening in research contexts remain narrow and largely unproven at scale.

Develop and implement research quality control procedures.

CI 23–35 · exposure 25 · augmentation 63 · importance 4.0/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	1/5	Research institutions have been slow to automate QC procedure development; adoption remains limited to pilot projects in well-resourced academic settings, and most organizations still rely on human experts to design and implement QC protocols.
Sector adoption velocity	claude-sonnet-5	2/5	Academic and social science research settings are slower adopters of AI automation for methodological/quality-control work compared to fast-moving sectors like finance or customer service; pilots exist but production-level adoption is limited.
Augmentation potential	claude-haiku-4-5-20251001	3/5	AI could usefully assist by generating drafts of procedural checklists, identifying common QC failures in literature, or flagging methodological issues in existing procedures, but a human researcher must ultimately design, validate, and take responsibility for the QC system.
Augmentation potential	claude-sonnet-5	4/5	AI tools can meaningfully assist by drafting QC checklists, flagging data anomalies, suggesting validation scripts, and summarizing best practices, substantially speeding up the human-led development and implementation process.
Task automatability	claude-haiku-4-5-20251001	2/5	Quality control procedure development requires domain expertise, judgment calls on acceptable error thresholds, and understanding of research methodology nuances that current AI cannot reliably perform end-to-end. While AI could assist in documenting or templating some procedural steps, the core work of designing and validating QC protocols remains substantially human-dependent.
Task automatability	claude-sonnet-5	2/5	Designing and implementing quality control procedures requires judgment about study design, domain-specific validity concerns, and institutional context that current AI cannot autonomously handle end-to-end. AI can assist with parts (checklists, data validation scripts) but not the full development/implementation cycle.
Adoption barriers	claude-haiku-4-5-20251001	4/5	Research quality control often falls under institutional review requirements, funder mandates, and regulatory compliance (IRB, data integrity standards), creating legal and organizational barriers. Institutional liability for QC failures typically requires human accountability and sign-off.
Adoption barriers	claude-sonnet-5	2/5	No licensing requirement exists for this task, but institutional/organizational norms (IRB, methodological rigor expectations) create moderate friction against fully automating quality control without human sign-off.
Cost vs. human wage	claude-haiku-4-5-20251001	2/5	The cost of AI systems capable of this work (domain-specific models, integration, extensive human oversight to ensure procedural validity) would exceed the cost of a research assistant performing the task directly, especially given the high cost of QC failures in research.
Cost vs. human wage	claude-sonnet-5	2/5	Because AI cannot fully replace the judgment and implementation oversight required, a human must still design and validate procedures, so AI only reduces some costs (e.g., drafting checklists) rather than replacing the bulk of paid labor.
Technical feasibility today	claude-haiku-4-5-20251001	2/5	No mature production systems exist that independently develop and implement research QC procedures; this task requires understanding of specific research contexts, institutional requirements, and domain standards that deployed AI tools have not demonstrated reliably. Some research software includes QC templates, but these require significant human customization and validation.
Technical feasibility today	claude-sonnet-5	2/5	There are no deployed products that autonomously develop and implement research QC procedures for social science; existing tools address narrow subtasks like data cleaning or statistical checks, not the holistic procedure design and rollout.

Supervise the work of survey interviewers.

CI 25–30 · exposure 25 · augmentation 50 · importance 3.4/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	2/5	Research organizations and survey firms are exploring call-quality analytics and basic performance dashboards, but human supervisors remain standard. Adoption of AI-driven supervision is slow and limited to analytics support, not replacement, reflecting sector conservatism around research integrity and personnel oversight.
Sector adoption velocity	claude-sonnet-5	2/5	Social science research organizations are moderate adopters of AI for data quality checks but slow to adopt AI for personnel supervision functions.
Augmentation potential	claude-haiku-4-5-20251001	3/5	AI can assist supervisors by automatically flagging performance outliers, extracting call transcripts, and generating quality-assurance dashboards, meaningfully reducing manual review time. However, the core tasks of coaching, hiring decisions, and resolving quality issues still require human judgment and interpersonal skill.
Augmentation potential	claude-sonnet-5	3/5	AI can help by automatically monitoring call quality, transcribing interviews, and flagging outliers, giving supervisors more efficient oversight tools.
Task automatability	claude-haiku-4-5-20251001	2/5	Supervising survey interviewers requires real-time quality assessment, interpersonal feedback, and contextual judgment about interviewer performance. While AI could flag some objective metrics (call duration, completion rates), it cannot reliably assess interview quality, provide meaningful coaching, or handle the nuanced personnel management aspects that define this task.
Task automatability	claude-sonnet-5	2/5	Supervising interviewers requires real-time judgment, coaching, and interpersonal management that current AI cannot fully replicate; only narrow subtasks like flagging data anomalies could be automated.
Adoption barriers	claude-haiku-4-5-20251001	4/5	Supervising interviewers carries implicit liability and quality-assurance responsibilities; many research organizations, especially those handling sensitive surveys or regulated data, expect human supervisors to review and sign off on survey quality and staff management decisions. Organizational norms and quality-control requirements create strong friction against full automation.
Adoption barriers	claude-sonnet-5	3/5	No formal licensing requirement, but organizational norms and accountability for interviewer performance and ethics typically require human oversight.
Cost vs. human wage	claude-haiku-4-5-20251001	2/5	Implementing AI quality monitoring, call recording, and analytics systems requires substantial setup and ongoing integration costs. The savings from partial automation of metrics tracking do not yet offset the infrastructure and human oversight needed, keeping total cost near or above a supervisor's loaded wage.
Cost vs. human wage	claude-sonnet-5	2/5	While AI monitoring dashboards are cheap to run, the human judgment and management components still require paid staff, keeping overall costs comparable to a human supervisor.
Technical feasibility today	claude-haiku-4-5-20251001	2/5	No deployed product reliably supervises human survey teams end-to-end. AI can monitor some call-center metrics automatically, but evaluating interviewer adherence to survey protocols, coaching quality, and personnel management remains primarily manual with only nascent automation aids.
Technical feasibility today	claude-sonnet-5	2/5	Some quality-monitoring and analytics tools exist to flag interview inconsistencies, but no deployed product performs holistic supervision of human interviewers reliably.

Allocate and manage laboratory space and resources.

CI 7–30 · exposure 13 · augmentation 38 · importance 3.0/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	2/5	Most academic and research institutions manage lab resources through legacy systems or administrative staff rather than AI-driven solutions. Adoption of AI for this function remains limited, with institutions moving slowly due to institutional inertia and the embedded human nature of negotiation and conflict resolution.
Sector adoption velocity	claude-sonnet-5	2/5	Research and academic lab environments have historically low digitization of physical resource management, with slow uptake of AI tools for such logistics.
Augmentation potential	claude-haiku-4-5-20251001	3/5	AI can meaningfully assist by generating scheduling suggestions, tracking inventory, and flagging resource conflicts, helping a research assistant allocate space more efficiently. However, the core task of stakeholder coordination and final allocation decisions still requires human judgment and presence.
Augmentation potential	claude-sonnet-5	2/5	AI could help with scheduling software or inventory tracking suggestions, but it offers only marginal assistance to the core physical/administrative allocation task.
Task automatability	claude-haiku-4-5-20251001	2/5	While AI could assist with scheduling and inventory tracking, allocating and managing lab space requires real-time decision-making, stakeholder negotiation, and physical coordination that current AI cannot fully automate. The task involves balancing competing resource demands and responding to dynamic needs that go well beyond data entry or simple optimization.
Task automatability	claude-sonnet-5	1/5	This requires physical presence, spatial judgment, and coordination with lab personnel to allocate equipment and space—no AI system can perform this end-to-end task itself.
Adoption barriers	claude-haiku-4-5-20251001	4/5	Lab resource allocation often involves institutional policies, safety compliance, and approval workflows that require human authority and accountability. Additionally, effective space management depends on building relationships and informal agreements with faculty and lab personnel, creating organizational friction against full automation.
Adoption barriers	claude-sonnet-5	3/5	While no license is strictly required, organizational authority, accountability for shared resources, and physical oversight create real friction against full automation.
Cost vs. human wage	claude-haiku-4-5-20251001	2/5	Current AI tools for scheduling and resource tracking would require significant integration and ongoing oversight to be useful, making total cost (inference, setup, human validation) comparable to or exceeding the cost of a research assistant managing these tasks directly.
Cost vs. human wage	claude-sonnet-5	1/5	AI cannot substitute for the physical and administrative labor involved, so there is no meaningful AI cost basis to compare against human wages for this task.
Technical feasibility today	claude-haiku-4-5-20251001	2/5	No deployed product reliably manages the full scope of lab resource allocation and space management independently. While calendar and inventory systems exist, they require substantial human oversight and cannot handle the nuanced interpersonal and logistical aspects of real lab operations.
Technical feasibility today	claude-sonnet-5	1/5	No deployed AI product manages physical laboratory space and resource allocation; this remains a human logistical and administrative function.

Obtain informed consent of research subjects or their guardians.

CI 4–15 · exposure 8 · augmentation 38 · importance 4.5/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	1/5	Adoption of AI for informed consent is minimal because regulatory and ethical requirements explicitly mandate human involvement. Research institutions cannot and do not substitute AI for human consent acquisition.
Sector adoption velocity	claude-sonnet-5	2/5	Academic and research institutions are cautious adopters of AI for compliance-sensitive tasks like consent, with slow uptake due to ethical and regulatory scrutiny.
Augmentation potential	claude-haiku-4-5-20251001	2/5	AI can assist by generating plain-language summaries of consent documents or providing reminders of key points to cover, but the core interpretive, responsive, and accountability dimensions of obtaining consent remain solidly human-driven and augmentation is limited.
Augmentation potential	claude-sonnet-5	3/5	AI can usefully generate consent form drafts, translate materials, or answer subject questions, augmenting the researcher's efficiency in preparing for consent conversations.
Task automatability	claude-haiku-4-5-20251001	1/5	Informed consent requires genuine human understanding, voluntary agreement, and legal/ethical accountability that cannot be meaningfully automated. An AI system cannot legally or ethically obtain true informed consent, as it cannot verify comprehension, assess capacity, or bear responsibility for the consent process.
Task automatability	claude-sonnet-5	2/5	AI can help draft consent forms and explain procedures, but obtaining actual informed consent requires interpersonal interaction, verifying understanding, and often in-person or verbal engagement with human subjects, which current AI cannot fully replace end-to-end.'
Adoption barriers	claude-haiku-4-5-20251001	5/5	Obtaining informed consent is a legally mandated requirement in research ethics and regulatory frameworks (e.g., FDA, IRB). A qualified human must document and be responsible for consent; this is a hard regulatory and liability barrier that prevents automation.
Adoption barriers	claude-sonnet-5	5/5	IRB and regulatory requirements (e.g., Common Rule, HIPAA) mandate that a qualified human researcher ensure and document genuine informed consent, making this a hard compliance barrier.
Cost vs. human wage	claude-haiku-4-5-20251001	2/5	AI might assist with drafting or explanation materials at low cost, but the core task—obtaining valid consent—still requires trained human research staff. Any cost savings from AI assistance are marginal compared to the required human labor.
Cost vs. human wage	claude-sonnet-5	2/5	While drafting consent documents is cheap via AI, the human-mediated interaction and legal/ethical verification steps still require paid staff time, limiting cost savings.
Technical feasibility today	claude-haiku-4-5-20251001	1/5	No deployed product reliably performs informed consent acquisition end-to-end. While AI can draft consent forms or assist in explaining information, actual consent requires a human agent capable of answering questions, assessing understanding, and taking legal responsibility.
Technical feasibility today	claude-sonnet-5	1/5	No deployed product performs the full consent-obtaining process (explaining risks, answering questions, verifying comprehension, securing signatures) reliably in research settings today.

Related occupations — Life, Physical & Social Science

How to read this

A high substitution score does not mean this job disappears — it means a large share of its current tasks face replacement pressure, so the mix of tasks is likely to change. High augmentation alongside substitution typically means the occupation reorganizes around the protected tasks. Wide confidence intervals mean the rater panel disagreed: treat those scores as open questions, not verdicts.

What would change this score

New model capabilities (automatability, feasibility), falling inference costs (cost ratio), regulation and licensing shifts (barriers), and measured sector adoption (velocity) all re-enter at every index release. Each release is recomputed, versioned and kept queryable — scores are claims with a date on them, not permanent labels.