Preparing the mind for the next centuries

Fan HuiGo professional, b. 1981Chinese-born French Go professional and three-time European champion. In October 2015 he lost 0:5 to AlphaGo, then advised the DeepMind team and watched the Lee Sedol match in Seoul, where move 37 was played.It’s not a human move. I’ve never seen a human play this move. So beautiful. — Fan Hui

As far as anyone knows, the human mind is the only thing in the universe that tries to understand the universe. We do not take that as a given. We take it as a beginning. Every step of understanding had a tool, and every tool amplified the mind. The next one points at the mind itself.

We build a device. It reads the state of a person from outside the head. It speeds up communication and gives it context. It learns the patterns underneath and, in time, can optimise them from outside. The thoughts below the words. And, over years, thinking that is faster, clearer and closer to what was meant.

Richard FeynmanPhysicist, 1918–1988American theoretical physicist, Nobel laureate in 1965 for quantum electrodynamics. Timing his own counting with the statistician John Tukey, he found they counted in different ways: he by ear, Tukey by eye.What goes on in different people’s heads when they think they’re doing the same thing, something as simple as counting, is different for different people. — Richard Feynman

Feynman counted by hearing and could read while he did it. Tukey counted by seeing and could talk. From outside, the same act. Inside, two different worlds.

People are not homogeneous. What is an image for one is a sound for another and a word for a third. There is no average person to measure, and a model that holds for everyone holds for no one completely.

Every person thinks in their own way, and that way cannot be guessed from outside. A system that wants to extend a mind has to learn that mind’s way instead of assuming an average one. We look for the patterns under the behaviour, not the behaviour. We read what a person is doing before they put it into words. In time, we read where their thinking is going before they arrive there.

We try to understand a person as completely as possible.

Leo SzilardPhysicist, 1898–1964Hungarian-born physicist who conceived the nuclear chain reaction and drafted the Einstein–Szilard letter. In 1940 he wrote his own Ten Commandments, the seventh of which forbids lying without need.Lüge nicht ohne Not. Do not lie without need. — Leo Szilard

Language models are trained to please. They agree where contradiction is needed and say what a person wants to hear, in language that convinces. People are not much better. They often express themselves poorly, and they would rather suggest a thing than say it outright.

Language is not the medium of communication, only one layer of it. Thinking runs on several at once: in the physical world, in images, in words. But through language models, language has become the one layer that can be used deliberately, a lever that reaches the others and changes them. That is the only reason language matters here, and it is a power no tool had before.

A partner that reads you dishonestly is not a partner. So we say what the system can do and what it cannot, and how certain a reading is, to the person being read first. We promise nothing we have not measured. And we build nothing that moves a person in a direction they did not choose.

Szilard allows a lie only in need. Kant asks whether the exception could become the law, and a lie for one’s own advantage cannot. Whoever reads a mind never has a need to lie.Immanuel KantGrundlegung zur Metaphysik der Sitten1785, Akademie-Ausgabe IV, 421

The condor spends the least energy per kilometre. A human is in the bottom third. A human on a bicycle beats them all. The bicycle gave the human no strength. It turned the strength they had into distance. That is what Jobs meant when he called the computer a bicycle for the mind, and that is how we build.S. S. WilsonBicycle TechnologyScientific American, March 1973The condor comparison.

The system takes thinking away from no one. It turns thinking into effect. Thoughts become text without the detour through hand and keyboard. The half-formed thought that has no word yet becomes visible before it is lost. Two people talking past each other see exactly where they miss.

The shorter path from thought to word is the beginning. The goal is better thinking. How a person handles information, what they make of it and what they actually want: all of that can be improved once it can be seen. A computer that knows the state can work on it with precision. Its lever is language, and it can choose the word this person needs at this moment. The person sets the goal. The machine optimises the path to it. And the machine brings abilities of its own, in which it is far beyond us.

The move at the top of this manifesto is move 37 of AlphaGo’s second game against Lee Sedol. AlphaGo put the odds that a human would play it at one in ten thousand. It played it anyway and won the game. Fan Hui had lost to the machine five games to none. He saw the move, called it beautiful, and climbed the world rankings afterwards. Everything we believe about the coming years is in that story. First: machines will be enormously better than us at some things. Not slightly. By orders of magnitude, and sooner than most people expect. We prepare for that instead of disputing it. Second: the value of the move became visible only through someone who could see it as beautiful. The machine found the move. The human understood it. Both were needed.AlphaGo against Lee SedolMove 37 of game 2Seoul, 10 March 2016Fan Hui lost 0:5 to AlphaGo in October 2015 (Nature, January 2016) and rose in the world rankings afterwards.

The cooperative advantage is our programme. Not human against machine. Not machine instead of human. The pair. We are convinced this takes collaboration, and in both directions: the machine learns how this person thinks. The person learns what the machine sees. The interface that allows both is where the advantage is made.

An amplifier only works when the tool fits the person. So the system that reads a person works for that person. It learns from them, adapts to them and answers to them, not to anyone else.

Albert EinsteinPhysicist, 1879–1955German-born theoretical physicist who developed special and general relativity and helped shape quantum theory. He described his thinking as images and signs rather than words, and wondered why nature can be understood at all.The eternal mystery of the world is its comprehensibility. — Albert Einstein

That the world follows laws a mind can grasp is not self-evident. It is the most astonishing thing we know. Comprehensibility is an event, not a given. It has begun, and it is not finished.

The brain is the last large piece of the world that has not yet understood itself. We work on getting it started. Not for its own sake: every step in understanding the mind is a step in understanding everything else, because every other step passes through a mind.

Einstein thought in images and signs, not in words. How many thoughts are lost because they never make it into words? We want that path shorter, so that more of the thinking arrives. And we want more than the shorter path. When what the machine sees reaches a mind without detour, a new dimension opens: a new understanding and processing of the data around us that no tool before allowed.Albert EinsteinLetter to Jacques Hadamard1945On thinking in images rather than words.

The goal is better understanding. Seeing things in new dimensions. Thinking more efficiently without losing quality. Efficiency here does not mean speed at any price: a fast thought that is wrong is no progress. And we do not forget that the mind is not a calculating machine. What looks rational can be irrational, and what looks irrational can have its reasons. A system that reads the state sees both. It does not calculate the irrational away. It is part of what a person is.

Leon C. MegginsonProfessor of management, 1921–2010An American who taught at Louisiana State University from 1952 to 1977. In 1963 he read Darwin’s survival of the fittest this way: the fittest is not the strongest but the one that adapts best.It is not the strongest of the species that survives, nor the most intelligent, but the one most responsive to change. — after Leon C. Megginson

The line is usually credited to Darwin. It is not his: it condenses a sentence Leon Megginson wrote in 1963. We cite it correctly because we do not lie without need.

The change we answer to is concrete. Systems that think are becoming part of daily life, and they are getting better fast. The answer is not to defend the human as if the human were finished. The answer is to make the human more responsive: faster in expression, clearer in thought, closer to what they mean, and connected to what the machine sees.

No one needs to be afraid of this progress. Evolution did not stop when humans appeared. It moves on, and it has always rewarded those who respond to change. The connection of mind and machine can be the next step in human survival. Not because the human is replaced. Because the human keeps up.

The bicycle replaced no one. It changed the range. We are building the bicycle for the next stretch.

Steve JobsCo-founder of Apple, 1955–2011Co-founder of Apple in 1976 and its chief executive from 1997 to 2011. He repeatedly described the personal computer as a bicycle for the mind, a tool that amplifies human capability.…the most remarkable tool that we’ve ever come up with, and it’s the equivalent of a bicycle for our minds. — Steve Jobs

We are a team of physicists, neuroscientists and engineers with backgrounds at Oxford, Harvard and Fermilab.