Astronomer Royal Martin Rees: How soon will robots take over the world?An explosion in artificial intelligence has sent us hurtling towards a post-human future, warns Martin ReesIn Davos a few years ago, I met a well-known Indian tycoon. Knowing I had the title Astronomer Royal, he asked: “Do you do the Queen’s horoscopes?” I responded, with a straight face: “If she wanted one, I’m the person she’d ask.” He then seemed eager to hear my predictions. I told him that markets would fluctuate and that there would be trouble in the Middle East. He paid rapt attention to these insights. But I then came clean. I said I was just an astronomer, not an astrologer. He immediately lost all interest in my predictions. And rightly so: scientists are rotten forecasters – worse, often, than writers of science fiction.
Nevertheless, 12 years ago, I wrote a book that I entitled Our Final Century? My publisher deleted the question-mark. The American publishers changed the title to Our Final Hour – Americans seek instant (dis)gratification. My theme was this: our Earth is 45 million centuries old, but this century is special. It’s the first when one species – ours – can determine the biosphere’s fate.
In the years since, a few forecasts have somewhat firmed up: the world is becoming more crowded – and warmer. There will be about 2 billion more people in 2050, and their collective “footprint” will threaten our finite planet’s ecology unless we can achieve more efficient use of energy and land. But we can’t predict the path of future technology that far ahead. Today’s smartphones would have seemed magic even 20 years ago, so in looking several decades ahead we must keep our minds open to breakthroughs that may now seem like science fiction. These will offer great hopes, but also great fears.
Society is more interconnected than ever, and consequently more vulnerable. We depend on elaborate networks: electric-power grids, just-in-time delivery, satnav, globally dispersed manufacturing, and so forth. Can we be sure that these networks are resilient enough to rule out catastrophic disruptions cascading through the system – real-world analogues of the 2008 financial crash? London would be instantly paralysed without electricity. Supermarket shelves would soon be bare if supply chains were disrupted. Air travel can spread a pandemic worldwide in days, causing havoc in the megacities of the developing world. And social media can amplify panic and rumour, literally at the speed of light.
The worry isn’t just accidental breakdowns. Malicious events can have catastrophic consequences. Cyber-sabotage efforts, such as “Stuxnet”, and frequent hacking of financial institutions have highlighted these concerns. Small groups – and even individuals – are more empowered than ever before.
And there are downsides to the huge advances in biotech, despite the bright prospects these offer for medicine and agriculture. There were reports last month that Chinese researchers had been gene-editing human embryos using a new technique called CRISPR, raising controversial ethical issues about “designer babies”. But more disquieting are the experiments at the University of Wisconsin and in the Netherlands that show it’s surprisingly easy to make an influenza virus more virulent and transmissible. Last October, the US federal government decided to cease funding these so-called “gain of function” experiments.
We held a debate on this recently in Cambridge. Those supporting “gain of function” research highlighted the need to study viruses in order to stay one step ahead of natural mutations. Others viewed the techniques as a scary portent of what’s to come. What would happen, for instance, if an ebola virus were modified to be transmissible through the air? And they worried that the risk of failure to contain the pathogens within the lab is too high to justify the knowledge gained.
There is a contrast here with the (also real) dangers of nuclear technology. Nuclear installations are sufficiently large-scale for bodies such as the International Atomic Energy Agency to regulate them effectively. It is hard to make a clandestine H-bomb. In contrast, millions with the capability to misuse biotech will have access to biomedical labs, just as millions can misuse cybertech today.
Indeed, biohacking is burgeoning even as a hobby and competitive game. (For instance, there is competition to develop plants that glow in the dark, and eventually make trees that could replace street lights.) The physicist Freeman Dyson foresees a time when children will be able to design and create new organisms just as routinely as his generation played with chemistry sets. I’d guess that this is comfortably beyond the “SF fringe”, but were even part of this scenario to come about, our ecology (and even our species) surely would not long survive unscathed.
Not all those with “bio” expertise will be balanced and rational. My worst nightmare is an “eco-fanatic”, empowered by the biohacking expertise that may be routine by 2050, who thinks that “Gaia” can only be saved if the human population is reduced. The global village will have its village idiots, and they will have global range.
In the early days of “molecular biology”, a group of academic scientists formulated the “Asilomar Declaration”, advocating a moratorium on certain types of experiments and setting up guidelines. Such self-policing worked back in the Seventies. But the field is now far more global, more competitive, and commercial pressures are stronger. So it’s doubtful that regulations imposed on prudential or ethical grounds could be enforced worldwide, any more than the drug laws can. So this is a real anxiety – number one in my estimation – and will raise the tension between privacy, freedom and security.
What about other future technologies — computers and robotics, for instance? There is nothing new about machines that can surpass our mental abilities in special areas. Even the pocket calculators of the Seventies could do arithmetic better than us. In the Nineties, IBM’s “Deep Blue” chess-playing computer beat Garry Kasparov, then the world champion. More recently, another IBM computer won a television game show that required wide general knowledge and the ability to respond to questions in the style of crossword clues.
We’re witnessing a momentous speed-up in artificial intelligence (AI) – in the power of machines to learn, communicate and interact with us. Computers don’t learn like we do: they use “brute force” methods. They learn to translate from foreign languages by reading multilingual versions of, for example, millions of pages of EU documents (they never get bored). They learn to recognise dogs, cats and human faces by crunching through millions of images — not the way a baby learns.
Deep Mind, a London company that Google recently bought for £400 million, created a machine that can figure out the rules of all the old Atari games without being told, and then play them better than humans.
It’s still hard for AI to interact with the everyday world. Robots remain clumsy – they can’t tie your shoelaces or cut your toenails. But sensor technology, speech recognition, information searches and so forth are advancing apace.
Google’s driverless car has already covered hundreds of thousands of miles. But can it cope with emergencies? For instance, if an obstruction suddenly appears on a busy road, can the robotic “driver” discriminate whether it’s a paper bag, a dog or a child? The likely answer is that it won’t cope as well as a really good driver, but will be better than the average driver — machine errors may occur but not as often as human error. The roads will be safer. But when accidents occur they will create a legal minefield. Who should be held responsible — the “driver”, the owner, or the designer?
And what about the military use of autonomous drones? Can they be trusted to seek out a targeted individual and decide whether to deploy their weapon? Who has the moral responsibility then?
AI will take over a wider range of jobs – not just manual work but accountancy, routine legal work, medical diagnostics and surgery. And the big question is then: will AI be like earlier disruptive technologies – the car, for instance – which created as many jobs as they destroyed? Or is it really different this time?
During this century, our society will be increasingly transformed by computers. But will they remain idiots savants or will they display near-human all-round capabilities? If robots could observe and interpret their environment as adeptly as we do, they would be perceived as intelligent beings that we could relate to. Would we then have a responsibility to them? Should we care if they are frustrated or bored? Maybe we’d have no more reason to disparage them as zombies than to regard other people in that way.
Experts disagree on how long it will take before machines achieve general-purpose human level intelligence. Some say 25 years. Others say “never”. The median guess in a recent survey was about 50 years.
Some of those with the strongest credentials think that the AI field is advancing so fast that it already needs guidelines for “responsible innovation”, just as biotech does.
And there is disagreement about the route towards human-level intelligence. Some think we should emulate nature and reverse-engineer the human brain. Others say that’s a misguided approach – like designing a flying machine by copying how birds flap their wings. But it’s clear that once a threshold is crossed, there will be an intelligence explosion. That’s because electronics is a million times faster than the transmission of signals in the brain; and because computers can network and exchange information much faster than we can by speaking.
In the Sixties, the British mathematician I J Good, who worked at Bletchley Park with Alan Turing, pointed out that a super-intelligent robot (were it sufficiently versatile) could be the last invention that humans need ever make. Once machines have surpassed human capabilities, they could themselves design and assemble a new generation of even more powerful machines — triggering a real “intelligence explosion”. Or could humans transcend biology by merging with computers, maybe losing their individuality and evolving into a common consciousness? In old-style spiritualist parlance, they would “go over to the other side”.
The most prominent evangelist for runaway super-intelligence – so-called “'singularity” – is Ray Kurzweil, now working at Google. He thinks this could happen within 25 years. But he is worried that he may not live that long. So he takes dozens of pills each day, and if he dies he wants his body frozen until this nirvana is reached.
I was once interviewed by a group of “cryonic” enthusiasts in California called the “society for the abolition of involuntary death”. They will freeze your body, so that when immortality is on offer you can be resurrected. I said I’d rather end my days in an English churchyard than a Californian refrigerator. They derided me as a “deathist”. (I was surprised later to find that three Oxford academics were cryonic enthusiasts. Two have paid full whack; a third has taken the cut-price option of just wanting his head frozen.)
Let me briefly deploy an astronomical perspective and speculate about the really far future – the post-human era. There are chemical and metabolic limits to the size and processing power of organic brains. Maybe humans are close to these limits already. But there are no such constraints on silicon-based computers (still less, perhaps, quantum computers): for these, the potential for further development could be as dramatic as the evolution from monocellular organisms to humans. So, by any definition of “thinking”, the amount and intensity that’s done by organic human-type brains will, in the far future, be utterly swamped by the cerebrations of AI. Moreover, the Earth’s biosphere in which organic life has symbiotically evolved is not a constraint for advanced AI. Indeed, it is far from optimal – interplanetary and interstellar space will be the preferred arena where robotic fabricators will have the grandest scope for construction, and where non-biological “brains” may develop insights as far beyond our imaginings as string theory is for a mouse.
Abstract thinking by biological brains has underpinned the emergence of all culture and science. But this activity – spanning tens of millennia at most – will be a brief precursor to the more powerful intellects of the inorganic post-human era. So, in the far future, it won’t be the minds of humans, but those of machines, that will most fully understand the cosmos – and it will be the actions of autonomous machines that will most drastically change our world, and perhaps what lies beyond.
Martin Rees is the Astronomer Royal.
http://www.telegraph.co.uk/culture/hay-festival/11605785/Astronomer-Royal-Martin-Rees-predicts-the-world-will-be-run-by-computers-soon.html
