深圳市灵感可视软件开发有限责任公司

可视化一切

Data visualization is often treated as a modern technol…

By

linggankeshi web-editor

—

Three Centuries of Data Visualization: From a Cholera Map to AI-Generated Charts

Data visualization is often treated as a modern technology. Stretch the timeline out, though, and it has existed for more than three hundred years — and many of the arguments we’re having today were had by our predecessors.

This article covers three things: where it came from, what AI is actually changing, and where the industry stands in 2026.

1. Foundations: Three Charts, 1786–1869

The starting point of modern data visualization can be pinned to three charts.

1786 — William Playfair’s The Commercial and Political Atlas

Playfair systematically used line and bar charts to express economic time series: imports and exports, national debt, trade balances. These weren’t the first charts ever drawn, but they were the first time “time series” was fixed as a set of conventions — time on the horizontal axis, value on the vertical, connected by a line. Those conventions are still with us.

1854 — John Snow’s Broad Street cholera map

During a London cholera outbreak, Snow plotted individual deaths on a street-level map. The clustering pointed directly at the Broad Street pump. The map’s value isn’t aesthetic — it’s that the map was an argument. Visualization was used to support causal inference, not merely to present known facts.

1869 — Charles Joseph Minard’s map of Napoleon’s Russian campaign

Minard compressed four information streams — army size, latitude and longitude, time, and temperature — into a single chart: band width encodes troop strength, black and red distinguish advance from retreat.

It’s frequently called “the best statistical graphic ever drawn.” That’s worth flagging: it’s a popular evaluation, not an experimental finding — attributed by later admirers rather than established by contemporary or empirical research. But its design merits are real: multivariate narrative, two-color encoding of direction, and prose reduced to a minimum.

2. Becoming a Discipline: 1967–2005

Those three charts proved visualization was useful. Turning it into a discipline took another century.

YearEventSignificance
1967Jacques Bertin, Semiology of GraphicsIntroduced “visual variables” (position, size, shape, color…) — the first systematic answer to “which encoding for which data”
1987NSF Workshop on Visualization in Scientific Computing (Feb 9–10)“Scientific visualization” established as a field; the resulting report became a founding document
1990First IEEE Visualization conferenceAn academic community formed
1999Card, Mackinlay & Shneiderman, Readings in Information VisualizationGave the canonical definition: “the use of computer-supported, interactive, visual representations of data to amplify cognition”
2005Thomas & Cook, Illuminating the PathIntroduced “visual analytics,” pairing interaction with automated analysis

The 1999 definition deserves attention because it sets out four requirements: computer-supported, interactive, visual representations of data, and aimed at amplifying cognition.

That’s why an animation engine like Manim doesn’t count as a visualization system — it produces video, and viewers can’t interact with it — while a tool like VisuAlgo does, because users can supply custom input and step through execution. Interactivity is a hard requirement.

3. Democratization: 2007–2020

This era’s keyword is “falling barriers.”

  • 2007 — ggplot2 brought the grammar of graphics to R users; declarative charting became the default approach
  • 2011 — D3.js made the browser a visualization platform; web charts could now match print quality
  • The 2010s — data journalism carried visualization out of academia and BI dashboards into public view

The result: the cost of producing a chart fell to nearly zero. Anyone who can write a few lines of code can produce a technically flawless chart.

But a new problem emerged alongside it: there are more charts than ever, and not proportionally more people who understand them.

4. What AI Is Changing

Since 2025, AI has moved deeply into three parts of the visualization pipeline.

Generation: from “I draw the chart” to “the chart appears”

In March 2026, Anthropic enabled Claude to generate inline interactive charts within a conversation; in October 2026 it shipped Claude Dashboards and Motion, pushing “conversational dashboards” into beta.

What changed: previously you chose a tool, wrote code, and tuned styling. Now you describe what you want.

Interaction: natural language becomes the query interface

Traditional BI requires users to understand the data model first — dimensions, measures, granularity — then operate an interface. Natural language removes that layer: users just ask “what was East China’s repeat purchase rate last quarter?”

What changed: the default entry point for BI is shifting from a drag-and-drop panel to a chat box.

Analysis: LLM-assisted visual analytics

Academia moved earlier. IEEE VIS 2026 hosts VISxGenAI (its second edition), and the organizers frame the questions this way:

Will autonomous agents replace human data analysts? Do visualization techniques designed for humans transfer to agent interaction? How do we preserve human agency while introducing agents?

These three questions matter far more than “can AI draw a chart?”

But one thing AI has not made cheaper

AI reduced the cost of producing a chart. It did not reduce the cost of knowing which chart to make.

Choosing the right encoding depends on the task (comparison, distribution, outliers, composition) and on domain knowledge. AI can’t make either judgment for you — it will render your idea beautifully, but if the idea used the wrong encoding, it simply renders the mistake more elegantly.

A concrete case: coloring two regions with a hundredfold difference in magnitude the same shade. No amount of polish makes that conclusion correct. That isn’t a tooling problem — it’s a judgment problem.

5. The Industry in 2026

Standardization: an unusually dense year

2026 saw a rare cluster of visualization standards published or taking effect:

StandardTitleDate
T/JNBDA 0017—2026Specification for Data Visualization Platform ConstructionPublished and effective 2026-08-03
GB/T 46237-2025Information technology — Capability maturity model for digital twinPublished 2025-10-05, effective 2026-05-01
T/SDIOT 064—2026Real-time IoT data mapping and rendering for building digital visualization platforms2026
T/ZGCIT 120—2026Guidelines for AI and digital-twin collaborative cultural tourism visualization platforms2026
T/ACCEM 745-2025Presentation and interaction for immersion liquid cooling data visualization softwarePublished 2025-12-23, effective 2026-01-22

T/JNBDA 0017—2026 is worth noting: it writes platform architecture, functional requirements, and platform acceptance into the standard, and explicitly lists “row- and column-level access control” and “AI agents” as technical pillars. A 2026 visualization platform standard encoding AI agents into its technical stack is itself an industry signal.

Policy: visualization recognized as a tradable data product

In February 2026, China’s National Data Administration and three other agencies issued a policy opinion (Guoshu Zhengce [2026] No. 6) explicitly supporting the development of “verification queries, data analysis reports, data indices, and data visualization” as analytical decision-making data products and services.

This is a national policy document naming data visualization directly, placing it alongside data indices and analysis reports. The implication: visualization is no longer just a phase in project delivery — it’s treated as a tradable data product form.

Academia: China’s deepest participation yet

IEEE VIS 2026 (November 9–13, Boston) adopts a federated format for the first time, with satellite venues in Paris and Tianjin, China. The Tianjin venue is organized by scholars including Jiawan Zhang, Shixia Liu, and Xiaoru Yuan.

For Chinese practitioners this means access to the field’s flagship conference without traveling abroad — previously a privilege for a few.

Commerce: the internationalization of Chinese BI

In Gartner’s Magic Quadrant for Analytics and Business Intelligence Platforms, 2026, Alibaba Cloud’s Quick BI was included for the seventh consecutive year, positioned as a Challenger — the only Chinese BI vendor on the list for seven straight years. Two points in its positioning stand out: a built-in data analysis agent, and usage-based Token billing layered on top of subscriptions.

The latter is a signal: AI analysis is shifting BI cost structures from per-seat subscription toward consumption-based pricing.

Open source: three foundations refreshed at once

  • Apache Superset 6.0 (2026-02-09): described by the project as its most significant visual and architectural overhaul, migrating fully to Ant Design v5
  • Grafana 12.4 (2026-02-25): Git Sync enters preview; dashboards start moving into GitOps
  • Apache ECharts 6.1 (2026-05-19); 6.0 shipped 2025-07-30

Most dashboards and BI platforms in China are built on these three, so their release cadence feeds directly into enterprise upgrade planning.

Conclusion: Three Takeaways

One — the barrier to producing visualizations keeps falling; the barrier to judgment hasn’t. Tools have evolved from hand-drawing to code to AI conversation. Deciding which chart to use still depends on task and domain knowledge, and that hasn’t changed in three hundred years.

Two — interactivity is the dividing line. The 1999 definition made interaction a requirement, and it still holds. Producing video doesn’t make something a visualization system; letting users change parameters, predict the next step, and construct their own input does.

Three — research settled this long ago: engagement determines outcomes, not the visualization itself. The engagement taxonomy in algorithm visualization (view → respond → change → construct → present) has been validated repeatedly. Simply letting students watch an animation has limited effect; having them change parameters and make predictions works.

Together these point to one conclusion: what we’re building isn’t a chart — it’s the process of understanding.

Standards numbers, policy document numbers, and product versions cited here come from public release information. Market-size figures are omitted because they cannot be traced to original research sources.

Sources

发表回复