Why should we fear a future where robots are smarter than us?
Mechanical minds will simply add to the rich tapestry of civilisation, J. Storrs Hall believes.
Sir Isaac Newton had a stroke of genius that led him to the laws of gravity. But Newton himself reminded us that he stood on the shoulders of giants – those of Kepler, Brahe, Galileo, and Copernicus to name but a very few. It took lifetimes of painstaking work to put together enough of the puzzle of knowledge for the new piece, the law of gravitation, to fit on to its expanding edge.
We appear to be on the verge of creating mechanical minds. Will they be stolidly robotic toilers or so brilliant as to drive the human race into relative uselessness? Slaves or our overlords – no one really knows. A recent Pew Research poll of futurists, economists and industrial analysts came back split 50-50 on the issue. Perhaps we are asking the wrong question.
A personal anecdote may be useful. In the 1990s I was associated with a major artificial intelligence (AI) project at Rutgers University. We were using sophisticated computational fluid dynamics codes – the same ones used by aircraft engineers – for an aircraft design program. Our program would propose a design, run the code to analyse it, and compare the result with other proposed designs.
To our great surprise, on the very first run our program reported it had found a design for an airliner that was 25% more efficient than any human engineer had ever invented. When we investigated, it turned out that the AI program had hit on an aeroplane design with “negative wing lengths” – something that cannot exist in the physical world.
Unlike a person, the program couldn’t see the foolishness of its design and had tried to extrapolate outside the envelope of its experience. And so the vast majority of our time and effort on that project was spent encoding just enough common sense that it wouldn’t make such idiotic mistakes again – 98% of intelligence is experience and common sense, colouring inside the edges of the envelope. Extrapolation is perilous.
How then are we to predict a future filled with intelligent machines?
Physicist Leo Szilard invented the concept of a nuclear chain reaction when he stepped off the kerb in Southampton Row on a foggy London morning in 1933. The basic mechanism came to him in a flash. He not only put together some old ideas, but set many new ones in motion. Like the neutrons of his conception, one idea triggered another in a chain reaction. Within a decade a working atomic pile – a carefully constructed environment in which neutrons could be fruitful and multiply – was built.
We already know how to build a superhuman intellect and have done so for 10,000 years – since the dawn of civilisation. You simply take a bunch of human-level intellects and put them in a box with enhanced communication and memory. We call the boxes “cities” and the memory “writing”. Carrying forward the metaphor, cities are places where ideas can undergo chain reactions. More recently we have tried variants ranging from universities and corporations to the internet, but the basic idea is the same – to create an environment that triggers chain reactions of ideas in human minds.
Negative wings might have been a brilliant stroke of genius had they worked. Genius is often a shot in the dark that happens to hit something. For all Szilard knew at the time there might not have been an element that would support a nuclear chain reaction and his idea would have fizzled. (Eventually, after a decade of work by hundreds of other physicists, it was found that uranium would do the trick). And consider how many flying machines failed before the Wright brothers. Experimentation is a crucial component of the advance of knowledge. Put enough thinkers in a box with no connection to the real world and you get an ivory tower – a superheated Twitter war – not genius.
One can make a superhuman intelligence by giving a human a pencil and paper (try multiplying two 10-digit numbers without). But the pencil, the paper, the multiplication, the digits, and indeed numbers, are ideas we have accumulated over the course of civilisation. They form tools – mental and physical – that make us more intelligent than we otherwise would be. If we have been building superhuman intellects and enhancing our own minds for all of recorded history, what is this new thing that is supposed to appear in the coming decades?
If it happens the way many of us expect, it will be machines that fit into the intellectual ecosystem at the human level. This has not happened yet by any means - and may not for a few decades. Virtually all AI today has to be provided with common sense – as did our aeroplane program – by humans. Only when a machine is able to learn for itself, extend its own knowledge, and give its common sense a grounding in real-world experience, will artificial intelligence be equal to a human’s.
So if the practical operation of human intelligence is the result of 10,000 years’ worth of painstakingly accumulated ideas, then it becomes clear that our new thinking machines will essentially be people. They will be an extension of, and ultimately important contributors to, our culture.
Each robot that creates a new idea will rely on the human legacy to do so and will contribute the result back as a tiny increment to the overall wealth of knowledge. IBM’s Watson program could only win Jeopardy by studying huge volumes of history, science, and pop culture trivia. It is currently studying (human) medical knowledge and, if all goes well, will contribute valuable diagnoses based on that knowledge. Industry analyst Gartner predicts that Watson will contribute 10% of IBM’s income by the end of the decade. This will not be subtracted from IBM’s current value, but added to it. The analogy to smart machines and the rest of the human endeavour is clear – thinking machines will add to our economies and well-being.
As individuals, civilisation has bequeathed us vastly more knowledge than any of one of us could have discovered alone. As more and more people have contributed, each of us has fared better, not worse. Each person is now a vanishingly small part of the overall medium of thought. What does it matter if part of the expanding whole is mechanical?
A fairly conservative projection of trends along the lines of Moore’s Law tells us that if useful ideas continue growing as they have for the past century, remarkable things are in prospect. Our fortunes rise as the foundation of our civilisation grows. So let’s let it grow.