Artificial intelligence and data science come from different traditions. Data science grew out of statistics: John Tukey argued in 1962 that learning from data deserved to be treated as a science, and William Cleveland published an action plan titled "Data Science" in 2001. Artificial intelligence began with a 1955 proposal for a Dartmouth summer study on whether machines could simulate learning and other features of intelligence.
Key takeaways
- Tukey's 1962 paper "The Future of Data Analysis" described data analysis as a science defined by a problem, not by a subject, and listed four forces acting on it.
- Cleveland's 2001 plan for statistics departments set six technical areas and gave each a share of effort, with theory capped at 20 percent.
- David Donoho's 2015 review describes the data science of that moment as a superset of statistics and machine learning, plus tools for scaling up to big data.
- The Dartmouth proposal concerns building machines that simulate intelligence, so the two fields have different starting questions even where their methods overlap.
Tukey's 1962 case for a science of data analysis
The earliest document in the data science line is a long paper in The Annals of Mathematical Statistics, volume 33, issue 1, March 1962, pages 1 to 67. Tukey wrote that, having watched mathematical statistics evolve, he had come to see his central interest as data analysis. By that he meant procedures for analyzing data, techniques for interpreting their results, ways of planning data gathering so analysis is easier and more accurate, and the machinery of mathematical statistics where it applies.
Tukey also argued that data analysis meets three tests of a science: intellectual content, organization in an understandable form, and reliance on experience as the final standard of validity. Mathematics, he said, fails the third test, because its standard is logical consistency. The four influences he named were the formal theories of statistics, faster computers and display devices, larger bodies of data in many fields, and the spread of quantification into more disciplines. Donoho notes that this list reads like the press releases of later data science initiatives.
Cleveland's 2001 action plan and its percentages
Cleveland, working from Bell Labs, wrote "Data Science: An Action Plan for Expanding the Technical Areas of the Field of Statistics" in International Statistical Review, volume 69, issue 1, pages 21 to 26, 2001. The plan told university statistics departments to reorient around the data analyst. Work was to be judged by how far it helps an analyst learn from data, directly through tools or indirectly through theory that leads to tools. Donoho reads this paper as perhaps the earliest clear use of "data science" in its modern sense, and he mentions an earlier proposal by Jeff Wu that statisticians rename their profession for a similar reason.
| Cleveland's technical area | Suggested share of effort |
|---|---|
| Multidisciplinary investigations | 25 percent |
| Models and methods for data | 20 percent |
| Theory | 20 percent |
| Computing with data | 15 percent |
| Pedagogy | 15 percent |
| Tool evaluation | 5 percent |
Donoho observed that several statistics departments could have fit all of their activity into the 20 percent that Cleveland allowed for theory. The plan was a call to change what the field spends its time on, and it did not describe a finished discipline.
Donoho's review of the data science moment
Donoho presented his paper at the Tukey Centennial workshop in Princeton on September 18, 2015, and the preprint hosted on an MIT course page was later published in the Journal of Computational and Graphical Statistics, volume 26, issue 4, pages 745 to 766, 2017. It was written days after the University of Michigan announced a USD 100 million Data Science Initiative to hire 35 faculty, and it asks whether data science differs from statistics at all.
His answer is that the field being built in universities was a superset of statistics and machine learning with technology added for large data, a choice he says was driven more by commercial than by intellectual developments. He proposes instead six divisions of activity, listed in the next section, which together he calls Greater Data Science. Machine learning therefore sits inside data science in his account, not beside it.
Six divisions of data science work
- Data exploration and preparation
- Data representation and transformation
- Computing with data
- Data modeling
- Data visualization and presentation
- Science about data science
Only one of the six is about building predictive models. The others cover the work of cleaning, restructuring, programming, presenting, and studying the practice itself, which is why a data science job can involve little model building.
Where artificial intelligence starts from a different question
The Dartmouth proposal, dated August 31, 1955 and reprinted in AI Magazine in 2006, had four authors: McCarthy, Minsky, Rochester, and Shannon. They conjectured that learning, and any other feature of intelligence, could in principle be specified in enough detail for a machine to imitate it. The AAAI reprint is the primary record.
Put side by side, the two founding texts ask different things. Tukey and Cleveland ask how a person can learn reliably from data. The Dartmouth authors ask whether a machine can be made to do things that normally need intelligence. Machine learning, where a program improves from data, is the region where the two questions meet, and that overlap is the main reason job titles and course names blur.
A rule of thumb for choosing the label
These questions are an editorial heuristic built on the sources above, not a definition from any standards body.
- Is the main product a finding, chart, or estimate that a person reads and decides on? That leans toward data science.
- Is the main product a system that takes inputs and produces actions or outputs on its own? That leans toward artificial intelligence.
- Does the work involve training a model from data? Both labels may apply, and the project description should name the task instead.
- Is most of the effort spent on data preparation and presentation? That matches Donoho's first and fifth divisions, whatever the job title says.
FAQ
Is data science a part of artificial intelligence?
No single authority settles this. Donoho describes data science as a superset of statistics and machine learning, which puts machine learning inside it. Artificial intelligence, as framed at Dartmouth, is a separate research aim, and the two overlap in machine learning.
Who coined the term data science?
Donoho credits Cleveland's 2001 paper as possibly the earliest use where the meaning is clearly the modern one, and he cites an earlier suggestion by Jeff Wu to rename statistics. Earlier uses of related terms exist, so the question has no single settled answer.
Why did Tukey call it data analysis instead of statistics?
He wanted a name for the whole activity of learning from data, including planning data collection and interpreting results, not only the mathematical theory of inference. Peter Huber later said Tukey stretched the term so far that it covered all of statistics.
Does this article say which field is better?
No. The sources compare origins and activities, and none of them ranks one field above the other.
Related reading
- How AI is used in research and discovery
- Computer vision as one branch of AI
- What deep learning is and does
- Artificial intelligence compared with computer science
More in AI Basics.
Last verified: October 2026
Sources
- The Future of Data Analysis, Tukey JW, The Annals of Mathematical Statistics 33(1), 1962
- Data Science: An Action Plan for Expanding the Technical Areas of the Field of Statistics, Cleveland WS, International Statistical Review 69(1), 2001
- 50 Years of Data Science, Donoho D, Journal of Computational and Graphical Statistics 26(4), 2017; preprint dated September 18, 2015
- A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence, McCarthy J, Minsky ML, Rochester N, Shannon CE, August 31, 1955, reprinted in AI Magazine 27(4), 2006