Humanity’s special power

Overview

All this time we have considered our intelligence as something that sets us apart from, and superior to, all other species, past and present, that have inhabited the earth. It might be time to move aside.

The chapter starts, as most do, with a “parable”: biological life on earth starting as a game between various species gods (Tree God and Scorpion God and Hominid God among others), and all the gods being incredulous about the hominid-god thinking his species could dominate despite having no superior physical attributes.

What is Intelligence?

A human brain can learn to navigate wider-ranging paths through a larger cross-section of reality than any other animal.

Random Thought: I think one of the fundamental flaws of human intelligence (human brain? Human mind?) was that we objectified the world around us. Where did that happen? Why didn’t indigenous people do it?

Intelligence is about two types of work: the work of predicting, and the work of steering.

Prediction and steering are entangled, but different.

As two humans get smarter, they will agree more and more on prediction, but they can steer toward different final destinations through no defect of intelligence.

Humans still have the edge at “generality” – ability to predict and steer across a broad array of domains.

OpenAI model o1 (and o3?) are much more general in their abilities.

Advantages machines have over biological brains:

Sheer speed: Transistors switch on and off billions of times per second compared to the fastest neurons at a 100 times per second. AI predicting and steering the world at least 10,000 times faster than humans would reduce us to statues equivalent to speaking a word per hour.

Really important for us to grasp that there’s more power in doing less and that may be what saves us

Copy-and-paste abilities: 20 years to grow a new human with a tiny fraction of human knowledge vs AI replicating ALL of it on demand

This feels like something that massively irks tech billionaires who are desperately trying to gain immortality through loading themselves into a digital world

Faster improvements: human brains ran into a bottleneck with baby heads getting too large to fit through female hips. [On the flip side, we are tirelessly helping our eventual masters by building much, much faster chips every day]

Larger memories. Human brains have more storage than a current laptop, but a datacenter beats human brains hands down.

Higher-quality thinking: Human brains fall prey to many subjective, systematic errors (e.g. arguing against conclusions we don’t like), in addition to not using our neurons effectively (e.g. inability to multiply 3 digit numbers in our heads). [I still wonder though: wouldn’t all these human errors ultimately be inextricably wired into AI?]

Self-experimentation and self-rewriting capabilities: AIs could make copies of themselves and experiment on them, graft new computational processes and build different versions of themselves much more quickly than humans can.

In conclusion, an “intelligence explosion” – an AI that makes a smarter AI that figures out how to make an even smarter AI is a possibility, and even an expectation. AI executives, while not entirely trustworthy, have indicated that they were planning to build “superintelligence”, so we are clearly not treating this like a risk.