Artificial intelligence: the move to science non-fiction

By Eloise Godden

‘A droid, concealed in society with a human exterior, wreaked havoc on a shopping mall in San Francisco. With machine guns enclosed in its arms, the robot ruthlessly shot at civilians, calling for the uprising of other machines in a robotic revolution. The secret services have enlisted 8,000 agents to deal with the threat… but is it enough?’ This synopsis would befit a Hollywood blockbuster, but recent advances in artificial intelligence (AI) make us wonder… how far is this from reality?

Recently, Jonas Degarve, a graduate student at Ghent University in Belgium, used a neural network (a brain-like computer) to create pseudo-consciousness. This neural network feeds on the plethora of data on the internet to link, interpret and learn. When given the phrase “brain coral”, the computer imagines a series of colourful, electric, squid-like tubes as if it was daydreaming. The distinction is that it can only imagine when instructed to imagine, not of its own accord. At some point in the future, a robot could decide to ‘think’ for itself. A super-intelligent, self-improving AI could become more powerful than humans, and thus self-sufficient, with no guarantee that it will adhere to human morality – the beginning of ‘malevolent AI’. Even if created in an ethical framework, the lack of ‘common sense’ evolved from biological roots would render this framework useless. Rather than turning on humans, as depicted in films, a malevolent AI takeover would probably be more subtle. For example, to maximise efficiency, it is possible that robots could redirect electricity away from services that matter to humans.

Images generated by Jonas Degrave's neural network when instructed to imagine (clockwise) "volcano", "broccoli", "gondolas" and "fox".
Images generated by Jonas Degrave’s neural network when instructed to imagine (clockwise): “volcano”, “broccoli”, “gondolas” and “fox”.

According to the AI expert community, a list of specific traits are required for intelligence. Goals like reasoning, perception and learning have been established. Planning ahead, creativity and social intelligence are harder to pinpoint. Machines have the capacity to simulate all of these skills, and many more. But humans have empathy, compassion and instinct. The origin of these concepts is unknown to us, – whether embedded in our DNA, from ethical and emotional systems in society or anything else – a fact that suggests AI could never recreate a human state. Does this indicate that self-sufficient machines are more likely to become malevolent than benevolent?

Even without being humanised, AI poses a real threat to jobs. A recent study from the University of Oxford concludes that 47% of jobs are at high risk of being computerised in the next couple of decades. Routine and manual jobs are the most vulnerable, but with the advancing dexterity, perception and planning machines are capable of, more unstructured jobs are at risk too. Not only is AI a cheap alternative to human labour (and as with all technology, is likely to become smaller and cheaper in the future), but robots could also be more efficient and exceed human capabilities.

Television programmes like Channel 4’s drama Humans explore the implications of androids living alongside people. With a notable 6.1 million viewers, the show’s popularity seems to reflect the public’s curiosity in AI. Even influential figures in the world of science are expressing concerns over AI progression. Stephen Hawking, Elon Musk and Steve Wozniak have signed a letter, along with 16,000 other signatories, opposing the use of autonomous weapons. Although many appreciate the benefits, such as increased precision leading to fewer civilian casualties, much of the AI expert community is concerned objectives have veered away from benefitting humankind. Moreover, on the prospect of an AI arms race, Joseph Weizenbaum argued in 1976 that AI should never replace professions that require empathy, using a soldier as one of his examples, years before AI weapons were even a prospect. As of now, it is illegal, although feasible, to use AI in warfare. At a recent UN conference the UK opposed a ban on the development of automated weapons.

Neural networks are already employed in speech recognition software in smartphones. AI is used in automated call centres. Google has created the first self-driving car. AI technology is already integrated in our lives, and much of it to our benefit. A robot uprising would not follow Hollywood’s ideas and would probably be insidious instead. However, the rapid development in this field requires careful regulation to ensure the positive use of AI and a cautionary approach to machines that can truly think for themselves.

Images: Jonas Degrave’s neural network