Build a Trajectory Forecast
Every lesson in this module has been preparation for this one. You now have a toolkit: you understand forecasting methods and their pitfalls, the structure of the AGI debate, the difficulty of defining and measuring AGI, the characteristics of transformative AI scenarios, the nature of deep uncertainty, the range of possible consequences, and the principles of preparation and long-run thinking. This lesson puts the toolkit to use. Your task is to construct a complete, reasoned AI trajectory forecast — not a vague prediction, but a structured analysis with explicit assumptions, calibrated probability estimates, stated policy implications, and honest acknowledgment of what you do not know.
What a Good Forecast Looks Like
A well-constructed forecast is not simply a guess about the future. It is a structured argument with identifiable components that can be evaluated, challenged, and updated. Before beginning the main activity, it is worth reviewing what distinguishes a rigorous forecast from an informal prediction. A rigorous forecast names its target precisely: rather than 'AI will be transformative,' it specifies 'AI will be capable of performing 50% of current US knowledge-work tasks at human cost or below by 2038.' A precise target can be evaluated when the time comes and can be falsified by evidence along the way. A rigorous forecast states its probability as a distribution: it does not say 'this will happen' or 'this might happen.' It assigns probability: '45% chance by 2035, 70% chance by 2045.' The distribution captures both the point estimate and the uncertainty around it. A rigorous forecast makes its key assumptions explicit: every forecast rests on assumptions about compute availability, algorithmic progress rates, economic incentives, regulatory decisions, and what capabilities the target requires. Making these explicit allows others to disagree with the assumptions specifically rather than arguing past each other, and allows the forecaster to identify which assumptions, if wrong, would most change the estimate. A rigorous forecast identifies update triggers: it specifies what kinds of evidence would cause the forecaster to revise upward or downward. A forecast that cannot be updated by any evidence is not a forecast — it is a commitment to a conclusion regardless of the facts. A rigorous forecast connects to policy implications: given the probability distribution, what actions are warranted? What would be different if the probability were twice as high or half as high?
Research on forecasting accuracy, beginning with Philip Tetlock's work on superforecasters, shows that accurate forecasting is a learnable skill. The key practices are: breaking large questions into smaller ones, actively seeking disconfirming evidence, updating on new information without over-reacting, and tracking one's own record to improve calibration over time. These practices apply to AI forecasting just as to any other domain.
Component Analysis Before Building
Before constructing your forecast, a brief review of the analytical components you will use. Capability trajectory: what is the current trajectory of AI capability? What metrics are most reliable indicators of progress toward your target? What discontinuities — algorithmic or infrastructural — could significantly accelerate or decelerate the trajectory? Bottlenecks and enablers: what factors could prevent your target from being reached even if technical progress continues? Regulatory bottlenecks, data availability constraints, energy requirements, adoption barriers, and social resistance are all relevant. Conversely, what factors could accelerate progress beyond the baseline trajectory? Comparative cases: are there historical or contemporary comparisons that help anchor your estimate? How long did it take for previous transformative technologies to go from research demonstration to widespread economic impact? What features of those cases are similar to and different from the AI case? Scenario weighting: which of the scenarios from Lesson 4 (Broad Automation, Scientific Acceleration, Concentration, Recursive Improvement) does your forecast most closely follow? Does your forecast span multiple scenarios, or does it primarily track one? How do the scenarios' different requirements affect your probability estimates?
Match each forecast component to the question it answers.
Terms
Definitions
Drag terms onto their definitions, or click a term then click a definition to match.
A student's forecast states: 'AI will completely transform society within the next decade.' What is the most important analytical weakness of this forecast?
After completing a forecast, a student identifies that her probability estimate changes dramatically depending on whether she assumes algorithmic efficiency improvements continue at the historical rate or slow significantly. What should she do with this finding?
Construct Your AI Trajectory Forecast
- This is the capstone activity of Lesson 9. You will produce a complete, structured AI trajectory forecast using the framework developed throughout this module.
- PART 1 — CHOOSE AND STATE YOUR TARGET (10 minutes)
- Select one of the following targets, or propose your own with approval:
- (A) AI performs 50% of current US knowledge-work tasks at human cost or below
- (B) An AI system scores above 85% on the ARC-AGI benchmark
- (C) AI-assisted drug discovery reduces average time from candidate identification to Phase II trial by 50%
- (D) A target of your own design — state it precisely and explain how it would be measured
- For your chosen target, write a one-paragraph explanation of why this target matters and what achieving it would imply.
- PART 2 — ANALYZE THE COMPONENTS (15 minutes)
- For your target, analyze each of the following:
- (a) Current trajectory: what is the trend, and what specific evidence supports it?
- (b) Top three bottlenecks that could prevent the target from being reached
- (c) Top two enablers that could accelerate progress beyond baseline
- (d) Most relevant historical comparison, and what you take from it
- (e) Which scenario from Lesson 4 your forecast most closely tracks
- PART 3 — ASSIGN YOUR PROBABILITY DISTRIBUTION (10 minutes)
- Provide probability estimates for your target being achieved by:
- - 2030
- - 2035
- - 2045
- - Never (via current approaches)
- Your four estimates should reflect a coherent view of the trajectory. Write one sentence for each estimate explaining the key reason you placed the probability where you did.
- PART 4 — STATE ASSUMPTIONS AND UPDATE TRIGGERS (10 minutes)
- List your three most important assumptions. For each assumption, state: what evidence would cause you to revise your probability upward, and what evidence would cause you to revise it downward.
- PART 5 — POLICY IMPLICATIONS (5 minutes)
- Given your probability distribution, write two concrete policy recommendations: one directed at individuals or organizations preparing today, one directed at government. Explain how your probability estimates specifically motivate each recommendation.
- PART 6 — PEER REVIEW
- Exchange forecasts with another student. Evaluate: Is the target precise? Are the assumptions explicit? Do the probability estimates reflect the analysis, or do they seem uncalibrated? Identify the single strongest argument that would update your peer's probability estimate and share it with them.
- Final reflection: What was the hardest part of building a rigorous forecast? What would you need to know to make your estimates more confident?