Data scientists

Clean and structure raw data to make them readable by software programs

Data cleaning and structuring is highly automatable—AI can handle missing values, standardize formats, detect outliers, and transform schemas. This is already a strength of modern data tools augmented by LLMs that can write cleaning scripts from natural language descriptions.

BLS evidence: Data scientists must 'clean' the raw data, a process by which they structure the data to make them readable by software programs.

Use data visualization software to present findings as charts, maps, and graphics

AI can generate publication-ready visualizations from data using tools like code-generating LLMs that produce matplotlib/ggplot/Tableau code. The main human role is selecting which insights to highlight and ensuring visual clarity for the audience.

BLS evidence: Data scientists 'Use data visualization software to present findings' and often present their findings as charts, maps, and other graphics.

Collect, categorize, and analyze data from various sources

AI systems can now autonomously collect data via APIs, categorize using LLMs, and perform standard analyses. Human oversight needed mainly for defining scope and validating unusual patterns, but the bulk of routine data collection and analysis is automatable.

BLS evidence: Data scientists typically 'Collect, categorize, and analyze data' and may use various methods to obtain data including access to databases or web-scraping tools.

Develop or recommend systems and enhance web-browsing functions

AI can write code for system enhancements and web functions, suggest architectural improvements, and implement features. Human developers still needed for integration decisions and handling edge cases, but AI substantially reduces the labor content.

BLS evidence: Data scientists with strong coding or engineering backgrounds may develop or recommend systems, build machine learning algorithms, and devise ways to enhance web-browsing functions.

Create, validate, test, and update algorithms and models for machine learning

Modern AI can generate, test, and iterate on ML models using AutoML frameworks and code generation. However, validating model appropriateness for business context and updating models based on shifting requirements still requires substantial human judgment.

BLS evidence: Data scientists 'Create, validate, test, and update algorithms and models' and develop algorithms to support programs for machine learning.

Conduct research for reports or academic journals

AI can draft literature reviews, synthesize findings, and structure research reports. However, formulating novel research questions, designing rigorous methodologies, and ensuring academic integrity require human oversight, though AI accelerates the writing and analysis phases significantly.

BLS evidence: Some data scientists conduct research for reports or academic journals.

Determine which data are available and useful for the project

AI can assess data availability, quality metrics, and relevance to objectives, but determining what's truly 'useful' requires understanding unstated business priorities and making judgment calls about ROI that benefit from human domain expertise.

BLS evidence: Data scientists typically 'Determine which data are available and useful for the project' and often begin by gathering or identifying relevant data sources.

Make business recommendations to stakeholders based on data analysis

AI can draft recommendations and identify patterns, but translating technical findings into actionable business strategy requires understanding organizational politics, risk tolerance, and implementation constraints that AI cannot reliably navigate without human guidance.

BLS evidence: Data scientists 'Make business recommendations to stakeholders based on data analysis' to help inform business decisions or process changes.

Communicate analyses to technical and nontechnical audiences

AI can generate clear written explanations and adapt technical depth, but live communication involves reading the room, handling unexpected questions, building trust with skeptical stakeholders, and adjusting messaging based on subtle social cues that AI cannot yet manage.

BLS evidence: Visualization techniques allow data scientists to clearly communicate their analyses to technical and nontechnical audiences, including colleagues, managers, and clients.