The Einstein Test: Why the Man Who Won a Nobel Prize for AI Thinks We Are Still Far From True…

2026-05-10 · Source: Artificial Intelligence in Plain English - Medium · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Fundamental Awareness, quick

Summary

Demis Hassabis, CEO of Google DeepMind and a 2024 Nobel laureate for his work on AlphaFold, proposes the "Einstein Test" as a benchmark for artificial general intelligence (AGI). AlphaFold, an AI system developed by his team, successfully predicted protein three-dimensional structures from amino acid sequences, a long-standing biological problem. Hassabis suggests that for an AI to demonstrate true intelligence, it must be trained on all human knowledge available up to 1911 and then asked a question that no human with that same knowledge base could answer. This test aims to assess an AI's ability to generate novel scientific discoveries, rather than merely processing existing information or mimicking human intelligence.

Key takeaway

For research scientists developing AGI, the Einstein Test shifts the focus from mere knowledge acquisition to genuine scientific discovery. You should consider designing AI systems and evaluation metrics that prioritize the generation of novel insights and theories, rather than just advanced pattern recognition or information synthesis. This perspective challenges current benchmarks and encourages a deeper exploration of AI's creative and problem-solving capabilities.

Key insights

True AGI requires an AI to generate novel scientific discoveries beyond its training data, not just mimic human intelligence.

Principles

• AGI must create new knowledge.
• Intelligence transcends data processing.

Method

Train an AI on pre-1911 human knowledge, then test its ability to answer questions unanswerable by humans of that era, demonstrating novel scientific discovery.

In practice

• Focus AI development on discovery.
• Design benchmarks for novelty generation.

Topics

• Demis Hassabis
• Google DeepMind
• Einstein Test
• Artificial General Intelligence
• AlphaFold

Best for: Research Scientist, AI Scientist, Director of AI/ML, General Interest

Related on AIssential

• AGI: What Gets Measured Gets Built — Current AGI evaluation methods are flawed, shaping AI to pass tests rather than exhibit true, safe intelligence.
• OpenAI researchers explain why math is the road to AGI — Mathematics serves as a crucial benchmark for AGI due to its demand for consistent, verifiable long-chain reasoning.
• [AINews] Humanity's Last Gasp — AI's progress in automation paradoxically correlates with increased human workload, highlighting a gap in generalizable learning.
• Artificial general intelligence must be assessed in its scientific and societal context — Assessing AGI requires considering both scientific and societal contexts, not just technical capabilities.
• Yann LeCun Pushes Back on AGI Hype at India AI Summit: "Superintelligence Isn't Close" — True AI requires a "world model" for physical interaction, a capability current LLMs and systems lack.
• AGI: Francois Chollet + Sam Altman — AI's rapid advancement necessitates evolving social contracts and re-evaluating human agency and intelligence.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence in Plain English - Medium.