Monday 24 August 2015

Artificial Intelligence: is it humanity's greatest 'existential risk'?

Continuing on from an earlier post which looked at 'existential risk' - and which included an interview with Arnold Schwarzenegger
Futures Forum: Climate change: is it humanity's greatest 'existential risk'?

... here is a piece from the Wikipedia entry for the Centre for the Study of Existential Risk at the University of Cambridge which opened that earlier post:

Among the global catastrophic risks to be studied by CSER are those stemming from possible future advances in artificial intelligence. The potential dangers of artificial general intelligence have been highlighted in early discussions of CSER, being likened in some press coverage to that of a robot uprising à la The Terminator.[5][6] Price explained to the AFP news agency, "It seems a reasonable prediction that some time in this or the next century intelligence will escape from the constraints of biology". He added that when this happens "we're no longer the smartest things around," and will risk being at the mercy of "machines that are not malicious, but machines whose interests don't include us".[7]

Centre for the Study of Existential Risk - Wikipedia, the free encyclopedia

The CSER has been very much in the mainstream media of late:
Sean Holden on Radio 4’s ‘Front Row’ this evening | CSER
Martin Rees: Can we prevent the end of the world? | TED Talk | TED.com

The Future of Humanity Institute at Oxford University is another leading body in the field:
Future of Humanity Institute - Wikipedia, the free encyclopedia
Toby Ord on the likelihood of natural and anthropogenic existential risks | Future of Humanity Institute

Nick Bostrom is a leading scientist at the Institute:
existential risks: threats to humanity's survival
Nick Bostrom's Home Page

His latest topped the FT summer reading list:

June 27, 2014 4:45 pm

Summer reading: Science

Superintelligence: Paths, Dangers, Strategies, by Nick Bostrom, Oxford University Press, RRP£18.99/$29.95
Bostrom, founding director of Oxford’s Future of Humanity Institute, casts a philosopher’s eye at the past, present and future of artificial intelligence. He believes that AI research could lead to runaway superintelligence that might threaten the very existence of human civilisation – and he outlines precautions we could take.

Summer reading: Science - FT.com
Superintelligence: Paths, Dangers, Strategies, by Nick Bostrom - FT.com
Superintelligence: paths, dangers, strategies - YouTube

The end of humanity: Nick Bostrom at TEDxOxford - YouTube
existential risks: threats to humanity's survival

What Happens When Machines Surpass Human Intelligence?

Whilst the subject of AI has been the mainstay of movies and of fiction - much of it very intelligent
Genocide of One: A Thriller: Kazuaki Takano, Philip Gabriel: 9780316226226: Amazon.com: Books

... it is being taken very seriously by the world of business
The dawn of artificial intelligence | The Economist
Killer robots, AI and the path to stranded IT assets | FT Alphaville

... and by governments such as the UK, which does not seem too keen to disallow 'autonomous weapons':
'Killer robots' with AI must be banned, urge Stephen Hawking, Noam Chomsky and thousands of others in open letter - News - Gadgets and Tech - The Independent

This is a particularly good summary of the issues:

Alarm over artificial intelligence 'arms race'


Fear of a future featuring self-aware machines is rising as artificially intelligent technology goes mainstream, writes Peter Munro.
Of all the reasons robots might rise up and destroy humanity, making paper clips is not an obvious one. But the popular theory goes that an artificially intelligent machine programmed to produce as many paper clips as possible, might one day decide to do away with its makers, lest they try to stop it from achieving its aim. Ultimately, the entire planet could be stripped of whatever resources the relentless robot needs to build the biggest pile of paper clips imaginable – not that anyone would be around to use them.  
Never mind an Arnold Schwarzenegger-style Terminator extinguishing the human race: Stationery could apparently bring on the robot-led apocalypse.
'They don't have their own desires. They do what they are programmed to do.'  
Professor Toby Walsh
Even a seemingly benign instruction to an artificial intelligence (AI) could inadvertently bring about our demise. Oxford academic Stuart Armstrong, from the Future of Humanity Institute, recently suggested that a supercomputer tasked to "prevent human suffering" could decide with lethal logic to "kill all humans" and so end our suffering altogether. Or that a mandate to "keep humans safe and happy" might be translated by a machine as "entomb everyone in concrete coffins on heroin drips".
An Arnold Schwarzenegger-style Terminator destroying humanity is an extreme scenario.
An Arnold Schwarzenegger-style Terminator destroying humanity is an extreme scenario.
Such doomsday scenarios might seem science fiction. But AI is suddenly everywhere, from personal assistants embedded in smartphones to driverless cars – an on-road trial of which will start in South Australia in November.
Super-intelligent technology has gone mainstream,  fuelled by an "AI arms race" among the likes of Google, Microsoft, Apple and Baidu. Google co-founder Sergey​ Brin​ has said he wants to build the equivalent of the sentient computer HAL from the film 2001: A Space Odyssey – except one that wouldn't kill people. In May, the company claimed it was on the brink of developing algorithms with the capacity for common sense, natural conversation and even flirtation. 
Super-smart machines already prescribe personalised treatments for cancer patients, write sports reports or beat human players in Atari video gamesBut despite the positive potential of AI – to help eradicate poverty, disease or the drudgery of daily work – fear of a future featuring self-aware machines looms large.   
An X-47B pilot-less drone combat aircraft is launched for the first time off an aircraft carrier, the USS George H. W. Bush, in 2013.
An X-47B pilot-less drone combat aircraft is launched for the first time off an aircraft carrier, the USS George H. W. Bush, in 2013. Photo: JASON REED
Theoretical physicist Stephen Hawking has  warned that "the development of full artificial intelligence could spell the end of the human race". He predicts computers will overtake human intelligence within the next century, rendering us obsolete.
Billionaire Elon​ Musk, founder of SpaceX and Tesla Motors, instead likens AI to "summoning the demon". This year, he donated $US10 million ($13.7 million) towards research into keeping AI safe, saying he struggled to sleep for fear Google would accidentally unleash an army of robots with the ability to annihilate the human race.
The prospect of killer robots was highlighted last week by more than 1000 AI experts and researchers – including Hawking and Musk – who warned of an emerging "military artificial intelligence arms race". Their joint letter, presented at the International Joint Conference on Artificial Intelligence in Buenos Aires, said weapons that select and engage targets without human intervention were feasible within years.
Alfred Boyadgis, co-founder of Forcite Helmet Systems, wearing a prototype of their AI ski helmet. The helmet will communicate with its surrounds and the internet to enhance the ski experience for the wearer.
Alfred Boyadgis, co-founder of Forcite Helmet Systems, wearing a prototype of their AI ski helmet. The helmet will communicate with its surrounds and the internet to enhance the ski experience for the wearer. Photo: Tony Walters
Also among the signatories was Toby Walsh, professor of AI at the University of NSW and the NICTA research organisation, who says "offensive autonomous weapons" will lower the threshold for waging war. "It's a technology that's going to allow presidents and prime ministers to think they can slip into battle without bodies coming home in bags," he tells Fairfax Media.
"It's not likely to make the world a better place – it's likely to escalate conflict and to cause more damage to humans."
The US military is already developing autonomous aerial vehicles capable of carrying out all steps of a strike mission, without a human operator, according to a May report in Nature magazine. 
Theoretical physicist Stephen Hawking has warned that "the development of full artificial intelligence could spell the end of the human race".
Theoretical physicist Stephen Hawking has warned that "the development of full artificial intelligence could spell the end of the human race". Photo: Jemal Countess
Professor Walsh says AI should be dedicated towards tackling pressing problems such as inequality and poverty, or the rising cost of healthcare. But such technology can also be used to inflict unnecessary harm. "Certainly, it would help assuage fears about killer robots if we didn't have killer robots," he says.
Potential danger closer
Machines already kill people, though in more mundane ways than in the movies. A 22-year-old worker in a German car factory died this month after being trapped by a robotic arm and crushed against a metal plate. The robot that killed him was part of an automated assembly line that can function without a human operator – but initial reports suggest human error was to blame.
Elon Musk  likens AI to "summoning the demon". This year, he donated $US10 million ($13.7 million) towards research into keeping AI safe.
Elon Musk likens AI to "summoning the demon". This year, he donated $US10 million ($13.7 million) towards research into keeping AI safe. Photo: Jason Merritt
In the United States, robots have caused at least 33 workplace deaths and injuries in the past 30 years, according to The New York Times.
But it's the potential emergence of super-intelligent and self-aware machines that sparks greater alarm. A recent Oxford University report found almost one in five AI experts believe such technology will ultimately pose an "existential threat" to humanity. According to the 550 people surveyed, machines will be able to work in most professions at least as well as humans between 2040 and 2075. By the end of this century, many believe AI is likely to greatly exceed the cognitive performance of humans across the board. 
But Professor Walsh says this is not necessarily a bad thing. "We will be superseded. I don't see any reason to suppose our intelligence is the peak intelligence you will find in the universe and that we won't be able to make computers smarter or better than us," he says.
Apple co-founder Steve Wozniak, whose company has helped AI technology to become mainstream.
Apple co-founder Steve Wozniak, whose company has helped AI technology to become mainstream. Photo: Dominic Lorrimer
AI could help combat global warming or bring First World healthcare to impoverished countries, delivering doctors to patients via smartphones, he says. People should be more worried about the global financial crisis or changing labour market than about super-intelligent, self-aware machines, he adds. "I'm willing to accept it is possible computers will become smart enough that an emergent form of consciousness arises within them. But at the moment they are very far from being sentient.
"They don't have their own desires. They do what they are programmed to do. Your accountancy program doesn't wake up and say 'I'm bored of adding up numbers, I'm going to become a journalist today'."
Computer sentience is a common theme in science fiction, from the surprisingly romantic filmHer, where a man falls in love with his operating system, through to 2014's Transcendence – where Johnny Depp's character uploads his own consciousness online to live beyond death, in the form of a fully-fledged sentient computer.
That hypothetical moment is called the "singularity". It's considered to be the moment where we have an AI that not only thinks independently but is capable of creating new and improved versions of itself, evolving at an accelerating rate of growth.
Beyond that point, we don't know what might happen, says tech chief Nick Howden. His Melbourne-based company Real Thing has created a mobile device for people with poor vision, which operates in a similar fashion to "intelligent assistants" such as Google Now, Apple's Siri or Microsoft's Cortana. 
He reckons self-aware machines will emerge within the next 50 years. "That moment is going to change the world, one way or another," he says. "There are lots of scenarios where it is benevolent and looks upon us kindly and starts looking to improve our lives and cure cancer. 
"But if that entity is created with the will to destroy – if it's a military drone or something like that – you can end up with a disaster that might wipe us from the face of the earth. I like to think that if you create something intelligent enough, it will think through the meaning of life and conclude that 'I should be good'. But you had better make sure the first one you create is friendly, because if it's not you're in trouble."
Ethical issues need resolving
Never mind paper clips. What will happen when the first self-driving car kills a pedestrian? 
It's the type of ethical dilemma that has long bedevilled humans. An autonomous vehicle has to decide between crashing into a person with a pram crossing the road, thereby killing a parent and child, or avoiding them by steering into a wall and killing their passenger instead. The legal issues alone are overwhelming. Who should be sued? The robot manufacturers? Computer programmer? Car company? 
Stephen Hawking has warned that success in creating AI would be "the biggest event in human history". But it might also be the last, unless we start considering such risks. "In the short term, people are concerned about who controls AI, but in the long term the concern will be whether AI can be controlled at all," he said at the Zeitgeist 2015 conference in London, in May.
Autonomous cars are already on the road in California – equipped, for now at least, with a safety driver, brakes and steering wheel. Google says its self-driving vehicles will significantly reduce the 94 per cent of accidents caused by human error, along with liberating people from unproductive hours spent in traffic and improving the mobility of those who can't drive. So what's one computer-driven death against a 1000 caused by human drivers? 
The tech company is separately developing computers with personality traits – creating machines capable of happiness, fear, thoughtfulness or derision. 
Applied AI expert Mustafa​ Suleyman​ – whose start-up company DeepMind was bought by Google in 2014 for $US500 million – says the "preposterous" debate over killer robots has stolen attention from such developments. Speaking at a conference on machine intelligence in June, he argued AI was "going to be a hugely powerful tool that we control and that we direct, whose capabilities we limit," comparing it to a washing machine or tractor. 
"We're building them to empower humanity and not to destroy us."
But what are the chances such an entity would remain content to take directions, asks American neuroscientist Sam Harris. Writing on his blog, in May, he suggested a super-intelligent AI will need to be instilled with values commensurate with our own. "But whose values should count?," he asks. "Should everyone get a vote in creating the utility function of our new colossus?"
Oxford University philosopher Nick Bostrom​, in his 2014 book Superintelligence: Paths, Dangers, Strategies, suggests an AI could instead be created with only limited ambitions, or programmed to be super-intelligent and benevolent – despite the difficulties of expressing such values in computer code. 
A final possibility, Bostrom suggests, is telling an AI to figure out for itself how to be safe.  
Professor Walsh says scientists are already trying to work out ways to program ethics into a computer. But he questions whether we're capable of choosing a clear moral code for machines to follow. Humanity has spent millennia debating ethical issues without yet devising a simple set of instructions, even for such basic concepts as gender equality. "If Sony are going to be making robots to sell around the world, do they have different sets of ethics for different parts of the world?" he asks.
"We don't have the answers yet but we are going to have to come up with them because computers will be putting themselves in situations where they will be needing to make ethical decisions – life-and-death decisions. Our whole legal framework is going to have to change, because there's going to be a new type of entity out there."
He hopes that an artificially intelligent machine might ultimately learn to do the right thing, in the same way we do – with mixed success, admittedly. "We have people from all around the world coming to Australia and we are able to work in a cohesive way in society, so there is no reason to suppose we can't throw robots into the same mix," he says. 
"Maybe we will hold them to higher ethical standards than we do for humans, which will be a good thing."
Pace of development increasing 
Artificial intelligence won't wipe out humanity, not before 2065 at least. "I think it's a bit dramatic to say that it's going to kill everyone," says Alfred Boyadgis. "I don't think it's going to kill anyone at all, at least not for the next 50 years or so."
So we have 50 years left to live? "Yeah," he says, laughing. Boyadgis is the co-creator of a smart snow helmet, which comes equipped with an artificially intelligent personal assistant. She can tell the wearer the weather and snow conditions, pinpoint his friends on the slopes or coach him to become a better skier.
She's called "Vikki", after the similarly-named supercomputer in the film I, Robot, which seeks to enslave all of humanity. "We're theatrical sci-fi fanatics here, so we picked something fun," Boyadgis says, laughing again.  
We're at the first-floor headquarters of Forcite Helmet Systems, in a shopping complex on a busy road in Alexandria, in Sydney's inner west. Six men with laptops sit silently in a small room. There's an empty Tim Tam packet on the table.
The rapid development of artificial intelligence (AI) is taking place in thousands of similarly inauspicious offices around the world – aided by the spread of big data and technology.
Forcite Helmet Systems' first prototype was a motorcycle helmet for police. The tech company is now developing a smart helmet for emergency service officers, which will identify risks such as a fire and relay that information back to a local area command.
Other projects include a helmet for miners that can detect whether someone is using heavy machinery incorrectly, or a motorbike helmet that can alert nearby riders to an oil slick on the road.
"We use artificial intelligence in our day-to-day life and it is nothing to be afraid of at all," Boyadgis says. "It's building products such as this that will help us have more time to enjoy our sports or enjoy our driving, or be safer."
The Forcite Alpine Helmet will be released on the crowdfunding website Kickstarter in October. Boyadgis describes Vikki as a form of "conversational AI". "Something that is artificially intelligent has situational awareness and it's autonomous, so it knows what's happening around it and can make judgments," he says.
I try her on and she starts telling me the temperature and the conditions on the Kamikaze run at Blue Cow. She sounds a bit posh and stilted – but then we've only just met.
The computer within the helmet utilises satellite and online data, along with GPS and accelerometers to track the wearer's posture, speed and balance. Boyadgis says Vikki can determine the wearer's skiing ability and advise them on how best to tackle a particular slope. She's certainly smart but not self-aware – not yet, anyway.
"AI that is aware of itself and decides that humanity is obsolete is something that is a bit more fictional, but we are moving towards it," he says. "Things like that can help us cure diseases, go to other planets and develop technology. But then there is that other side, where it could potentially be very dangerous."

Alarm over artificial intelligence 'arms race'

No comments: