Continued skepticism from academics and journalists
Tracker summary: However, although this word has entered discourse, most people—academics, politicians, government employees, and the media—continue to underestimate the pace of progress. Skeptics are still pointing out the things Agent-3-mini can’t do.
At a glance
- Assessment: Not Yet Testable
- Confidence in assessment: 85%
- Predicted timing: May to July 2027
- Primary source: AI 2027 archived scenario: May 2027: National Security, pp. 13, 14; July 2027: The Cheap Remote Worker, p. 16; Late 2026: AI Takes Some Jobs, p. 8
What AI 2027 Predicted
However, although this word has entered discourse, most people—academics, politicians, government employees, and the media—continue to underestimate the pace of progress.
Source: May 2027: National Security, pp. 13-14 (the sentence begins on p. 13 and continues on p. 14).
Skeptics are still pointing out the things Agent-3-mini can’t do.
Source: July 2027: The Cheap Remote Worker, p. 16.
Earlier context, Late 2026: AI Takes Some Jobs, p. 8:
The mainstream narrative around AI has changed from “maybe the hype will blow over” to “guess this is the next big thing,” but people disagree about how big.
Tracker summary: The tracked passage concerns underestimating the pace of progress and continuing skepticism in May to July 2027, not a claim that nobody recognized major AI change in 2026.
Scenario source: archived version and page-number reference: May 2027: National Security, pp. 13, 14; July 2027: The Cheap Remote Worker, p. 16; Late 2026: AI Takes Some Jobs, p. 8.
How We Track This
Tracker operational test: Collect prospective, population-specific forecasts of concrete capability milestones and compare them with actual achievement dates in May to July 2027. Before observing the outcomes, define milestone descriptions, completion criteria, the populations sampled and the scoring method. Preserve forecast dates, sample sizes, recruitment, weighting and uncertainty; do not select only milestones that later arrive earlier than expected.
For each milestone and population, measure the share whose forecast placed achievement later than the independently verified date. The majority-underestimation claim concerns people, not the number of skeptical articles. Report results separately for the public, academics, politicians, government employees and media respondents where comparable samples exist. Explain which populations each result can represent and where evidence is missing. Report uncertainty and mixed results rather than requiring every named group to pass, imposing a two-milestone minimum or making a universal confidence-bound cutoff part of the prediction.
Contemporaneous observations of continued skepticism in July 2027 can address that part of the narrative, but pointing out real limitations does not itself establish an incorrect forecast of progress.
Current Evidence
Not Yet Testable (85%): May to July 2027 is still ahead. No comparable prospective forecasts matched to achievements in that window are supplied here. The evidence below concerns earlier attitudes or labor outcomes, not whether the named populations underestimate the pace of progress during the predicted period. Confidence concerns this classification, not an 85% chance that skepticism will persist.
March 2025: AAAI Presidential Panel survey: The primary report, Introduction and the Community Opinion subsections of Artificial General Intelligence and AI Perception vs. Reality describes 475 respondents overall. Of those answering the AGI questions, 76% regarded scaling current approaches to AGI as unlikely or very unlikely to succeed. Of those answering the perception questions, 79% disagreed that current perceptions match AI research and development; only 36% of all survey respondents chose that theme. These are dated views from a self-selected research-community sample, not representative estimates for all academics, journalists or policymakers, and not scored forecasts demonstrating underestimation during May-July 2027.
March 5, 2026: labor-market results, not a skepticism survey: Anthropic’s primary study reports no systematic unemployment increase among highly exposed workers and tentative evidence of slower hiring for younger workers in exposed occupations. These measured labor outcomes cannot establish researchers’ or journalists’ beliefs about future capability.
Counterevidence & Limitations
Caution, disagreement and observations about a model’s current limits do not by themselves demonstrate incorrect forecasts. Conversely, isolated alarmist or skeptical articles cannot establish what most people believe. Compare predictions recorded before a capability outcome with that outcome, rather than classifying tone alone.
The observation window is May to July 2027. Earlier attitudes cannot resolve that outcome. Nonrepresentative samples, unmatched milestone definitions and uncertain achievement dates limit what can be inferred about a population; missing observations are not evidence that its forecasts were accurate.
What Would Change Our Assessment
- Confirm under the forecast test: Prospective forecasts matched to verified achievement dates in May to July 2027 support majority underestimation in the populations they represent. State the population coverage, milestone-specific findings and uncertainty; do not generalize a narrow sample to everyone.
- Remain Not Yet Testable: Comparable forecasts or milestone outcomes are unavailable, including after the window if the only problem is missing measurement.
- Contradict the forecast claim: Adequate comparable observations show that the populations assessed did not generally place achievement later than it occurred. Mixed or imprecise results remain unresolved, rather than isolated skeptical comments counting as confirmation.
Update History
| Date | Update |
|---|---|
| 2026-11-09 | 2026-11-09 · Correction: status Confirmed (95%) → Not Yet Testable (85%). The May to July 2027 outcome requires prospective, population-specific milestone forecasts matched to actual achievement dates. Earlier attitudes cannot establish underestimation. Removed the added all-group, two-milestone and universal lower-95%-bound pass conditions; uncertainty will be reported rather than converted into those extra requirements. |
| 2026-11-07 | Correction: The scenario concerns underestimation in May to July 2027. The AAAI survey records earlier attitudes; labor-market outcomes do not measure those attitudes or prove underestimation. April 2025 criticism is reception of the scenario, not population evidence. Added the prospective population-and-milestone test and corrected the scenario locators. |
| 2026-03-13 | Institutional skepticism persists alongside rapid capability gains. Pattern matches prediction — skeptics update slowly relative to capability progress. 2026-11-07 · Correction: The March 2025 AAAI survey describes earlier attitudes, not majority underestimation of outcomes in May to July 2027. |
| 2025-12 | Despite GPT-5 launch and coding agent proliferation, prominent academics and journalists continue questioning transformative AI claims. |
| 2025-04 | LessWrong “Thoughts on AI 2027” (April 9) includes critical analysis; separate post argues forecast is “implausible” citing FutureSearch data showing 20M actual paid subscribers vs 27M predicted. Early community engagement highlights specific calibration misses. 2026-11-07 · Classification correction: Reception of this particular scenario, not population evidence of institutional skepticism. See Max Harms, April 9, 2025 and Comments on AI 2027. The subscriber comparison above is the latter author’s criticism, not an independent measurement of subscriber numbers. |