sciencetech-blog

comments

Elon Musk recently likened artificial intelligence to 'summoning the demon'.

The SpaceX founder, along with other scientists such Stephen Hawking, are concerned by the rapid pace of progress in machine intelligence.

But computers may not be as clever as we believe. In fact, a study in the US suggests that artificial intelligence could be fooled by simple optical illusions.

Computer scientists from the University of Wyoming and Cornell University in New York were able to hack the way a computer views objects using unique images that appear as static to humans.  In these experiments, the machine was almost completely convinced it had labelled these images correctly

The findings have wide implications, because they mean hackers could someday exploit machines that rely on their ability to recognise their surroundings.

Computer scientists from the University of Wyoming and Cornell University developed a range of images that are unrecognisable to humans, but meaningful to computers.

'It is easy to produce images that are completely unrecognisable to humans, but that state-of-the-art [deep neural networks] believe to be recognisable objects,' the team wrote in a paper posted to ArXiv.

The team used one of these deep neural network (DNN), known as AlexNet, to test their theory. 

AlexNet was created by researchers at the University of Toronto.

Pictured is another version of the optical illusion created by computer scientists. 'It is easy to produce images that are completely unrecognisable to humans, but that state-of-the-art [deep neural networks] believe to be recognisable objects,' the team wrote in a paper posted to ArXiv

They found that a computer's DNN can often say, with a high degree of confidence, that static - like that seen on older TVs - represents, for instance, a cat or a starfish. 

According to a report by Adrienne LaFrance in The Atlantic, these optical illusions exploit the particular way that computers 'see', in the same way as optical illusions can trick humans.

And the confusion is caused by the differences in how computers and humans perceive the world around them.

Humans identify objects by drawing relationships between different features, whereas a DNN looks at parts of an image on a pixel level that distinguish it from other areas.

This means that an image that looks static to a human could clearly appear to be a cat to a computer.

'One way to think about it is this,' co-author Jeff Clune said. 'These DNNs are fans of cubist art.

The findings could have major implications for security when it comes to AI. For instance, people could use these optical illusions to communicate secretly via machines by hiding messages in images

'They want to see an eye, a nose, and a mouth in the image to call it a face, but they don't particularly care where those things are. 

'The mouth can be above the eyes and to the left of the nose.'

This could have major implications for security when it comes to AI. For instance, people could communicate secretly via machines by hiding messages in images.

It could also leave computers vulnerable to hackers who could easily exploit their vision systems.

While machines may be fooled by optical illusions now, humans may not always be able to find ways to trick them. 

Earlier this month, physicist Stephen Hawking said that our desire to create intelligent could bring about our demise.

'Looking further ahead, there are no fundamental limits to what can be achieved,' continued Professor Hawking.

'There is no physical law precluding particles from being organised in ways that perform even more advanced computations than the arrangements of particles in human brains.'

ELON MUSK COMPARES AI TO 'SUMMONING THE DEMON' 

Earlier this year, Elon Musk (pictured) similarly likened artificial intelligence to 'summoning the demon'.The Tesla and Space X founder previously warned that the technology could someday be more harmful than nuclear weapons

Earlier this year, Elon Musk likened artificial intelligence to 'summoning the demon'.

The Tesla and Space X founder previously warned that the technology could someday be more harmful than nuclear weapons.

And speaking at the Massachusetts Institute of Technology (MIT) AeroAstro Centennial Symposium in October, Musk described artificial intelligence as our 'biggest existential threat'.  

He said: 'I think we should be very careful about artificial intelligence. If I had to guess at what our biggest existential threat is, it's probably that. So we need to be very careful with artificial intelligence.

'I'm increasingly inclined to think that there should be some regulatory oversight, maybe at the national and international level, just to make sure that we don't do something very foolish.

'With artificial intelligence we're summoning the demon. You know those stories where there's the guy with the pentagram, and the holy water, and … he's sure he can control the demon? Doesn't work out.'

In August, he warned that AI could to do more harm than nuclear weapons.

Tweeting a recommendation for a book by Nick Bostrom called Superintelligence: Paths, Dangers, Strategies that looks at a robot uprising, he wrote: 'We need to be super careful with AI. Potentially more dangerous than nukes.'  

In a later tweet, Musk wrote: 'Hope we're not just the biological boot loader for digital superintelligence. Unfortunately, that is increasingly probable.'

Mr Musk has previously claimed that a horrific 'Terminator-like' scenario could be created from research into artificial intelligence.

The 42-year-old is so worried, he is investing in AI companies, not to make money, but to keep an eye on the technology in case it gets out of hand.

In March, Mr Musk made an investment San Francisco-based AI group Vicarious, along with Mark Zuckerberg and actor Ashton Kutcher. 

In October 2013, the company announced it had developed an algorithm that 'reliably' solves modern Captchas - the world's most widely used test of a machine's ability to act human.