# Polling charts for “I want an AI that keeps working”

The dumbbell chart compares 2024 and 2025 for Canada, the United States, China, Japan, Indonesia and Great Britain on a 0–100% scale. Hollow orange markers indicate 2024 and solid teal markers indicate 2025. Equal values use concentric markers at the same position; they are not offset. They measure agreement that AI products and services have more benefits than drawbacks. They are selected country results, not regional averages.

- 2025 edition, 2024 survey: [Stanford AI Index, chapter 8](https://hai.stanford.edu/assets/files/hai_ai-index-report-2025_chapter8_final.pdf#page=7), figure 8.1.2; percentages other than Japan and Great Britain also explicitly listed in chapter highlight 4. Great Britain is 46% in 2024 and 43% in 2025, verified against Ipsos AI Monitor 2025, p. 15; the label follows the survey’s geographic coverage. Japan is 48% in both years, verified against the 2024 and 2025 columns of Ipsos AI Monitor 2025, p. 15. Chapter page 6 gives 23,685 respondents across 32 countries and April–May 2024 fieldwork.
- 2026 edition, 2025 survey: [Stanford AI Index, chapter 9](https://hai.stanford.edu/assets/files/ai_index_report_2026_chapter_9_public_opinion.pdf#page=5), figure 9.1.2, printed p. 364. The 2025 values match the plotted points. Five are cross-checked against [Ipsos AI Monitor 2025](https://resources.ipsos.com/rs/297-CXJ-795/images/Ipsos-AI-Monitor-2025.pdf), p. 15. China (87%) is from Stanford's figure: p. 363, footnote 1 explains that Ipsos supplied China data that may not appear in published Ipsos reports. The chapter reports March 21–April 4, 2025 fieldwork, 23,216 adults and 30 countries for the published survey.

The CSV contains chart-level percentages transcribed from the sources, not respondent-level microdata. Report years and survey years are separate columns. Online survey samples have country-specific representativeness limits. Neither survey identifies the cause of country differences or measures preference for open-weight models. These are the requested AI Index editions, not a claim to have the latest September 2026 polling.

Run `python3 assets/stanford-ai-polling/render.py` from the published directory (or the site’s `public` directory) with matplotlib installed to recreate the PNG. Source notes in the essay use Markdown inside an HTML `div` with `align="center"`, since Markdown alone has no centring syntax; rendering requires support for inline HTML.
