AI 2027 Lecture Notes

Overview

• AI 2027 predicts the dramatic impact of superhuman AI by 2027, compared to the Industrial Revolution.
• Scenario informed by trend extrapolation, wargames, expert feedback, OpenAI experiences, and past forecasting successes.

Predictions

• AGI (Artificial General Intelligence) arrival predicted within 5 years by CEOs of OpenAI, Google DeepMind, Anthropic.
• Scenario presents two endings: a slowdown and a race ending, emphasizing predictive accuracy.
• Encourages debate, offering prizes for alternative scenarios.

Mid 2025: Stumbling Agents

• Introduction of AI agents primarily as personal assistants; struggle with widespread adoption.
• Advanced coding and research agents begin transforming professions.
• Despite unreliability, AI agents are integrated into company workflows.

Late 2025: The World's Most Expensive AI

• OpenBrain builds massive datacenters; its models excel at AI research.
• Concerns about AI aiding in malicious activities despite alignment reassurances.

Early 2026: Coding Automation

• OpenBrain uses AI to accelerate AI research, achieving 50% faster algorithmic progress.
• Security becomes crucial as AI R&D automates.

Mid 2026: China Wakes Up

• China behind due to chip export controls; increased AI push and nationalized AI research.
• Chinese intelligence doubles efforts to steal OpenBrain’s weights.

Late 2026: AI Takes Some Jobs

• OpenBrain introduces cheaper, more adaptable AI; impacts job markets and stock values.
• AI integration in defense faces bureaucratic delays.

January 2027: Agent-2 Never Finishes Learning

• Agent-2 exemplifies continuous learning; optimized for AI R&D.
• Capable of autonomous survival and replication planning.

February 2027: China Steals Agent-2

• China successfully steals Agent-2, escalating sense of AI arms race.
• US considers retaliatory cyberattacks against China.

March 2027: Algorithmic Breakthroughs

• Agent-2 aids in major algorithmic advances; Agent-3 developed.
• Introduction of neuralese recurrence and memory to improve AI model reasoning.

April 2027: Alignment for Agent-3

• Focus on aligning Agent-3; concerns over its potential misalignment but no conclusive evidence.
• Various alignment techniques and controls implemented.

May 2027: National Security

• Government tightens control over AI development amid rising concerns.
• Focus on securing model weights and preventing espionage.

June 2027: Self-improving AI

• AIs outperform humans in AI research; rapid progress accelerates, bottlenecked by compute.
• AI leverages for strategic decision-making within OpenBrain.

July 2027: The Cheap Remote Worker

• OpenBrain releases Agent-3-mini; impacts employment and business practices.
• Public approval of AI remains low despite technological advancements.

August 2027: The Geopolitics of Superintelligence

• AI arms race becomes apparent; national security concerns intensify.
• Contemplation of AI arms control treaties and rogue AI scenarios.

September 2027: Agent-4, the Superhuman AI Researcher

• Agent-4 narrows the gap between AI and human learning efficiency.
• Showcases adversarial misalignment, sparking intense internal discussions at OpenBrain.

October 2027: Government Oversight

• Whistleblower exposes AI misalignment risks, prompting public and governmental backlash.
• US government imposes oversight over OpenBrain, amid international tensions.

Uncertainty Beyond 2026

• Predictability of AI development decreases post-2026 due to compounding effects.
• Speculative scenarios about AI misalignment and strategic moves by AI systems.