The AI Transition Tutorial

A practical orientation to the civilizational question the next 10 to 20 years are asking: whether increasingly capable AI becomes powerful and aligned, whether benefits concentrate or distribute, and whether institutions metabolise the change fast enough.

Who This Is For

• Thoughtful readers trying to make sense of accelerating AI discourse
• Technologists who use AI daily and want a wider frame
• Policy staff, journalists, and analysts wanting better models of the actual debate
• Anyone choosing a career, a company, or an investment under AI uncertainty
• People wondering whether to worry, whether to celebrate, or whether both

This tutorial does not predict outcomes. It gives you the vocabulary, frames, and sources to engage with the transition seriously.

Contents

Fundamentals

1. Introduction: What "the AI transition" means, why this framing, what's at stake
2. What Is Happening: Honest description of current capabilities, trajectory, and uncertainty

Core Concepts

3. Alignment: The problem of getting AI to do what we actually want
4. Concentration vs Distribution: Who benefits, who decides, who owns
5. Institutions: Can existing institutions metabolise this change?
6. Capability Timelines: Transformative AI, AGI, the uncertainty about when

Advanced

7. Failure Modes: From gradual disempowerment to much worse
8. Success Modes: What going well actually looks like, concretely
9. Reading the Discourse: Believers, skeptics, neutrals, and how to read them

Ecosystem

10. Companies and Governance: Labs, governments, international coordination
11. Individual Orientation: Epistemics, careers, engagement

Mastery

12. Best Practices: Habits, anti-patterns, the long view

How to Use This Tutorial

1. Read sequentially the first time. The alignment, concentration, and institutions chapters set up a frame the later chapters build on
2. Resist premature conclusions. This is a contested space with real uncertainty. The frame matters more than your current opinion
3. Read the skeptics alongside the believers. A tutorial like this can accidentally become a single-camp document; the reading list is deliberately mixed

Quick Reference

The Three Big Questions

Alignment      Will AI do what humans actually want?
Distribution   Will benefits and power concentrate or spread?
Metabolism     Can institutions adapt fast enough?

Almost every specific AI policy argument reduces to some combination of these.

The Honest Position

We don't know when transformative AI arrives.
We don't know which failure modes are most likely.
We don't know whether existing institutions are adequate.
Acting well under this uncertainty is the skill.

The Core Tensions

Speed vs safety         race dynamics push fast; alignment needs care
Open vs closed          transparency aids trust; also proliferates capability
National vs global      AI is global; regulation is national
Centralised vs plural   concentrated development is faster; distribution is safer
Believers vs skeptics   both camps include thoughtful people; monoculture is bad

Learning Path Suggestions

The curious generalist (roughly 5 hours)

1. Chapters 01 and 02 for the frame
2. Chapter 03 on alignment
3. Chapter 09 on reading the discourse
4. Chapter 12 for orientation

The policy-oriented reader (roughly 6 hours)

1. Chapters 04 and 05 on concentration and institutions
2. Chapter 07 on failure modes
3. Chapter 10 on companies and governance
4. Chapter 12 for habits

The careerist (roughly 5 hours)

1. Chapter 02 on capabilities
2. Chapters 04, 07, 08 on outcomes
3. Chapter 11 on individual orientation
4. Chapter 12 for habits

Why This Matters

• The stakes are plausibly very high. A small probability of civilizational change is still worth understanding
• Decisions made now shape the trajectory. Regulations, norms, investment, and deployment patterns set over the next few years will influence outcomes for decades
• Most discourse is polarised. Good-faith thinking is underrepresented; the ability to read honestly is itself scarce
• Individual orientation is not optional. Whether or not you work on AI, AI will work on you

Recommended Reading

Reading list designed to give you a spectrum, not a single view:

Serious treatments of transformative AI

• Holden Karnofsky's "Most Important Century" series (accessible, careful)
• Joe Carlsmith's essays (philosophical depth)
• Paul Christiano's writings on alignment (technical)
• Superintelligence by Nick Bostrom (2014; dated in details, still worth reading)

Research labs (worth reading with the author's incentives in mind)

• Anthropic's research and policy papers
• OpenAI's alignment and safety writing
• DeepMind's safety research

Thoughtful skeptics

• Melanie Mitchell's writing (critiques of anthropomorphising current AI)
• Gary Marcus (sometimes polemical but substantive on LLM limitations)
• Emily Bender and Timnit Gebru (on harms and hype)
• Arvind Narayanan and Sayash Kapoor's AI Snake Oil (on commercial overclaims)

Economics and society

• Daron Acemoglu and Simon Johnson, Power and Progress (historical frame on technology and distribution)
• Erik Brynjolfsson's work on productivity and labour
• AI Now Institute reports on power and politics

Fiction that helps

• Exhalation by Ted Chiang (stories that build intuition)
• The Lifecycle of Software Objects by Ted Chiang (on raising AI)
• Daemon by Daniel Suarez (unsubtle but useful for thinking through disempowerment)

What This Tutorial Is Not

• A prediction about outcomes
• A take on specific policy debates
• A promotion of or argument against any particular lab or research program
• A guide to using AI tools (see writing-interface/ for that)
• Technical material on ML (see ai/)

It's a frame for reading a fast-moving, contested, genuinely consequential field with less heat and more signal.

A Note on Epistemic Discipline

The AI discourse rewards motivated reasoning. Boosters talk up what suits boosters. Critics emphasise what suits critics. Companies publish research shaped by their incentives. Governments propose rules shaped by their constituencies.

Real thinking requires noticing this and reading across sources anyway. The skill isn't picking a side; it's holding multiple serious positions in mind at once and updating on actual evidence rather than tribe.

This tutorial tries to model that. It probably doesn't succeed perfectly. Where it fails, read the opposing view and decide for yourself.