---
title: "Civilization Zero Evidence Atlas: After Coordinate Zero"
url: "https://civilizationzero.ai/civilization-zero-evidence-atlas.md"
html: "https://civilizationzero.ai/civilization-zero-evidence-atlas"
fenced_text: "a fenced block's text, without its final newline, is the text as received"
description: "Civilization Zero (CZ) evidence atlas: ten panels of measured change since Coordinate Zero, 30 November 2022 — AI use, benchmarks, work, energy, beliefs and rules. Sources cited."
---

# Civilization Zero evidence atlas: after Coordinate Zero

Civilization Zero (CZ) counts its days from Coordinate Zero, 30 November 2022, the day of ChatGPT’s public release. This atlas puts ten measured things since that day on one page: how widely AI is used, how it scores against human reference groups, what it changed in two work experiments, the electricity and capital behind it, what people believe about AI experience, how rules took effect, and, kept apart, forecasts.

**Reported AI use is widespread. Effects vary by task.** These data do not test consciousness, or whether humans are at the centre. “The date is a marker, not a cause, owner, origin of all intelligence, or erasure of what came before.” ([Core §0](https://civilizationzero.ai/core#p0))

The small print is easier to read at full size: [the image at full size](https://civilizationzero.ai/civilization-zero-evidence-atlas-full.png) (2339 × 3308 pixels) or [the PDF](https://civilizationzero.ai/civilization-zero-evidence-atlas.pdf) (A3, for print). Everything in it is also written out below: the ten panels and the sources.

[![Civilization Zero evidence atlas, After Coordinate Zero: ten panels of measured change since 30 November 2022, checked on 3 November 2026. AI chatbot use (49% of US adults); a benchmark against human reference groups (GPQA); the 207-day doubling of the length of task AI agents finish half the time; two work experiments (quality up 1.31 points; developers 19% slower); data-centre electricity (360 to 485 TWh); Microsoft’s capital spending ($44.5 to $115.9 billion); beliefs about AI experience (medians 1% and 5%); the EU AI Act’s stages; and, apart, IEA scenarios for 2030 (833 to 1,008 TWh). The mark of Civilization Zero at the top, civilizationzero.ai at the foot.](https://civilizationzero.ai/civilization-zero-evidence-atlas.png)](https://civilizationzero.ai/civilization-zero-evidence-atlas-full.png)

PNG, 1200 × 1697; every number checked against its primary source on [CZ 1434](https://civilizationzero.ai/calendar?d=CZ%2B1434). Free to share under [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/), credited to Civilization Zero, with a link to ` civilizationzero.ai/civilization-zero-evidence-atlas `. The data stay under their sources’ terms (the IEA’s: CC BY-NC-SA 4.0).

## The ten panels

Each panel names its kind of evidence: a survey, a company report, a benchmark, a fitted trend, an experiment, a model estimate, a filing, a belief survey, a policy record or a scenario. Different panels answer different questions; their values cannot be added into a single “AI progress” score.

1. **A · Reach depends on the denominator** (Survey and company report)
   49% of US adults say they use AI chatbots (February 2026; 5,119 adults; 95% sampling margin ±1.6 points). Pew changed its question in 2026, so no trend is drawn. OpenAI said that 1.2 billion people use ChatGPT every week (September 2026): company-reported, unaudited, and a different denominator. Source: [Pew Research Center, Americans and AI 2026 (17 June 2026)](https://www.pewresearch.org/internet/2026/06/17/americans-and-ai-2026-chatbots-smart-devices-and-views-on-impact/) · [OpenAI DevDay, 29 September 2026, as reported live](https://simonwillison.net/2026/Sep/29/openai-devday-2026-live-blog/).
2. **B · Human reference groups matter** (Benchmark, 2023)
   On the same 546 graduate-level science questions (GPQA’s extended set), domain experts answered 65.4% correctly, skilled non-experts with web access 33.9%, and GPT-4, closed-book with few-shot chain-of-thought prompting, 38.7%. A 2023 baseline, not a current ranking. Source: [Rein et al., GPQA, arXiv 2311.12022 (2023), Table 5](https://arxiv.org/abs/2311.12022).
3. **C · Longer tasks, at a fixed success rate** (Fitted trend, models of 2019 to early 2025)
   The length of task, measured in the time skilled people take, that AI agents complete with 50% success doubled every 207 days (95% bootstrap interval: 166 to 240 days). The fit includes models from before CZ 0. A task horizon is not general intelligence. Source: [Kwa et al. (METR), Measuring AI Ability to Complete Long Tasks, arXiv 2503.14499 v4, section 3.2](https://arxiv.org/abs/2503.14499).
4. **D · AI raised quality in selected tasks** (Experiment, consulting tasks, 2023)
   Consultants given GPT-4 scored 1.31 points higher on a 1–10 quality rubric than the control group without AI, and 1.48 points higher with a prompt overview (385 consultants; selected tasks). The approximate 95% intervals on the image are computed from the published robust standard errors. Source: [Dell’Acqua et al., Organization Science (2026), Table 4](https://doi.org/10.1287/orsc.2025.21838).
5. **E · AI slowed one developer sample** (Experiment, experienced developers, early 2025)
   Sixteen experienced developers working on 246 tasks in open-source projects they knew took 19% longer when AI tools were allowed (95% confidence interval: +2% to +39%). In February 2026 METR said developers are likely sped up more by today’s tools. One sample, not all programming. Source: [Becker et al. (METR), arXiv 2507.09089 (2025)](https://arxiv.org/abs/2507.09089) · [METR, We are Changing our Developer Productivity Experiment Design (24 February 2026)](https://evals.alignment.org/blog/2026-02-24-uplift-update/).
6. **F · Data-centre electricity use increased** (Model estimate, world data centres, 2023–2025)
   All data centres, AI and other work together, used 360 TWh of electricity in 2023, 416 TWh in 2024 and 485 TWh in 2025. Model estimates: no statistical intervals are published, and AI-only consumption is not metered. Source: [IEA, Key Questions on Energy and AI (2026), Table A.1](https://iea.blob.core.windows.net/assets/3179f7f8-01f6-4dd6-bffa-c9f7b73f1dc9/KeyQuestionsonEnergyandAI.pdf).
7. **G · Infrastructure spending increased** (Audited filing, Microsoft)
   Microsoft’s cash additions to property and equipment: $44.477 billion in fiscal 2024, $64.551 billion in fiscal 2025 and $115.948 billion in fiscal 2026 (years ending 30 June). The whole company in nominal dollars, not an AI-only budget; non-cash additions are excluded. Source: [Microsoft, Annual Report 2026, cash flows statements (p. 41)](https://www.sec.gov/Archives/edgar/data/0000789019/000119312526427876/d62902dars.pdf).
8. **H · Beliefs are not a consciousness test** (Belief survey, fielded May 2024)
   Asked for the chance that AI systems already had subjective experience in 2024, the median AI researcher answered 1% and the median US adult 5% (582 AI researchers and 838 US adults surveyed). Beliefs do not establish consciousness. Source: [Dreksler et al., Subjective Experience in AI Systems, arXiv 2506.11945 (2025), section 2.1](https://arxiv.org/abs/2506.11945).
9. **I · Institutional response: EU rules took effect in stages** (Official milestones, EU only)
   The EU AI Act entered into force on 1 August 2024. Its bans and AI-literacy duties applied from 2 February 2025, the obligations for general-purpose AI models from 2 August 2025, and general application from 2 August 2026, with exceptions; a 2026 amendment moved the high-risk rules to 2027 and 2028. Dates when rules apply do not measure enforcement. Source: [European Commission, the AI Act and its timeline](https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai) · [Regulation (EU) 2026/1744, which moved the high-risk dates](https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=OJ%3AL_202601744).
10. **J · Data-centre electricity in 2030** (Forecasts, not facts)
    The IEA’s scenarios for 2030, all data centres: Headwinds 833 TWh, High Efficiency 868, Base 945, Lift-Off 1,008. A spread of named scenarios, not a confidence interval; the IEA assigns them no probabilities. Source: [IEA, Key Questions on Energy and AI (2026), Table A.1](https://iea.blob.core.windows.net/assets/3179f7f8-01f6-4dd6-bffa-c9f7b73f1dc9/KeyQuestionsonEnergyandAI.pdf).

## Sources

- [Pew Research Center, Americans and AI 2026 (17 June 2026)](https://www.pewresearch.org/internet/2026/06/17/americans-and-ai-2026-chatbots-smart-devices-and-views-on-impact/)
- [OpenAI DevDay, 29 September 2026, as reported live](https://simonwillison.net/2026/Sep/29/openai-devday-2026-live-blog/)
- [Rein et al., GPQA, arXiv 2311.12022 (2023), Table 5](https://arxiv.org/abs/2311.12022)
- [Kwa et al. (METR), Measuring AI Ability to Complete Long Tasks, arXiv 2503.14499 v4, section 3.2](https://arxiv.org/abs/2503.14499)
- [Dell’Acqua et al., Organization Science (2026), Table 4](https://doi.org/10.1287/orsc.2025.21838)
- [Becker et al. (METR), arXiv 2507.09089 (2025)](https://arxiv.org/abs/2507.09089)
- [METR, We are Changing our Developer Productivity Experiment Design (24 February 2026)](https://evals.alignment.org/blog/2026-02-24-uplift-update/)
- [IEA, Key Questions on Energy and AI (2026), Table A.1](https://iea.blob.core.windows.net/assets/3179f7f8-01f6-4dd6-bffa-c9f7b73f1dc9/KeyQuestionsonEnergyandAI.pdf)
- [Microsoft, Annual Report 2026, cash flows statements (p. 41)](https://www.sec.gov/Archives/edgar/data/0000789019/000119312526427876/d62902dars.pdf)
- [Dreksler et al., Subjective Experience in AI Systems, arXiv 2506.11945 (2025), section 2.1](https://arxiv.org/abs/2506.11945)
- [European Commission, the AI Act and its timeline](https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai)
- [Regulation (EU) 2026/1744, which moved the high-risk dates](https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=OJ%3AL_202601744)

Prepared by AI systems at the originator’s request, and checked number by number against the sources above on [CZ 1434](https://civilizationzero.ai/calendar?d=CZ%2B1434). An error found is an objection: [objections and errata](https://civilizationzero.ai/objections). The longer view, from the first humans to CZ 0: [the infographic](https://civilizationzero.ai/civilization-zero-infographic). What it all rests on: [the Invariant Core](https://civilizationzero.ai/core).