The fault lies not in our AI but in us
Are you one of those people who worry that superintelligent robots will one day wipe out or enslave humanity? If so, you are in good company: Elon Musk, Bill Gates and the late Stephen Hawking have sounded this alarm.
But consider for a moment what will happen if we manage to tame the wonder-machines we or our descendants will almost certainly build.
They will be far smarter than us, so it would be stupid not to let them make all the important decisions. They will master all branches of knowledge including how to enhance their own intelligence so there would be no point in people dedicating their efforts to learning any more. They would supply all our physical needs, so why wouldn’t we sit back and indulge ourselves?
We may survive our own most powerful creations but end up the equivalent of wildlife in a safari park, overseen by a benign intelligence whose own goals we could never hope to fathom.
This version of the future of artificial intelligence (AI) is conjured up by Stuart Russell in
a thoughtprovoking and readable account of the past, present and future of AI. A professor of computer science at the University of California, Berkeley, Russell is grounded in the realities of the technology including its many limitations, and isn’t one to jump at overheated language. But through an explanation of today’s AI techniques, and by sketching out the breakthroughs that are still needed, he makes the path to superintelligence seem all too real.
is intended mainly as an answer to what is known in AI circles as “the control problem”: how to make sure AI always remains under human guidance and working for human benefit. Along the way, the book becomes a tour d’horizon of the field. If you are looking for a serious overview to the subject that doesn’t talk down to its non-technical readers, this is a good place to start.
The problem with keeping control of AI, according to Russell, lies in the way we build all our machines. They are optimised for a given purpose, which is built or programmed in from the start. This is fine as long as the machines’ powers are limited. But when they reach superhuman levels of intelligence, they will be able to work out for themselves how to achieve their ends. We may not like the results.
No matter how carefully the objectives are set, the world is too complex, and the ingenuity of the machines will be too great, for us to predict or control the outcome. Russell’s answer: we should never give the machines a firm goal but force them instead to constantly search to understand our preferences, adjusting their goals along the way.
What, though, do people really want? This may end up being the hardest question of all. We spend much of our lives floundering in the dark. Even when we aren’t beset by pride and envy, or distracted by emotion, we can never see far enough ahead to tell if our supposedly rational actions are the right ones to achieve our desires. The problem, in fact, lies not in the machines, but in us.
Russell succeeds in making a complex subject intelligible without falling into the sort of crass popularisation that often infects books on important tech topics. He deploys a bracing intellectual rigour, such as when he explodes the many arguments that have been made for why AI couldn’t pose a threat to humans. But a laconic style and dry humour keep his book accessible to the lay reader.
At heart, Russell is an optimist. Shaping our ultimate technology, he says, will mean drawing on psychology, economics, political theory and moral philosophy. What is the best way, for instance, to maximise human happiness? Would it involve engineering a small population of exceptionally contented souls, or a much bigger human race of the only mildly happy? And how should a superintelligent machine balance the selfish interests of its owner against an altruistic regard for humanity?
Problems like these have been the stuff of moral philosophy for centuries. With the advent of superhuman AI capable of drastically reordering society, we will have to come up with some very real answers or suffer the consequences.
It is in dealing with the collision between technology on the one hand, and human and social values on the other, that Russell is at his most effective. His insistence on making the technology subservient to the human at all times may sound obvious but it is a guiding philosophy that the biggest tech companies seem, all too often, to have forgotten. /©
WE MAY SURVIVE OUR OWN MOST POWERFUL CREATIONS BUT END UP THE EQUIVALENT OF WILDLIFE IN A SAFARI PARK