Judgment in the Age of AI · Article 40
A chart can contain correct numbers and still create a false impression through axes, grouping or missing context. AI often optimises for apparent clarity, but clarity does not guarantee honest proportions, valid comparisons or a supported conclusion.
Disassemble the chart from outside to inside
- Title: Does it describe the data or announce a conclusion for the data?
- Axes: Are they truncated, dual, logarithmic or unevenly spaced?
- Denominator: Are we seeing counts, proportions, per-person values or growth?
- Time: Were start and end points chosen to favour a story?
- Groups: Does an average conceal differences by place, age or another factor?
- Uncertainty: Where are sample size, error and missing data?
Run a redraw test
Show the same data as a table, with a zero baseline or with another defensible grouping. If the story changes dramatically, the conclusion depends on presentation. Then ask someone who has not seen the original chart to write what the data alone suggests.
A responsible chart makes the important point easy to see and its limitations difficult to hide.
References
Discover more from Geoffrey Chen
Subscribe to get the latest posts sent to your email.