Why Scenarios, Not Predictions
Prediction assumes the future is deterministic. Scenarios recognize that it is contingent — it depends on human decisions, historical accidents, and feedback loops that no one fully controls. The scenario method was developed by Shell Oil in the 1970s to navigate oil shocks. Today it is used by governments, central banks, and AI laboratories to think the unthinkable.
The five scenarios below are not equally probable. They are equally possible — and each already has embryonic evidence in the present.
Scenario 1: Controlled Acceleration
Triggers: Regulations like the European AI Act become the global standard. The US passes federal AI legislation. China maintains intense development but within national borders. AI companies accept independent audits as the norm.
What happens: AI advances rapidly, but within regulatory guardrails. Language models reach capabilities close to narrow AGI in specific domains (medicine, law, engineering), but with mandatory human oversight. The parameter arms race slows; efficiency becomes the new competitive frontier.
For Brazil: A favorable scenario. PL 2.338/2023 would place the country in alignment with international standards. Brazilian companies could compete in niches of regulated AI — especially agribusiness and health.
Estimated probability (Metaculus, Jun/2026): 34%
Scenario 2: Geopolitical Fragmentation
Triggers: US-China tech war deepens. Chip exports become the primary geopolitical weapon. Europe develops its own models for sovereignty. Emerging countries are caught between two incompatible ecosystems.
What happens: Two (or three) parallel AI internets. American models dominate the West; Chinese models dominate the Global South via tech diplomacy. Incompatible APIs, data that cannot cross borders, regulations that fragment the global market.
For Brazil: A scenario of pressure. Brazil would need to choose sides or attempt strategic neutrality — a historically difficult position to sustain.
Estimated probability: 28%
Scenario 3: Extreme Concentration
Triggers: The cost of training frontier models continues to rise exponentially. Only 3-5 companies in the world can compete at the frontier. Regulation favors incumbents via expensive compliance requirements.
What happens: Frontier AI becomes a global oligopoly. Small companies and countries depend on big tech APIs for any advanced application. Open source maintains vitality in smaller models, but the frontier is inaccessible.
For Brazil: A scenario of moderate risk. The country already depends on foreign infrastructure for advanced AI. Concentration would deepen this dependency.
Estimated probability: 22%
Scenario 4: Radical Democratization
Triggers: Model efficiency advances faster than compute costs. 7B parameter models reach capabilities that today require 70B. Cheap inference hardware proliferates. Open source dominates common use cases.
What happens: Powerful AI runs on ordinary laptops and smartphones. The competitive advantage shifts from "having the model" to "knowing how to use the model." Entry barriers fall; proliferation of applications across all sectors and countries.
For Brazil: The most favorable scenario of all. With the third largest developer base in the world and a culture of technological adaptation, Brazil would benefit disproportionately from democratization.
Estimated probability: 12%
Scenario 5: Disruption by Incident
Triggers: A serious accident involving autonomous AI (healthcare system, critical infrastructure, financial market) triggers an emergency global regulatory reaction. Or: an unexpected capability advance creates widespread moral and political panic.
What happens: Temporary or permanent moratorium on certain applications. Rush to reactive regulation without solid technical basis. AI market enters recession for 2-4 years.
For Brazil: A paradoxically opportunistic scenario. Countries not at the frontier suffer less immediate impact. Brazil could position itself as a responsible regulator if it has a solid regulatory framework at the moment of the incident.
Estimated probability: 4%
What the Scenarios Have in Common
Three constants appear in all scenarios: the importance of sovereign data, the need for specialized human capital, and the weight of regulatory decisions made today. No favorable scenario for Brazil happens passively — all require deliberate action on at least one of these three fronts.
The future of AI in 2030 is not being written in the laboratories of San Francisco or Beijing. It is being co-written by every industrial policy decision, every investment in technical education, and every law passed — or postponed — in the world's parliaments.
