---
id: FCST-SIGNALS-001
title: "Signals, Indicators and Inflection Points"
alternative_names:
  - "AI Leading Indicators"
  - "Tripwires and Early-Warning Indicators for AI Progress"
category: "The Future → Indicators"
volume: XII
chapter: 67
entry_type: "Technical Article"
level: 4
status: "Draft"
last_reviewed: 2026-10-15
version: "0.1.0"
prerequisites:
  - "FCST-METHOD-001"
related_entries:
  - "AGI-TIME-001"
  - "ASI-EXP-001"
  - "ECON-COMP-001"
  - "CAP-BENCH-001"
  - "FCST-NEAR-001"
  - "FCST-SCEN-001"
  - "FCST-UNCERT-001"
  - "FCST-REG-001"
---

# Signals, Indicators and Inflection Points

## 1. Concise Definition
A **signal** (or indicator) is a measurable quantity chosen in advance because its movement should change a forecast; an **inflection point** is a point at which the rate or direction of a tracked quantity changes materially, such as a transition from exponential to saturating growth. This article sets out how indicators for AI progress are chosen, what each one can and cannot show, and how pre-committed thresholds keep updating honest.

## 2. Overview & Significance
Forecasts and scenarios ([FCST-SCEN-001](./FCST-SCEN-001_medium_and_long_term_scenarios.md)) are only useful if observations can update them. Without pre-specified indicators, observers tend to interpret any new development as confirming their existing view. The discipline described here is to state, before the data arrive, which readings would move which forecast and by roughly how much.

## 3. Historical Development
Early-warning indicator practice comes from economics, public health, and intelligence analysis. In AI, indicators have mostly been published capability benchmarks and scaling datasets. Epoch AI's compute-trend analysis (Sevilla et al., 2022) turned training compute into a tracked series; METR's time-horizon measure (Kwa et al., 2025) added a capability series expressed in human task duration, intended to be comparable across model generations. Dynamic forecasting benchmarks such as ForecastBench (Karger et al., 2024) add a series measuring a model's ability to forecast. Expert surveys (Grace et al., 2024) provide a periodic belief indicator.

## 4. Technical / Conceptual Explanation

### 4.1 Leading, coincident and lagging indicators
- *Leading:* resource inputs (training compute, power procurement, chip supply, capital). They change before capability and are the focus of projections such as Sevilla et al. (2024).
- *Coincident:* measured capability (benchmark and task-length series).
- *Lagging:* economic and social effects (productivity, employment, incidents), and the stated beliefs of experts.

### 4.2 Detecting a trend change
For a series $y_t$ with log-linear growth $\ln y_t=\alpha+\beta t+\varepsilon_t$, a change in slope can be tested by comparing the fit of one slope against two slopes with a breakpoint $\tau$, for example with a likelihood-ratio or information-criterion comparison. Two cautions apply. First, with few observations (frontier models appear a few times per year), detection lags the true change by many months. Second, the observed series is a *measurement* of capability, and a change in the measurement (task suite, scoring, evaluation tooling) can mimic a change in the trend. The METR paper itself reports that the trend "may have accelerated in 2024" while noting a seven-month average doubling since 2019 (Kwa et al., 2025), which illustrates the difficulty of distinguishing a slope change from noise on a short window.

### 4.3 Saturation
Benchmarks saturate: once the best systems score near the maximum, the series stops discriminating. A flat benchmark line after saturation is not evidence that capability has plateaued. Indicators should therefore be retired and replaced on a stated schedule, with overlap periods to allow re-calibration (see [CAP-BENCH-001](../vol04_capabilities/CAP-BENCH-001_ai_evaluation.md) on contamination and measurement problems).

### 4.4 Pre-committed thresholds
For each indicator, a record states: the series and its publisher; a baseline value and date; a threshold; and the forecast or scenario it informs. Because the baseline for several series was not verified in this draft, baselines are captured at the first review of each record.

## 5. Types, Components & Classifications

| Indicator | Class | Publisher/source | What a change would suggest | Main caveat |
| :--- | :--- | :--- | :--- | :--- |
| Largest documented training compute | Leading | Epoch AI database (Sevilla et al., 2022, 2024) | Resource scaling pace | Closed labs undisclosed; estimates |
| Power and data-centre procurement for training | Leading | Operator and utility announcements; Epoch analysis | Whether power binds first | Announcements are not delivered capacity |
| 50% task time horizon | Coincident | METR (Kwa et al., 2025) | Autonomy on long software tasks | Software tasks only; suite changes |
| Dynamic forecasting benchmark standing | Coincident | ForecastBench (Karger et al., 2024) | Model judgment quality | Benchmark design may change |
| Share of AI research performed by AI systems | Coincident (hypothesised) | No established standard series | Relevance of Scenario C | Not currently measured in a standard way: VERIFICATION NEEDED |
| Aggregate expert survey dates | Lagging | AI Impacts survey series (Grace et al., 2024) | Shifts in researcher beliefs | Beliefs follow events |
| Labour-market and productivity statistics | Lagging | National statistical agencies | Realised economic effect | Slow, noisy, hard to attribute |
| Regulatory timetables | Institutional | Official journals | Constraint on deployment (Scenario D) | Dates change; check primary text |

## 6. Applications & Concrete Examples
- *Update rule for the time-horizon record.* In `FCST-HORIZON-001` ([FCST-NEAR-001](./FCST-NEAR-001_near_term_outlook_0_3_years.md)), if METR's published horizon rises by more than a factor of two within 12 months, the record resolves true and the Scenario B/C boundary is revisited; if it rises by less than a factor of two over 18 months, confidence in continued fitted trends is reduced.
- *Compute indicator.* If a documented run above $2\times10^{29}$ FLOP appears before 2029-10-15, `FCST-COMPUTE-001` fails and the Epoch 2030 feasibility projection (Sevilla et al., 2024) is treated as having arrived early.
- *Survey indicator.* A shift in the 50% date of 13 years between 2022 and 2023 (Grace et al., 2024) shows how much a belief indicator can move within a year; it should be read with the capability events of that year, not as an independent signal.

## 7. Advantages, Limitations & Unresolved Questions
- *Goodhart effects.* Once an indicator is widely used as a target, laboratories may optimise for it, weakening its meaning.
- *Selection of indicators after the fact* defeats the purpose; this article's list is fixed at the stated date and additions are logged.
- *Independence.* Several indicators share a cause (for example, the same compute growth), so agreement among them is weaker evidence than it looks.
- *Unresolved:* a validated standard measure of AI contribution to AI research; indicators for non-software capability; indicators of deployed (as opposed to demonstrated) capability.

## 8. Current State of Knowledge
As of 2026-10-15, the compute series, the METR time-horizon series, a dynamic forecasting benchmark, and a recurring expert survey exist as public series with the properties described above, per the cited sources (dated 2022-2025). The state of each series today was not re-measured for this draft; use `MODEL-001`, `FACT-001` and `TIME-001` for current values. No standard indicator of AI-driven acceleration of AI research was identified. [VERIFICATION NEEDED: a literature search for such measures was not completed.]

## 9. Future Trajectory & Strategic Outlook
Indicator sets will likely be extended (more domains, more agentic measures) and some will be retired at saturation. The following record applies the methods above to one leading indicator.

### Forecast record

| Field | Content |
| :--- | :--- |
| **Forecast ID** | `FCST-COMPGROWTH-001` |
| **Forecast Statement** | The largest training compute in Epoch AI's database will grow by at least a factor of two between 2026-10-15 and 2028-10-15. |
| **Target Event** | Growth in the maximum documented training compute |
| **Operational Definition** | Let $M_0$ be the maximum training-compute value in Epoch AI's published database on 2026-10-15 (recorded at first review) and $M_2$ the maximum on 2028-10-15. Resolves true if $M_2\ge 2M_0$. If the database is discontinued, resolves "not tested". |
| **Time Horizon** | 2028-10-15 |
| **Assumptions** | Continued investment in frontier training; laboratories' disclosure or Epoch's estimates continue to cover the largest runs; the deep-learning-era doubling of about 6 months for notable systems (Sevilla et al., 2022) is not a guide to the single largest run, so a lower bar (factor of two over two years) is used. |
| **Confidence Level** | High (about 80-90%; editorial judgment) |
| **Alternative Scenarios** | Scenario A: constraints bind and the largest runs plateau; or runs continue but go undisclosed, so the database does not record them. |
| **Review Schedule** | 2027-01-15 (baseline), 2027-10-15, 2028-10-15 (resolution) |
| **Status** | `Active` |

## 10. Related Entries & Knowledge Graph
- **Foundational Concepts**:
  - [[FCST-METHOD-001] Forecasting AI Progress: Methods and Calibration](./FCST-METHOD-001_forecasting_ai_progress_methods_and_calibration.md): Scoring and record format.
  - [[CAP-BENCH-001] AI Evaluation, Benchmarking & Contamination](../vol04_capabilities/CAP-BENCH-001_ai_evaluation.md): Why capability indicators saturate and mislead.
- **Related Methods & Technologies**:
  - [[ECON-COMP-001] The Economics of Compute & Scaling](../vol09_economy/ECON-COMP-001_compute_economics.md): Leading resource indicators.
  - [[AGI-TIME-001] AGI Forecasting & Timeline Methodology](../vol05_agi/AGI-TIME-001_agi_forecasting.md): Timelines the indicators update.
- **Advanced Implications & Future Trajectories**:
  - [[FCST-SCEN-001] Medium- and Long-Term Scenarios](./FCST-SCEN-001_medium_and_long_term_scenarios.md): Scenarios the indicators discriminate.
  - [[FCST-NEAR-001] Near-Term Outlook (0-3 years)](./FCST-NEAR-001_near_term_outlook_0_3_years.md): Dated records using these indicators.
  - [[FCST-UNCERT-001] The Uncertainties That Matter Most](./FCST-UNCERT-001_the_uncertainties_that_matter_most.md): Uncertainties indicators should target.
  - [[ASI-EXP-001] The Intelligence Explosion Hypothesis](../vol06_asi/ASI-EXP-001_intelligence_explosion.md): The mechanism whose indicator is not yet standardised.
  - [[FCST-REG-001] Forecast Register](../reference/FCST-REG-001_forecast_register.md): Record status.

## 11. References & Foundational Literature
1. Sevilla, J., Heim, L., Ho, A., Besiroglu, T., Hobbhahn, M., & Villalobos, P. (2022). *Compute Trends Across Three Eras of Machine Learning*. arXiv:2202.05924; IJCNN 2022. https://arxiv.org/abs/2202.05924. [Training-compute series with an approximately 6-month doubling in the deep-learning era.]
2. Sevilla, J., Besiroglu, T., Cottier, B., You, J., Roldán, E., Villalobos, P., & Erdil, E. (2024). *Can AI scaling continue through 2030?* Epoch AI, 2024-08-20. https://epoch.ai/blog/can-ai-scaling-continue-through-2030. [Resource-constraint projections; power identified as the first binding constraint.]
3. Kwa, T., West, B., Becker, J., et al. (2025). *Measuring AI Ability to Complete Long Software Tasks*. arXiv:2503.14499. https://arxiv.org/abs/2503.14499. [Time-horizon indicator; seven-month doubling; possible 2024 acceleration.]
4. Karger, E., Bastani, H., Yueh-Han, C., Jacobs, Z., Halawi, D., Zhang, F., & Tetlock, P. E. (2024). *ForecastBench: A Dynamic Benchmark of AI Forecasting Capabilities*. arXiv:2409.19839 (v5, 2025-02-28). https://arxiv.org/abs/2409.19839. [Dynamic benchmark of 1,000 questions; experts outperformed the top LLM on a 200-question subset.]
5. Grace, K., Stewart, H., Sandkühler, J. F., Thomas, S., Weinstein-Raun, B., Brauner, J., & Korzekwa, R. C. (2024). *Thousands of AI Authors on the Future of AI*. arXiv:2401.02843. https://arxiv.org/abs/2401.02843. [Belief indicator; 13-year shift in the 50% date between survey waves.]

## 12. Evidential Classification Matrix
| Section / Claim | Evidence Category | Supporting Source | Confidence / Status |
| :--- | :--- | :--- | :--- |
| Compute series and doubling | Well Supported | Sevilla et al. (2022) | High for period studied |
| Power as first binding constraint | Forecast | Sevilla et al. (2024) | Report-stated |
| Time-horizon series | Preliminary | Kwa et al. (2025) | Moderate |
| Dynamic forecasting benchmark | Preliminary | Karger et al. (2024) | Moderate |
| Belief shift of 13 years | Well Supported (survey fact) | Grace et al. (2024) | High |
| Indicator classification and trend-break methods | Editorial method (no specific citation) | -- | Moderate |
| `FCST-COMPGROWTH-001` | Forecast | Sevilla et al. (2022, 2024) | 80-90%, editorial |

## 13. Revision History
- **v0.1.0** (2026-10-15): Initial draft. Citations checked against arXiv/Crossref/publisher pages where marked; not independently fact-checked.