A plain-language guide to superintelligence
The Last Invention
Artificial intelligence is the first tool that might learn to build better tools than we can. This is the story of that idea.
ScrollPart one · Everything before
Seven entries out of millions. Watch the dates.
-
c. 1,000,000 BCE
Fire
Heat, light and cooked food. The first source of energy we controlled that wasn’t our own bodies.
-
c. 3500 BCE
The wheel
First on a potter’s table, then under a cart.
-
c. 3200 BCE
Writing
Memory that outlives the person who had the thought.
-
c. 1440
The printing press
The same page, a thousand times over.
-
1712
The steam engine
Work without muscle, wind or water.
-
1903
Powered flight
Twelve seconds off the ground at Kitty Hawk.
-
1947
The transistor
A switch with no moving parts. We have since made more of them than of any other object.
- From fire to the wheel took about a million years. From the first flight to the Moon took sixty-six.
-
Part two · Thinking machines
Then we started building tools that think.
-
1950
The imitation game
Alan Turing asks whether machines can think, and proposes a test instead of an answer.
-
1956
A name for the field
A summer workshop at Dartmouth College gives the work its name: artificial intelligence.
-
1997
Deep Blue
A chess computer beats the world champion, Garry Kasparov, over six games.
-
2012
AlexNet
A deep neural network wins an image-recognition contest by a wide margin. Deep learning takes over the field.
-
2016
AlphaGo
A program beats Lee Sedol at Go, a game many expected to resist computers for another decade.
-
2017
The transformer
A network design that learns from text at enormous scale. It is the T in GPT.
-
2022
ChatGPT
A chatbot reaches an estimated hundred million users in two months. AI becomes something everyone has tried.
-
20??
The last one
A machine that out-thinks us at everything, including building machines. This entry writes the rest of the ledger.
“…the first ultraintelligent machine is the last invention that man need ever make.”
Good was a mathematician who had broken codes at Bletchley Park alongside Alan Turing. His argument was short: a machine smarter than us would also be better than us at designing machines, including smarter ones.
He attached one condition, and it is the part that gets quoted less often: only if the machine is docile enough to tell us how to keep it under control.
The idea in four parts
Understand superintelligence
What is superintelligence?
An intellect that outperforms the best human minds in practically every field. Not a faster calculator: a better thinker. How it differs from AI and AGI.
Read the explainer →The intelligence explosion
If a machine can improve its own design, each improvement makes the next one easier. The 1965 argument, the case for it, and the reasons to doubt it.
Read the argument →A timeline of AI
From a 1943 model of the neuron to chatbots with hundreds of millions of users: the dates that mattered in artificial intelligence, and why.
See the timeline →The vocabulary
AGI, ASI, alignment, takeoff, singularity. The terms people argue with, each defined in a sentence or two of plain English.
Open the glossary →Short answers
Questions people ask
What does “the last invention” mean?
It is shorthand for superintelligence. The phrase comes from the mathematician I. J. Good, who argued in 1965 that a machine more intelligent than any person could design even better machines, so people would never need to invent anything again.
Who said AI would be the last invention?
Irving John Good, a British mathematician who worked with Alan Turing at Bletchley Park. He wrote it in “Speculations Concerning the First Ultraintelligent Machine”, published in 1965. Nick Bostrom and James Barrat later brought the phrase to a wider audience.
What is superintelligence?
An artificial intelligence that greatly exceeds human ability across practically every field, including scientific research, strategy and social skills. It is a step beyond artificial general intelligence (AGI), which means roughly human-level ability across the board. More on superintelligence.
What is the difference between AI, AGI and superintelligence?
AI is any software that performs tasks we associate with intelligence. Today’s systems are strong at some tasks and weak at others. AGI would match a capable person across most intellectual tasks. Superintelligence would exceed the best people at all of them.
Does superintelligence exist yet?
No. Current AI systems can write, code and reason through many problems, but they do not outperform human experts across the board, and they still depend on people to build the next version. Researchers disagree about whether superintelligence is years away, decades away, or possible at all.
What is the intelligence explosion?
A feedback loop. An AI that is good enough at AI research improves itself, the improved version is better still at improving itself, and capability rises quickly. Whether this would take days or decades is one of the main open questions. The full argument.
Why would superintelligence be dangerous?
A system far more capable than its makers would be hard to correct if its goals differed from theirs, even slightly. Good made the point himself: his argument holds only if the machine is docile enough to tell us how to keep it under control. The research field built around this problem is called AI alignment.
Why is this site on a .si domain?
.si is the country code for Slovenia. It also happens to be the initials of superintelligence.
Status of the last invention
Not yet.
Last checked today · still uninvented