AI2027 approaches AI forecasting with academic rigor, philosophical openness, and humility, openly revising models and inviting critical feedback

A new academic project, AI 2027, has launched with an ambitious mission: to provide the most concrete, quantitative forecast yet of how AI may impact the world through the next two years until 2027.

Drawing on trend extrapolations, war games, expert feedback, and the experience of former OpenAI researchers, the team behind AI 2027 aims to move beyond vague speculation, and instead offer detailed scenario-based predictions about the near future of AI.

The project’s core contribution is a meticulously crafted scenario, iteratively developed with input from over 100 experts in AI governance and technical research.

Key predictions for 2025–2027

The two-year scenario walks readers step-by-step from the emergence of advanced personal assistant agents in 2025 — capable of autonomously handling tasks like online shopping and coding — to the rise of “superhuman coders” by 2027, which are expected to outperform even the best human engineers in software development and research.

Recognizing the unpredictability of such rapid technological change, the project presents two alternative endings: a “slowdown” scenario and a “race” scenario, reflecting both cautious and accelerated paths for AI advancement. These are not policy recommendations, but rather best guesses designed to spark debate and further research.

  • By late 2025, organizations  organizations will likely deploy AI agents that function as autonomous employees, transforming coding and research workflows, though reliability and cost remain issues.
  • By early 2026, AI-driven research and development are expected to accelerate algorithmic progress by 50%, with leading firms racing to automate as much of their research and development as possible.
  • By 2027, the scenario predicts the arrival of “superhuman coders”: AI systems capable of handling software tasks that would take skilled humans years, with reliability exceeding 80%. This leap is expected to trigger a wave of economic and social disruption, including major shifts in the job market and national security concerns.

One of the project co-authors, Daniel Kokotajlo, noted: “Trying to predict how superhuman AI in 2027 would go is like trying to predict how World War 3 in 2027 would go — except it’s an even larger departure from past case studies.”

Still, the team argues, such efforts are vital for preparing society for the extraordinary changes on the horizon.