The Era of Infinite Intelligence
Superintelligent AI might also prove completely uncontrollable and destroy humanity, whether intentionally or as mere collateral damage in the path of achieving other goals.
The clashing viewpoints about the potential and dangers of peak AI live at the heart of the battle of techno-optimists and doomsayers, accelerationists vs doomers.
No matter what side you fall on about the promise or danger of superintelligence, we don’t need it to achieve the extraordinary outcomes in the first scenario. We only need to achieve infinite intelligence, and there’s a path to get there in just a few years.
Here’s the difference between the two:
- Superintelligence is an AI vastly more intelligent than the most intelligent human. It is a singular, centralized system capable of extreme knowledge and understanding across any and every domain. Superintelligence solves problems with apparent elegance through superior intellect and without human involvement.
- Infinite intelligence is a massively scaled system with intelligence in a specific field or fields equivalent to a human college graduate. It is more specialized and decentralized than superintelligence. Infinite intelligence solves problems with brute force alongside some human guidance.
There are about 1 million mathematicians in the US workforce today. Imagine a world where AI can do graduate level math, and we could spin up a billion instances of those digital mathematicians. We’d have 1,000x more artificial brains working on mathematical problems than we do now. If we focused a firehose of AI instances with graduate level math knowledge, we’d probably solve some important problems, and you could extend the same approach to chemistry (only about 80,000 chemists in the US workforce), physics (20,000 physicists), biology (100,000 biologists), and more.
That’s infinite intelligence. It wins through brute force vs superintelligent elegance, but winning is winning. All human excellence is brute force. There are 7 billion of us. Only a handful participate in meaningful breakthroughs in society. We’re surrounded by the output of brute force intelligence. Infinite intelligence seems a most natural extension of the brute force human experience.
The infinite monkey theorem: At scale, AI can even win by accidental brute force.
How Far are We From Infinite Intelligence?
Not far. We may already be living it.
Google’s DeepMind used a large language model to solve a stubborn math problem through what seems a method in evolutionary brute force. The model treated the math problem like a puzzle. Per the Nature paper explaining the breakthrough, the model “searches for programs that describe how to solve a problem, rather than what the solution is.”
With better models and more efficient compute, infinite intelligence may only be a few years away.
One catalyst is achieving AGI, which could happen soon. Jack Kendall, CTO at Rain AI predicted in my interview with him that we’ll have AGI by 2025 using a specific definition:
- A single network architecture that can in principle learn any combination of data modalities – e.g. speech, vision, language. We have this now.
- The ability to use prior experience to learn new things rapidly. That’s causal understanding.
- The ability to reason about things and perform tasks it hasn’t seen before. That’s symbolic reasoning.
Symbolic reasoning includes the ability to do math, which is critical to graduate level intelligence in any field.
Note that Jack’s definition of AGI doesn’t require the AI to understand like a human. That’s human-level intelligence. AGI only requires that a machine can learn new things and perform tasks it hasn’t seen before, whether that means the AI thinks like a human or not.
By achieving Jack’s AGI, we have all the pieces for infinite intelligence.