Case Study 1: Three Futures — Scenarios for AI in 2035

Introduction

Scenario planning is a tool used by governments, businesses, and researchers to prepare for uncertain futures. Instead of trying to predict what will happen, scenario planning asks: What could happen? How would we respond? What should we prepare for regardless of which path unfolds?

In this case study, we develop three detailed scenarios for AI in 2035 — not as predictions, but as structured thought experiments. Each scenario traces specific consequences for the four anchor examples from this book. Your task is to analyze the scenarios, evaluate their plausibility, and identify what each one implies about the choices we should be making today.

Scenario A: "The Steady Hand" — AI as Infrastructure

The World in 2035

AI has become infrastructure — as ubiquitous and unremarkable as electricity. Most people interact with AI dozens of times a day without thinking about it, just as they flip light switches without thinking about the power grid. This happened not through a single breakthrough but through a decade of steady improvement: AI systems became more reliable, more accurate, and more deeply integrated into existing tools and processes.

Crucially, governance kept pace with development. The EU AI Act proved effective, and its framework was adopted (with local adaptations) by over 60 countries. A binding international agreement on AI safety, signed in 2031, established minimum standards for high-risk AI systems and created an international inspectorate with real enforcement power. Industry pushed back but ultimately adapted, the way automakers adapted to seatbelt requirements.

ContentGuard has evolved into a multilingual, culturally adaptive system. It now works with local advisory boards in 40 countries that provide cultural context for moderation decisions. Its error rates have dropped substantially, particularly in non-English languages, though edge cases in satire, political speech, and cultural context remain challenging. Transparency reports are published quarterly and audited by independent organizations.

MedAssist AI is now standard in radiology departments worldwide. Performance gaps across skin tones have been largely (though not entirely) closed through diversified training data and mandatory pre-deployment equity audits. Physicians have been trained to use the tool as a partner, not a replacement, for clinical judgment. Liability frameworks clearly establish shared responsibility between the AI developer, the hospital, and the physician.

Priya graduated years ago. AI writing tools are now integrated into word processors the way spell-check was in the 2000s. Universities have adapted: assessment has shifted toward oral examinations, project-based work, and demonstrations of process rather than just final products. Students are taught to use AI effectively and critically, including understanding its limitations.

CityScope Predict was adopted in some cities and rejected in others. Cities that adopted it were required (by regulation) to publish their algorithms, submit to independent audits, and establish civilian oversight boards. The system reduced response times for some emergency services but had minimal impact on crime rates — largely because the underlying social determinants of crime are not addressable through patrol allocation.

Key Features of This Scenario

  • Governance is effective and adaptive
  • AI is beneficial but not transformative at a civilizational level
  • Inequities persist but are mitigated
  • The biggest AI story is boring: incremental improvement, like sanitation or aviation

Scenario B: "The Great Divide" — AI Amplifies Inequality

The World in 2035

AI capabilities have advanced dramatically — foundation models are far more capable than anything available in 2025. But the benefits have been captured disproportionately by wealthy countries, large corporations, and privileged populations. The gap between AI haves and have-nots has widened into a chasm.

Governance failed to keep pace. The international AI treaty negotiations collapsed in 2029, undermined by U.S.-China competition and industry lobbying. Each country has its own regulations, many of which are poorly enforced. The compute divide has deepened: training frontier models now costs billions of dollars, putting independent development out of reach for all but a handful of entities. Open-source alternatives exist but lag significantly behind proprietary systems.

ContentGuard operates effectively in English, Mandarin, Spanish, and a few other high-resource languages. In most other languages, moderation remains crude, leading to both over-censorship and under-moderation. The platform has withdrawn from some smaller markets entirely, finding them unprofitable. Independent content platforms have emerged in some regions but struggle to compete.

MedAssist AI is extraordinary — in well-funded hospitals. In wealthy-country teaching hospitals, AI-assisted diagnosis has reduced certain types of medical errors significantly. But access is deeply unequal. Rural hospitals, developing-country health systems, and under-resourced clinics cannot afford the latest systems and often rely on older, less accurate versions. The performance gap across demographics has improved in high-income settings but worsened in settings using outdated systems.

Priya's successors face a divided landscape. Students at well-resourced universities have access to sophisticated AI tutoring systems that provide genuinely personalized instruction. Students at underfunded institutions have access to basic tools but little training in how to use them critically. The gap in AI literacy tracks existing educational inequalities.

CityScope Predict has proliferated globally, often with inadequate oversight. In some democracies, robust governance frameworks have mitigated the worst risks. In countries with weak institutions, predictive policing systems operate with virtually no accountability, reinforcing discriminatory patterns and suppressing dissent.

Key Features of This Scenario

  • AI capabilities are impressive but concentrated
  • Governance is fragmented and inadequate
  • The technology amplifies existing inequalities
  • The Global South bears disproportionate costs

Scenario C: "The Pivot Point" — Crisis Forces a Reckoning

The World in 2035

A series of AI-related crises between 2028 and 2032 forced a fundamental reassessment. A deepfake video of a world leader nearly triggered a military confrontation before it was identified as synthetic. An AI trading system caused a severe financial market crash. And a widely deployed medical AI system misdiagnosed a rare side effect in a common medication, resulting in hundreds of preventable deaths across multiple countries before the error was identified.

The crises triggered intense public backlash and rapid regulatory response — some of it thoughtful, some of it reactive and counterproductive. Several countries enacted broad AI moratoriums. Venture capital investment in AI plummeted. Public trust cratered.

The result, by 2035, is a world that has pulled back from AI's frontier but invested heavily in AI safety, transparency, and governance. AI systems are more tightly regulated than any previous technology. Every high-risk AI deployment requires pre-market safety approval (similar to pharmaceutical regulation). International cooperation, galvanized by shared crisis, produced a binding AI safety treaty in 2033.

ContentGuard was temporarily shut down during the deepfake crisis, replaced by expanded human moderation teams. The reconstituted system operates under strict regulatory oversight, with mandatory real-time auditing and explicit liability for moderation failures. Free speech advocates argue the new regime is overly restrictive.

MedAssist AI was nearly destroyed by the medical misdiagnosis crisis. The reconstituted system operates under a regulatory framework modeled on pharmaceutical approval — extensive pre-market testing, ongoing post-deployment monitoring, and clear liability for failures. The system is more reliable but less innovative, as the regulatory burden discourages rapid iteration.

Priya's successors live in a world where AI tools in education are more restricted and more carefully supervised. AI tutoring is used, but under strict guidelines about appropriate use. Some educators argue the restrictions have gone too far, preventing beneficial applications.

CityScope Predict was banned outright in several countries following revelations that it had been used to target political dissidents in multiple jurisdictions. Predictive policing remains deeply controversial, and most democratic countries have imposed strict limitations or bans.

Key Features of This Scenario

  • Crisis drives governance, but at a cost
  • Regulation is comprehensive but sometimes overcorrective
  • Innovation slows; safety improves
  • Public trust in AI remains low

Analysis Questions

1. Rank the three scenarios from most to least desirable. Then rank them from most to least plausible. Are your rankings different? Why?

2. For each scenario, identify one policy decision that could be made today (in 2025-2026) that would make that scenario more or less likely.

3. None of the three scenarios represents a world where AI safety is fully solved. What does this suggest about the relationship between technological progress and the challenges of governing that progress?

4. The "Pivot Point" scenario involves crisis driving governance reform. Is governance-by-crisis a reliable approach? What are its strengths and weaknesses compared to proactive governance? Can you think of historical examples of each?

5. Create a fourth scenario — your own "most likely" version of 2035. Describe the state of AI governance, the fate of each anchor example, and the role of AI literacy. What assumptions underlie your scenario?

Connections

  • Chapter 2 (History of AI): The "AI winters" from Chapter 2 are the historical precedent for the Pivot Point scenario — cycles of hype, crisis, and retrenchment.
  • Chapter 13 (Governing AI): All three scenarios are fundamentally about governance — the difference between them is largely about how effectively governance responds to technological change.
  • Chapter 19 (Global Perspectives): The Great Divide scenario illustrates the global equity concerns from Chapter 19 at their most extreme.
  • Chapter 20 (AI Safety): The Pivot Point scenario illustrates what happens when safety concerns are validated by real-world failures.

Reflection Prompt

Which scenario would you most want to live in? Which scenario do you think you are currently living in the early stages of? What is one thing you could do — as a citizen, a professional, or a community member — to steer toward a better outcome?