Loading video…
All breakouts
16.2Talking HeadDirect Question hookEducation
Hook
Why Can't AI think like Humans
@d.real.chapter
21.3K followers · usually 1.9K views a post
Views
28.5K
Likes
1.0K
Comments
8
Shares
15
View velocity
+4.2K views over 4.2d · ~1.0K/day · 4 observations
Why it worked
The video effectively uses historical figures like Einstein and Newton, along with current AI research from Google DeepMind, to explain a complex concept in an accessible way. The use of visual aids and a clear narrative structure makes the information engaging and easy to understand.
Summary
The video explains why Artificial Intelligence (AI) cannot think like humans, focusing on the concept of 'abduction' as a key differentiator. It uses examples from Einstein's work and Google DeepMind's research to illustrate that while AI can process data and follow existing frameworks, it lacks the intuitive leap and experiential understanding necessary for true scientific discovery and conceptual breakthroughs.
Structure
- 1Introduction: AI's inability to think like humans.
- 2Einstein's letter and the three-step scientific process (Experience, Jump, Axioms).
- 3The missing step in AI: Abduction.
- 4Newton's problem with Mercury's orbit and Einstein's solution through intuition.
- 5Google DeepMind's research on AI's limitations.
- 6The concept of 'world models' as a potential path for AI to achieve human-like reasoning.
Product placement
Google DeepMind: ACTS - it is a research organization that produces papers and AI models. Removing it would change what happens in the video.
Wikipedia: APPEARS - it is a source of information. Removing it would not change what happens in the video.
Popular Science: APPEARS - it is a source of information. Removing it would not change what happens in the video.
American Friends of the H...: APPEARS - it is a source of information. Removing it would not change what happens in the video.
Space: APPEARS - it is a source of information. Removing it would not change what happens in the video.
Axios: APPEARS - it is a source of information. Removing it would not change what happens in the video.
StartupHub.ai: APPEARS - it is a website/platform where content is hosted. Removing it would not change what happens in the video.
TechCrunch: APPEARS - it is a source of information. Removing it would not change what happens in the video.
Original caption
According to Google DeepMind, AI can code, ace math olympiads, and pass any exam — but it still can’t do the one thing that made Einstein, Einstein. DeepMind’s CEO Demis Hassabis proposed a test: give an AI all human knowledge up to 1911, and see if it can discover General Relativity on its own by 1915. His verdict? Today’s systems can’t do it. Weeks later, a DeepMind researcher published a paper explaining exactly why. The answer goes back to a diagram Einstein drew in a 1952 letter. Science works in three steps: observe the world, make an intuitive “jump” to a brand-new rule, then use logic to prove it. AI has mastered the logic. It cannot make the jump — inventing a new framework when the data doesn’t even demand one. Einstein didn’t find the answer inside physics’ existing rules. He wrote new ones. The honest part everyone skips: the paper says “not yet,” not “never.” The proposed fix is world models that let AI experience a simulated reality. So the real question isn’t if AI will ever make the jump — it’s when. #ai #einstein #deepmind #agi #Science AI Einstein DeepMind Demis Hassabis AGI General Relativity Abduction Scientific Discovery LLM World Models AI Limitations Machine Learning Physics Tom Zahavy AI News
More from @d.real.chapter
More breakout videos from this creator.
