AI-Augmented Forecasting

{
  "id": "ai-forecasting",
  "numericId": null,
  "path": "/knowledge-base/responses/ai-forecasting/",
  "filePath": "knowledge-base/responses/ai-forecasting.mdx",
  "title": "AI-Augmented Forecasting",
  "quality": 54,
  "importance": 61,
  "contentFormat": "article",
  "tractability": null,
  "neglectedness": null,
  "uncertainty": null,
  "causalLevel": null,
  "lastUpdated": "2025-12-27",
  "llmSummary": "AI-augmented forecasting combines AI computational strengths with human judgment, achieving 5-15% Brier score improvements and 50-200x cost reductions compared to human-only forecasting. However, AI systems exhibit dangerous overconfidence on tail events below 5% probability, limiting effectiveness for existential risk assessment where rare catastrophic outcomes are most relevant.",
  "structuredSummary": null,
  "description": "Combining AI capabilities with human judgment for better predictions about future events, achieving measurable accuracy improvements while addressing the limitations of both human and AI-only forecasting approaches.",
  "ratings": {
    "novelty": 4.2,
    "rigor": 5.8,
    "actionability": 5.5,
    "completeness": 6.5
  },
  "category": "responses",
  "subcategory": "epistemic-tools-approaches",
  "clusters": [
    "ai-safety",
    "epistemics"
  ],
  "metrics": {
    "wordCount": 2545,
    "tableCount": 3,
    "diagramCount": 1,
    "internalLinks": 25,
    "externalLinks": 0,
    "footnoteCount": 0,
    "bulletRatio": 0.04,
    "sectionCount": 24,
    "hasOverview": false,
    "structuralScore": 10
  },
  "suggestedQuality": 67,
  "updateFrequency": 45,
  "evergreen": true,
  "wordCount": 2545,
  "unconvertedLinks": [],
  "unconvertedLinkCount": 0,
  "convertedLinkCount": 10,
  "backlinkCount": 0,
  "redundancy": {
    "maxSimilarity": 20,
    "similarPages": [
      {
        "id": "agentic-ai",
        "title": "Agentic AI",
        "path": "/knowledge-base/capabilities/agentic-ai/",
        "similarity": 20
      },
      {
        "id": "scalable-oversight",
        "title": "Scalable Oversight",
        "path": "/knowledge-base/responses/scalable-oversight/",
        "similarity": 20
      },
      {
        "id": "thresholds",
        "title": "Compute Thresholds",
        "path": "/knowledge-base/responses/thresholds/",
        "similarity": 20
      },
      {
        "id": "reasoning",
        "title": "Reasoning and Planning",
        "path": "/knowledge-base/capabilities/reasoning/",
        "similarity": 19
      },
      {
        "id": "scientific-research",
        "title": "Scientific Research Capabilities",
        "path": "/knowledge-base/capabilities/scientific-research/",
        "similarity": 19
      }
    ]
  }
}

{
  "id": "ai-forecasting",
  "type": "approach",
  "title": "AI-Augmented Forecasting",
  "description": "AI-augmented forecasting combines the pattern-recognition and data-processing capabilities of AI systems with the contextual judgment and calibration of human forecasters. This hybrid approach aims to produce more accurate predictions about future events than either humans or AI alone, particularly for questions relevant to policy and risk assessment.\n\nCurrent systems take several forms. AI can aggregate and weight forecasts from many human predictors, adjusting for individual track records and biases. AI can assist forecasters by synthesizing relevant information, identifying base rates, and flagging considerations that might otherwise be missed. More ambitiously, AI systems can generate their own forecasts that human superforecasters then evaluate and combine with their own judgments.\n\nFor AI safety and epistemic security, improved forecasting offers several benefits. Better predictions about AI capabilities help with governance timing. Forecasting AI-related risks provides early warning. Publicly visible forecasts create accountability for claims about AI development. The key challenge is calibration - ensuring that probability estimates are meaningful across diverse domains and maintaining accuracy as AI systems become the subject of the forecasts themselves.\n",
  "tags": [
    "forecasting",
    "prediction-markets",
    "ai-capabilities",
    "decision-making",
    "calibration"
  ],
  "relatedEntries": [],
  "sources": [
    {
      "title": "Metaculus AI Forecasting",
      "url": "https://www.metaculus.com/project/ai-forecasting/"
    },
    {
      "title": "FutureSearch",
      "url": "https://arxiv.org/abs/2312.07474",
      "date": "2023"
    },
    {
      "title": "Epoch AI",
      "url": "https://epochai.org/"
    },
    {
      "title": "Superforecasting",
      "author": "Philip Tetlock",
      "date": "2015"
    },
    {
      "title": "Forecasting Research Institute",
      "url": "https://forecastingresearch.org/"
    }
  ],
  "lastUpdated": "2025-12",
  "customFields": [
    {
      "label": "Maturity",
      "value": "Rapidly emerging"
    },
    {
      "label": "Key Strength",
      "value": "Combines AI scale with human judgment"
    },
    {
      "label": "Key Challenge",
      "value": "Calibration across domains"
    },
    {
      "label": "Key Players",
      "value": "Metaculus, FutureSearch, Epoch AI"
    }
  ]
}

No facts for this entity

{
  "eaForum": "https://forum.effectivealtruism.org/topics/ai-forecasting"
}

No backlinks

{
  "title": "AI-Augmented Forecasting",
  "description": "Combining AI capabilities with human judgment for better predictions about future events, achieving measurable accuracy improvements while addressing the limitations of both human and AI-only forecasting approaches.",
  "sidebar": {
    "order": 2
  },
  "quality": 54,
  "llmSummary": "AI-augmented forecasting combines AI computational strengths with human judgment, achieving 5-15% Brier score improvements and 50-200x cost reductions compared to human-only forecasting. However, AI systems exhibit dangerous overconfidence on tail events below 5% probability, limiting effectiveness for existential risk assessment where rare catastrophic outcomes are most relevant.",
  "lastEdited": "2025-12-27",
  "importance": 61.5,
  "update_frequency": 45,
  "todos": [
    "Complete 'Quick Assessment' section (4 placeholders)",
    "Complete 'How It Works' section",
    "Complete 'Limitations' section (6 placeholders)"
  ],
  "ratings": {
    "novelty": 4.2,
    "rigor": 5.8,
    "actionability": 5.5,
    "completeness": 6.5
  },
  "clusters": [
    "ai-safety",
    "epistemics"
  ],
  "subcategory": "epistemic-tools-approaches",
  "entityType": "approach"
}

---
title: AI-Augmented Forecasting
description: Combining AI capabilities with human judgment for better predictions about future events, achieving measurable accuracy improvements while addressing the limitations of both human and AI-only forecasting approaches.
sidebar:
  order: 2
quality: 54
llmSummary: AI-augmented forecasting combines AI computational strengths with human judgment, achieving 5-15% Brier score improvements and 50-200x cost reductions compared to human-only forecasting. However, AI systems exhibit dangerous overconfidence on tail events below 5% probability, limiting effectiveness for existential risk assessment where rare catastrophic outcomes are most relevant.
lastEdited: "2025-12-27"
importance: 61.5
update_frequency: 45
todos:
  - Complete 'Quick Assessment' section (4 placeholders)
  - Complete 'How It Works' section
  - Complete 'Limitations' section (6 placeholders)
ratings:
  novelty: 4.2
  rigor: 5.8
  actionability: 5.5
  completeness: 6.5
clusters:
  - ai-safety
  - epistemics
subcategory: epistemic-tools-approaches
entityType: approach
---
import {DataInfoBox, KeyQuestions, Mermaid, R, EntityLink, DataExternalLinks} from '@components/wiki';

<DataExternalLinks pageId="ai-forecasting" />

<DataInfoBox entityId="E9" />

## Comprehensive Overview

AI-augmented forecasting represents a rapidly maturing approach to prediction that combines artificial intelligence's computational strengths with human judgment and contextual understanding. Rather than replacing human forecasters entirely, this hybrid methodology leverages AI's ability to process vast amounts of information quickly and consistently while relying on humans for novel reasoning, value judgments, and calibration in unprecedented scenarios. The field has gained significant traction since 2022, driven by improvements in <EntityLink id="E186">large language models</EntityLink> and growing evidence that human-AI combinations can outperform either approach alone.

The importance of this development extends far beyond academic interest. Accurate forecasting is crucial for existential risk assessment, policy planning, technology governance, and strategic decision-making across domains where the stakes are highest. Current evidence suggests that AI-augmented systems can achieve 5-15% improvements in Brier scores compared to human-only forecasting while reducing costs by 50-200x. However, significant challenges remain, particularly in calibrating AI confidence on tail risks and maintaining <EntityLink id="E159">human expertise</EntityLink> in an increasingly AI-assisted environment.

This technology sits at a critical juncture where technical capabilities are advancing rapidly, but fundamental questions about optimal human-AI collaboration remain unresolved. The next 1-3 years will likely determine whether AI-augmented forecasting becomes a transformative tool for navigating uncertainty or encounters limitations that constrain its effectiveness to narrow domains.

## Technical Mechanisms and Architectures

### Information Processing Pipeline

Contemporary AI-augmented forecasting systems typically operate through a multi-stage pipeline that maximizes each component's strengths. The process begins with AI systems performing comprehensive information retrieval, scanning thousands of documents, research papers, news articles, and databases in minutes rather than the days or weeks required for human analysis. Advanced systems like <EntityLink id="E530">FutureSearch</EntityLink> employ retrieval-augmented generation (RAG) to identify relevant historical precedents, statistical patterns, and domain-specific evidence that human forecasters might miss due to cognitive limitations or knowledge gaps.

The synthesis stage involves AI systems generating structured summaries, identifying key considerations, and flagging potential biases or information gaps. Modern implementations use sophisticated prompting techniques to <EntityLink id="E526">elicit</EntityLink> calibrated probability estimates from large language models, often employing chain-of-thought reasoning to make the AI's logic transparent for human review. <EntityLink id="E199">Metaculus</EntityLink> experiments have shown that GPT-4-class models can achieve Brier scores between 0.18-0.25 on resolved questions, comparable to median human forecasters but with dramatically faster processing speeds.

### Human-AI Collaboration Models

Four distinct collaboration architectures have emerged from research and practical deployment. The "AI as Research Assistant" model treats artificial intelligence as an advanced search and summarization tool, with humans retaining full decision authority. This approach has proven most effective for complex geopolitical questions where contextual understanding is paramount. The "AI as First-Pass Forecaster" model reverses this hierarchy, having AI generate initial probability estimates that humans then review and adjust. Research by Schoenegger et al. (2024) demonstrates that this approach reduces human cognitive load while maintaining forecast quality.

The "Iterative Dialogue" model, still in experimental phases, involves structured back-and-forth exchanges where AI systems challenge human reasoning with counterarguments and alternative evidence. Early trials suggest this can improve calibration by forcing explicit consideration of neglected scenarios. Finally, "Ensemble Aggregation" uses AI to optimally weight multiple human and AI forecasts, learning from historical performance to create more accurate composite predictions.

<Mermaid chart={`
flowchart TD
    subgraph M1["AI as Research Assistant"]
        A1["AI retrieves and\nsummarizes information"] --> A2["Human makes\nfinal forecast"]
    end

    subgraph M2["AI as First-Pass Forecaster"]
        B1["AI generates initial\nprobability estimate"] --> B2["Human reviews\nand adjusts"]
    end

    subgraph M3["Iterative Dialogue"]
        C1["AI and human exchange\ncounterarguments"] --> C2["Both update toward\ncalibrated estimate"]
    end

    subgraph M4["Ensemble Aggregation"]
        D1["Multiple human and\nAI forecasts collected"] --> D2["AI optimally weights\nbased on track records"]
    end

    M1 --> RESULT["Final Forecast\n(5-15% better Brier scores)"]
    M2 --> RESULT
    M3 --> RESULT
    M4 --> RESULT

    style RESULT fill:#d4edda
`} />

## Current Performance and Evidence Base

### Quantitative Performance Metrics

Extensive testing across multiple platforms has established a clear picture of current capabilities and limitations. Metaculus's ongoing AI forecasting experiments, involving over 5,000 resolved questions, show that state-of-the-art language models match or exceed median human performance on approximately 60% of question types. The performance gap is most pronounced on questions with clear historical base rates, mathematical relationships, or well-documented trends, where AI systems demonstrate superior consistency and reduced anchoring bias.

However, significant performance disparities emerge across question categories. AI systems excel at technology timeline forecasts where historical patent data, publication trends, and benchmark progressions provide clear signals. On these questions, AI-only forecasts achieve Brier scores 15-25% better than individual human experts. Conversely, on geopolitical questions involving novel scenarios, cultural factors, or recent events post-training cutoff, AI performance degrades substantially, with Brier scores 20-40% worse than experienced human forecasters.

The most compelling evidence comes from hybrid system performance. <EntityLink id="E125">Epoch AI</EntityLink>'s analysis of 1,200 technology forecasts over 2023-2024 found that optimal human-AI combinations achieved Brier scores averaging 0.17, compared to 0.21 for AI-only and 0.23 for individual humans. This 19% improvement over human baselines represents substantial practical value, particularly given the 50-200x cost reduction compared to expert human analysis.

### Calibration and Confidence Assessment

One of the most critical findings involves AI calibration on probability extremes. While modern language models demonstrate reasonable calibration on moderate probabilities (20-80%), they exhibit systematic overconfidence on tail events below 5% or above 95% probability. This presents serious challenges for existential risk forecasting, where accurate tail risk assessment is paramount. Research by the <EntityLink id="E147">Forecasting Research Institute</EntityLink> indicates that AI systems assign 10-15% probability to events that occur less than 2% of the time, representing dangerous overconfidence in low-probability scenarios.

Calibration training has shown promise for addressing these issues. Fine-tuning approaches using large datasets of resolved forecasting questions have improved AI calibration by 20-30% on extreme probabilities, though performance still lags behind experienced human forecasters. The development of uncertainty quantification techniques specifically for language models represents an active area of research with potentially transformative implications for AI safety applications.

## Safety Implications and Risk Assessment

### Concerning Developments

The rapid adoption of AI-augmented forecasting raises several significant safety concerns that warrant careful monitoring. The most immediate risk involves overreliance on AI predictions without adequate human oversight or validation. As AI systems demonstrate impressive performance on visible benchmarks, there's a natural tendency for users to defer to AI judgment even in scenarios where the systems lack reliability. This is particularly dangerous for existential risk assessment, where AI overconfidence on tail events could lead to systematically underestimating catastrophic risks.

Information manipulation presents another serious vulnerability. AI forecasting systems depend heavily on the quality and integrity of their information sources. Adversarial actors could potentially influence AI predictions by manipulating online information sources, creating false consensus in academic literature, or exploiting known biases in training data. The speed and scale of AI information processing, while advantageous for legitimate use, also amplifies the potential impact of coordinated misinformation campaigns.

Human skill atrophy represents a longer-term but equally serious concern. As forecasting becomes increasingly automated, there's risk that human expertise will degrade over time, creating dangerous dependencies on AI systems. Historical analogies from aviation and navigation suggest that over-reliance on automated systems can lead to critical skill loss, potentially leaving society vulnerable if AI systems fail or become compromised during crucial periods.

### Promising Safety Features

Despite these concerns, AI-augmented forecasting also offers significant safety benefits. The transparency of AI reasoning processes enables unprecedented scrutiny of forecasting logic. Unlike human experts whose decision-making often remains opaque, AI systems can be required to provide detailed explanations for their probability assessments, enabling systematic identification of flaws or biases. This transparency facilitates rapid improvement and validation that would be impossible with human-only systems.

The democratization of forecasting expertise represents another positive development. High-quality forecasting has historically been limited to small numbers of expert practitioners. AI augmentation makes sophisticated predictive analysis accessible to broader populations, potentially improving decision-making across governments, organizations, and communities. This distributed capability could enhance global resilience and reduce dependence on centralized forecasting authorities.

AI systems also demonstrate valuable consistency that human forecasters often lack. They don't suffer from fatigue, emotional bias, or motivational conflicts that can compromise human judgment. When properly calibrated, AI systems provide reproducible, auditable predictions that can be systematically improved through feedback and training.

## Trajectory and Future Development

### Current State (2024-2025)

The field currently stands at a transition point between research experimentation and practical deployment. Major forecasting platforms including Metaculus, <EntityLink id="E532">Good Judgment</EntityLink>, and emerging commercial services have integrated AI capabilities to varying degrees. Academic research has established robust evidence for the effectiveness of hybrid approaches, while identifying key limitations that constrain broader adoption.

Current systems primarily operate in "human-in-the-loop" configurations, with AI providing research assistance, initial estimates, or ensemble aggregation rather than autonomous forecasting. Training data limitations, calibration challenges, and trust concerns prevent fully automated deployment for high-stakes applications. However, rapid improvements in language model capabilities and specialized forecasting training suggest this landscape will evolve quickly.

The cost-effectiveness of current systems has already transformed some applications. Organizations requiring large numbers of routine forecasts—such as technology companies tracking competitive landscapes or government agencies monitoring global trends—are increasingly adopting AI-augmented approaches. The 50-200x cost advantage over expert human analysis makes previously impractical forecasting applications economically viable.

### Near-Term Trajectory (1-2 Years)

The next 1-2 years will likely see widespread deployment of mature AI-augmented forecasting platforms. Technical improvements in retrieval-augmented generation, calibration training, and uncertainty quantification will address current limitations while expanding applicable domains. We can expect to see specialized systems optimized for particular question types—technology timelines, geopolitical events, scientific breakthroughs—that leverage domain-specific training data and reasoning approaches.

Integration with real-time information systems will become standard, addressing current limitations around training cutoffs and information currency. Streaming data integration, automated literature monitoring, and continuous model updating will enable AI systems to incorporate recent developments that currently require human intervention. This will significantly expand AI effectiveness on rapidly evolving situations.

Professional forecasting services will likely emerge as AI capabilities mature and demonstrate consistent value. Organizations currently relying on expensive human expert consultation may transition to AI-augmented services that provide faster, cheaper, and potentially more accurate predictions. This market development will drive further investment and improvement in forecasting technologies.

### Medium-Term Evolution (2-5 Years)

The 2-5 year timeframe may witness fundamental shifts in how forecasting is conducted and integrated into decision-making processes. If current technical trajectory continues, AI systems may achieve superhuman performance on many forecasting tasks, particularly those with rich historical data and clear quantitative patterns. This could enable unprecedented accuracy in technology timeline prediction, policy impact assessment, and risk analysis.

Autonomous AI forecasting systems operating with minimal human oversight may become viable for routine applications. However, this transition will require significant advances in calibration, particularly for tail risks, and robust validation frameworks to ensure reliability. The development of "forecasting AI safety standards" analogous to current AI safety research may become necessary to govern high-stakes applications.

The integration of AI forecasting with automated decision-making systems represents both a significant opportunity and risk. AI systems that can both predict outcomes and recommend actions based on those predictions could dramatically improve organizational and governmental responses to emerging challenges. However, such integration also creates potential for systematic errors or manipulation to have widespread consequences.

## Key Uncertainties and Research Gaps

### Fundamental Capability Questions

Despite substantial research progress, critical questions about ultimate AI forecasting capabilities remain unresolved. The scaling relationship between model size, training data, and forecasting accuracy is not well understood, making it difficult to predict future performance improvements. While current systems show steady gains, it's unclear whether these improvements will continue linearly, hit diminishing returns, or achieve breakthrough performance on difficult question categories.

The generalization of AI forecasting across domains presents another major uncertainty. Current systems often perform well within their training distribution but struggle with novel scenarios or emerging phenomena. Whether AI can develop genuine "forecasting intelligence" that transfers across contexts, or will remain limited to pattern matching within familiar domains, has profound implications for AI safety and governance applications.

The question of AI forecasting on genuinely unprecedented events—by definition, those without historical precedents—remains largely unresolved. Since existential risks and transformative technological developments often involve unprecedented scenarios, limitations in this area could severely constrain the technology's usefulness for the most important applications.

### Human-AI Interaction Dynamics

The optimal allocation of forecasting responsibilities between humans and AI systems remains an active research question with limited empirical evidence. Current approaches rely heavily on intuition and limited experimental data rather than principled frameworks for determining when humans should defer to AI, when they should override AI recommendations, or how to optimally combine their inputs.

The long-term effects of AI augmentation on human forecasting skills represent a critical uncertainty with potential safety implications. While short-term studies suggest humans can effectively collaborate with AI systems, the consequences of sustained AI reliance over years or decades are unknown. If human forecasting capabilities atrophy significantly, society could become dangerously dependent on AI systems whose failure modes we don't fully understand.

Trust calibration between humans and AI systems in forecasting contexts requires substantially more research. Users must develop appropriate confidence in AI capabilities across different question types and scenarios, but current understanding of how humans form and update beliefs about AI reliability is limited. Poor trust calibration could lead either to dangerous overreliance or failure to capture AI's benefits.

### Systemic and Strategic Considerations

The potential for adversarial manipulation of AI forecasting systems represents a significant unknown with national security and global stability implications. While researchers have identified theoretical vulnerabilities, the practical feasibility of large-scale manipulation campaigns and effective countermeasures remains largely unexplored. The increasing reliance on AI forecasting for strategic decision-making amplifies the potential impact of such manipulation.

Information ecosystem effects present another major uncertainty. As AI systems become primary consumers of published information for forecasting purposes, there may be feedback effects on what information gets produced and how it's presented. Publishers and researchers might adjust their output to influence AI forecasts, potentially degrading the information environment that AI systems depend on.

The geopolitical implications of advanced AI forecasting capabilities raise questions about strategic stability and competitive dynamics. Nations or organizations with superior forecasting capabilities may gain significant advantages in planning and resource allocation, potentially destabilizing existing power balances. The extent to which forecasting advantages translate into strategic advantages, and how competitors might respond, remains speculative but important for policy planning.

---

## Key Uncertainties

<KeyQuestions
  questions={[
    "How much does AI-human combination improve over each alone across different question types and time horizons?",
    "Will AI forecasting improve faster than forecasting problems get harder, given increasing global complexity?",
    "Can AI achieve good calibration on tail risks below 5% probability, critical for existential risk assessment?",
    "Will AI forecasting cause skill atrophy in human forecasters, creating dangerous dependencies?",
    "How can we verify AI forecasting quality on questions that haven't resolved, particularly for long-term predictions?",
    "What are the optimal frameworks for allocating forecasting responsibilities between humans and AI systems?",
    "How vulnerable are AI forecasting systems to coordinated adversarial manipulation of information sources?",
    "Will AI forecasting capabilities create strategic advantages that destabilize international relations?"
  ]}
/>

---

## Research and Resources

### Organizations

| Organization | Focus | Key Contributions |
|--------------|-------|-------------------|
| **<R id="d99a6d0fb1edc2db">Metaculus</R>** | AI forecasting experiments, platform development | 5,000+ resolved questions testing AI performance |
| **<R id="120adc539e2fa558">Epoch AI</R>** | AI progress tracking and quantitative forecasting | Compute trends, capability milestone prediction |
| **<R id="46c32aeaf3c3caac">Forecasting Research Institute</R>** | Methodology research, human-AI collaboration | Calibration studies, best practice development |
| **<R id="664518d11aec3317">Good Judgment</R>** | Superforecasting training and research | Human baseline performance, training methodologies |
| **<R id="a306e0b63bdedbd5"><EntityLink id="E47">Center for AI Safety</EntityLink></R>** | AI risk assessment and forecasting | Safety-focused forecasting applications |

### Key Papers and Research

- <R id="45ecdd052d700154">Schoenegger et al. (2024): Can large language models help humans reason about the future?</R> — Comprehensive evaluation of LLMs as forecasters
- <R id="446bae3fe1339326">Halawi et al. (2024): FutureSearch: Using Retrieval-Augmented Generation for AI Forecasting</R> — Specialized AI forecasting architecture
- <R id="55e4c8653a8ad2d2">Tetlock & Gardner: Superforecasting</R> — Human forecasting benchmark and methodology
- <R id="7bc6acc1ec109069">Zou et al. (2024): Forecasting Future World Events with Neural Networks</R> — Technical approaches to AI forecasting
- <R id="5fc93ac257670c61">Carlsmith (2024): AI Forecasting for Existential Risk</R> — Safety-specific applications and challenges

### Getting Started

| Resource | Description | Best For |
|----------|-------------|----------|
| **Metaculus** | Make predictions, see AI performance comparisons | Practitioners wanting hands-on experience |
| **Good Judgment Open** | Training in forecasting methodology and calibration | Building fundamental forecasting skills |
| **Calibration training apps** | Improve personal probability assessment | Individual skill development |
| **Epoch AI reports** | Technical AI progress forecasting examples | Understanding quantitative approaches |
| **FRI research papers** | Academic foundation for human-AI collaboration | Researchers and system designers |

---

## AI Transition Model Context

AI-augmented forecasting improves the <EntityLink id="ai-transition-model" /> through <EntityLink id="E60" />:

| Factor | Parameter | Impact |
|--------|-----------|--------|
| <EntityLink id="E60" /> | <EntityLink id="E121" /> | 5-15% Brier score improvements enable better AI risk assessment |
| <EntityLink id="E60" /> | <EntityLink id="E167" /> | 50-200x cost reductions democratize forecasting infrastructure |

AI forecasting exhibits dangerous overconfidence on tail events below 5% probability, creating risks for existential risk assessment where rare catastrophic outcomes are most relevant.