Re: Re: The Idea of Human-Level Intelligence
I’ve been thinking more about human-level intelligence, especially after revisiting my previous blog posts and reflecting on insights from Demis Hassabis. His discussions on AGI and creativity—particularly from his interviews on the Big Technology Podcast by Alex Kantrowitz and his speech at the Royal Academy of Arts—offer a fascinating lens to reconsider my earlier thoughts.
What still makes humans special today—or whether we are special at all—is the question that keeps pulling me back to this topic. If AI eventually replicates everything we consider unique to human intelligence, does that diminish our significance? Or will there always be something inherently irreplaceable about human thought, creativity, and intuition?
AGI
Hassabis defines AGI (Artificial General Intelligence) as a system capable of performing all or most human cognitive tasks with the ability to adapt to new situations seamlessly. This includes perception, reasoning, long-term memory, autonomous planning, and true creativity—not just recombination of known patterns but the ability to propose genuinely new inventions. Current AI, despite its impressive capabilities in specific domains, still struggles with coherence in reasoning, long-term memory retention, and the abstraction needed for true creativity.
AI & Creativity
Creativity is one area where my thoughts have evolved, especially after reflecting on Demis Hassabis' breakdown of AI creativity. He categorizes it into three levels:
- Interpolation: Recombining existing data into new but predictable ways—like early GAN models generating faces by blending real features.
- Extrapolation: Moving beyond known data into novel territory, as seen in AlphaGo’s Move 37, which revealed a non-human strategic insight within an existing framework.
- Invention: Creating entirely new concepts, paradigms, or scientific theories, akin to how humans invent new games rather than just mastering existing ones.
This ties into a hypothetical my co-founder and I discussed: What if an AI, given only a dictionary, could generate Shakespeare-level works—not by being trained on existing literature but by developing something truly original? That would signify genuine creative intelligence. If AI could reach that level of invention, it would redefine our understanding of machine creativity and perhaps challenge what we consider uniquely human.
Knowledge Transmission
Another key distinction is how humans accumulate and transmit knowledge. Hassabis emphasizes the challenge of integrating long-term memory into AI, which aligns with my earlier argument that human intelligence thrives on generational knowledge-building. Culture, language, and historical context shape how we innovate. AI may someday develop a similar knowledge-preserving mechanism, but it hasn’t yet.
Expanding the Definition of Intelligence
Revisiting the broader perspective on intelligence, my previous posts argued that intelligence isn’t one-size-fits-all. Examples from nature—such as collective intelligence in ant societies, knowledge transmission among dolphins, and the emergent problem-solving abilities of slime molds—challenge our human-centric view of intelligence. Hassabis’ work reinforces this idea: AI will need more than just problem-solving ability to reach AGI; it must also possess adaptability, contextual awareness, and autonomous goal-setting.
Final Thoughts
This brings us back to the fundamental question: Is "human-level intelligence" even the right benchmark? Perhaps the better measure is intelligence that can continuously evolve, question itself, and create new paradigms of thought. If an AI can reach the level of true invention—not just recombining data but generating entirely new ideas—we will have entered a fundamentally new era of intelligence.
For now, AI remains an incredible tool, but the essence of human creativity, intuition, and abstract reasoning remains uniquely ours.
