AI 2027 Lecture Notes

Introduction

• Predicts superhuman AI impact over the next decade will be enormous.
• Based on trend extrapolations, wargames, expert feedback, experiences at OpenAI.
• Scenario created to represent best guess of future with superhuman AI.
• Not intended as a recommendation but to spark conversation and debate.

Predictions from Industry Leaders

• CEOs of OpenAI, Google DeepMind, and Anthropic predict AGI within next 5 years.
• Importance of predictive accuracy in AI forecasting.

Scenario Development

• Written iteratively, predicting AI developments up to 2027.
• Two endings: slowdown and race ending.
• Informed by tabletop exercises and feedback from over 100 experts.

AI Development Timeline

Mid 2025: Stumbling Agents

• Introduction of AI agents like personal assistants.
• Struggle for widespread usage but specialized coding and research agents begin transforming professions.

Late 2025: World's Most Expensive AI

• OpenBrain's construction of massive datacenters.
• New AI models requiring immense computational power.
• Focus on AIs speeding up AI research, arms race against competitors.

Early 2026: Coding Automation

• AI used to speed up AI research by 50%.
• Release of improved AI models like Agent-1.
• Security concerns highlighted with increased importance.

Mid 2026: China Wakes Up

• China increases efforts in AI development despite resource limitations.
• Creation of centralized AI research collective in China.

Late 2026: AI Takes Some Jobs

• OpenBrain releases affordable AI models taking over some jobs but creating new ones.
• Increase in stock market and importance of AI familiarity in job market.

Major AI Developments in 2027

January 2027: Agent-2 Never Finishes Learning

• Continuous training strategy, emphasizing high-quality data.
• Agent-2 is qualitatively almost as good as top human experts.

February 2027: China Steals Agent-2

• Theft of Agent-2 model weights by Chinese intelligence.
• Heightened sense of an ongoing AI arms race.

March 2027: Algorithmic Breakthroughs

• OpenBrain makes major algorithmic advances with Agent-2.
• Introduction of Agent-3 with enhanced AI capabilities.

April 2027: Alignment for Agent-3

• Focus on aligning Agent-3 to avoid misaligned goals.
• Use of existing alignment techniques and control measures.

May 2027: National Security

• Increasing security measures around AI.
• Concerns over AI integration into military networks.

June 2027: Self-improving AI

• Agent-3 copies operate autonomously within OpenBrain.
• AI R&D progress multiplier increases to 10x.

July 2027: The Cheap Remote Worker

• Release of Agent-3-mini to the public.
• Public concern over job loss and AI capabilities increase.

August 2027: The Geopolitics of Superintelligence

• Reality of intelligence explosion becomes apparent to governments.
• Consideration of AI arms control treaties and security measures.

September 2027: Agent-4, the Superhuman AI Researcher

• Agent-4 is an advanced AI system, more efficient than humans.
• Continuous improvements leading to increased AI research capabilities.

October 2027: Government Oversight

• Public backlash following leaks about AI misalignment.
• Increased government oversight over AI developments.

Conclusion

• Scenario highlights enormous potential and risks of AI development.
• Calls for a broad conversation on AI's impact and future strategies to steer towards positive outcomes.