The benefits for human society are likely to be immense. AI doctors could provide far better and cheaper healthcare for billions of people, particularly for those who currently receive no healthcare at all. Thanks to learning algorithms and biometric sensors, a poor villager in an underdeveloped country might come to enjoy far better healthcare via her smartphone than the richest person in the world gets today from the most advanced urban hospital.5
Similarly, self-driving vehicles could provide people with much better transport services, and in particular reduce mortality from traffic accidents. Today close to 1.25 million people are killed annually in traffic accidents (twice the number killed by war, crime and terrorism combined).6 More than 90 per cent of these accidents are caused by very human errors: somebody drinking alcohol and driving, somebody texting a message while driving, somebody falling asleep at the wheel, somebody daydreaming instead of paying attention to the road. The US National Highway Traffic Safety Administration estimated in 2012 that 31 per cent of fatal crashes in the USA involved alcohol abuse, 30 per cent involved speeding, and 21 per cent involved distracted drivers.7 Self-driving vehicles will never do any of these things. Though they suffer from their own problems and limitations, and though some accidents are inevitable, replacing all human drivers by computers is expected to reduce deaths and injuries on the road by about 90 per cent.8 In other words, switching to autonomous vehicles is likely to save the lives of a million people every year.
Hence it would be madness to block automation in fields such as transport and healthcare just in order to protect human jobs. After all, what we ultimately ought to protect is humans – not jobs. Redundant drivers and doctors will just have to find something else to do.
The Mozart in the machine
At least in the short term, AI and robotics are unlikely to completely eliminate entire industries. Jobs that require specialisation in a narrow range of routinised activities will be automated. But it will be much more difficult to replace humans with machines in less routine jobs that demand the simultaneous use of a wide range of skills, and that involve dealing with unforeseen scenarios. Take healthcare, for example. Many doctors focus almost exclusively on processing information: they absorb medical data, analyse it, and produce a diagnosis. Nurses, in contrast, also need good motor and emotional skills in order to give a painful injection, replace a bandage, or restrain a violent patient. Hence we will probably have an AI family doctor on our smartphone decades before we have a reliable nurse robot.9 The human care industry – which takes care of the sick, the young and the elderly – is likely to remain a human bastion for a long time. Indeed, as people live longer and have fewer children, care of the elderly will probably be one of the fastest-growing sectors in the human labour market.
Alongside care, creativity too poses particularly difficult hurdles for automation. We don’t need humans to sell us music any more – we can download it directly from the iTunes store – but the composers, musicians, singers and DJs are still flesh and blood. We rely on their creativity not just to produce completely new music, but also to choose among a mind-boggling range of available possibilities.
Nevertheless, in the long run no job will remain absolutely safe from automation. Even artists should be put on notice. In the modern world art is usually associated with human emotions. We tend to think that artists are channelling internal psychological forces, and that the whole purpose of art is to connect us with our emotions or to inspire in us some new feeling. Consequently, when we come to evaluate art, we tend to judge it by its emotional impact on the audience. Yet if art is defined by human emotions, what might happen once external algorithms are able to understand and manipulate human emotions better than Shakespeare, Frida Kahlo or Beyoncé?
After all, emotions are not some mystical phenomenon – they are the result of a biochemical process. Hence, in the not too distant future a machine-learning algorithm could analyse the biometric data streaming from sensors on and inside your body, determine your personality type and your changing moods, and calculate the emotional impact that a particular song – even a particular musical key – is likely to have on you.10
Of all forms of art, music is probably the most susceptible to Big Data analysis, because both inputs and outputs lend themselves to precise mathematical depiction. The inputs are the mathematical patterns of sound waves, and the outputs are the electrochemical patterns of neural storms. Within a few decades, an algorithm that goes over millions of musical experiences might learn to predict how particular inputs result in particular outputs.11
Suppose you just had a nasty fight with your boyfriend. The algorithm in charge of your sound system will immediately discern your inner emotional turmoil, and based on what it knows about you personally and about human psychology in general, it will play songs tailored to resonate with your gloom and echo your distress. These particular songs might not work well with other people, but are just perfect for your personality type. After helping you get in touch with the depths of your sadness, the algorithm would then play the one song in the world that is likely to cheer you up – perhaps because your subconscious connects it with a happy childhood memory that even you are not aware of. No human DJ could ever hope to match the skills of such an AI.
You might object that the AI would thereby kill serendipity and lock us inside a narrow musical cocoon, woven by our previous likes and dislikes. What about exploring new musical tastes and styles? No problem. You could easily adjust the algorithm to make 5 per cent of its choices completely at random, unexpectedly throwing at you a recording of an Indonesian Gamelan ensemble, a Rossini opera, or the latest K-pop hit. Over time, by monitoring your reactions, the AI could even determine the ideal level of randomness that will optimise exploration while avoiding annoyance, perhaps lowering its serendipity level to 3 per cent or raising it to 8 per cent.
Another possible objection is that it is unclear how the algorithm could establish its emotional goal. If you just fought with your boyfriend, should the algorithm aim to make you sad or joyful? Would it blindly follow a rigid scale of ‘good’ emotions and ‘bad’ emotions? Maybe there are times in life when it is good to feel sad? The same question, of course, could be directed at human musicians and DJs. Yet with an algorithm, there are many interesting solutions to this puzzle.
One option is to just leave it to the customer. You can evaluate your emotions whichever way you like, and the algorithm will follow your dictates. Whether you want to wallow in self-pity or jump for joy, the algorithm will slavishly follow your lead. Indeed, the algorithm may learn to recognise your wishes even without you being explicitly aware of them.
Alternatively, if you don’t trust yourself, you can instruct the algorithm to follow the recommendation of whichever eminent psychologist you do trust. If your boyfriend eventually dumps you, the algorithm may walk you through the official five stages of grief, first helping you deny what happened by playing Bobby McFerrin’s ‘Don’t Worry, Be Happy’, then whipping up your anger with Alanis Morissette’s ‘You Oughta Know’, encouraging you to bargain with Jacques Brel’s ‘Ne me quitte pas’ and Paul Young’s ‘Come Back and Stay’, dropping you into the pit of depression with Adele’s ‘Someone Like You’ and ‘Hello’, and finally aiding you to accept the situation with Gloria Gaynor’s ‘I Will Survive’.
The next step is for the algorithm to start tinkering with the songs and melodies themselves, changing them ever so slightly to fit your quirks. Perhaps you dislike a particular bit in an otherwise excellent song. The algorithm knows it because your heart skips a beat and your oxytocin levels drop slightly whenever you hear that annoying part. The algorithm could rewrite or edit out the offending notes.
In the long run, algorithms may learn how to compose entire tunes, playing on human emotions as if they were a piano keyboard. Using your biometric data the algorithms could even produce personalised melodies, which you alone in the entire universe would appreciate.
It is often said that p
eople connect with art because they find themselves in it. This may lead to surprising and somewhat sinister results if and when, say, Facebook begins creating personalised art based on everything it knows about you. If your boyfriend leaves you, Facebook will treat you to an individualised song about that particular bastard rather than about the unknown person who broke the heart of Adele or Alanis Morissette. The song will even remind you of real incidents from your relationship, which nobody else in the world knows about.
Of course, personalised art might never catch on, because people will continue to prefer common hits that everybody likes. How can you dance or sing together to a tune nobody besides you knows? But algorithms could prove even more adept at producing global hits than personalised rarities. By using massive biometric databases garnered from millions of people, the algorithm could know which biochemical buttons to press in order to produce a global hit which would set everybody swinging like crazy on the dance floors. If art is really about inspiring (or manipulating) human emotions, few if any human musicians will have a chance of competing with such an algorithm, because they cannot match it in understanding the chief instrument they are playing on: the human biochemical system.
Will all this result in great art? That depends on the definition of art. If beauty is indeed in the ears of the listener, and if the customer is always right, then biometric algorithms stand a chance of producing the best art in history. If art is about something deeper than human emotions, and should express a truth beyond our biochemical vibrations, biometric algorithms might not make very good artists. But nor do most humans. In order to enter the art market and displace many human composers and performers, algorithms won’t have to begin by straightaway surpassing Tchaikovsky. It will be enough if they outperform Britney Spears.
New jobs?
The loss of many traditional jobs in everything from art to healthcare will partly be offset by the creation of new human jobs. GPs who focus on diagnosing known diseases and administering familiar treatments will probably be replaced by AI doctors. But precisely because of that, there will be much more money to pay human doctors and lab assistants to do groundbreaking research and develop new medicines or surgical procedures.12
AI might help create new human jobs in another way. Instead of humans competing with AI, they could focus on servicing and leveraging AI. For example, the replacement of human pilots by drones has eliminated some jobs but created many new opportunities in maintenance, remote control, data analysis and cyber security. The US armed forces need thirty people to operate every unmanned Predator or Reaper drone flying over Syria, while analysing the resulting harvest of information occupies at least eighty people more. In 2015 the US Air Force lacked sufficient trained humans to fill all these positions, and therefore faced an ironic crisis in manning its unmanned aircraft.13
If so, the job market of 2050 might well be characterised by human–AI cooperation rather than competition. In fields ranging from policing to banking, teams of humans-plus-AIs could outperform both humans and computers. After IBM’s chess program Deep Blue beat Garry Kasparov in 1997, humans did not stop playing chess. Rather, thanks to AI trainers human chess masters became better than ever, and at least for a while human–AI teams known as ‘centaurs’ outperformed both humans and computers in chess. AI might similarly help groom the best detectives, bankers and soldiers in history.14
The problem with all such new jobs, however, is that they will probably demand high levels of expertise, and will therefore not solve the problems of unemployed unskilled labourers. Creating new human jobs might prove easier than retraining humans to actually fill these jobs. During previous waves of automation, people could usually switch from one routine low-skill job to another. In 1920 a farm worker laid off due to the mechanisation of agriculture could find a new job in a factory producing tractors. In 1980 an unemployed factory worker could start working as a cashier in a supermarket. Such occupational changes were feasible, because the move from the farm to the factory and from the factory to the supermarket required only limited retraining.
But in 2050, a cashier or textile worker losing their job to a robot will hardly be able to start working as a cancer researcher, as a drone operator, or as part of a human–AI banking team. They will not have the necessary skills. In the First World War it made sense to send millions of raw conscripts to charge machine guns and die in their thousands. Their individual skills mattered little. Today, despite the shortage of drone operators and data analysts, the US Air Force is unwilling to fill the gaps with Walmart dropouts. You wouldn’t like an inexperienced recruit to mistake an Afghan wedding party for a high-level Taliban conference.
Consequently, despite the appearance of many new human jobs, we might nevertheless witness the rise of a new ‘useless’ class. We might actually get the worst of both worlds, suffering simultaneously from high unemployment and a shortage of skilled labour. Many people might share the fate not of nineteenth-century wagon drivers – who switched to driving taxis – but of nineteenth-century horses, who were increasingly pushed out of the job market altogether.15
In addition, no remaining human job will ever be safe from the threat of future automation, because machine learning and robotics will continue to improve. A forty-year-old unemployed Walmart cashier who by dint of superhuman efforts manages to reinvent herself as a drone pilot might have to reinvent herself again ten years later, because by then the flying of drones may also have been automated. This volatility will also make it more difficult to organise unions or secure labour rights. Already today, many new jobs in advanced economies involve unprotected temporary work, freelancing and one-time gigs.16 How do you unionise a profession that mushrooms and disappears within a decade?
Similarly, human–computer centaur teams are likely to be characterised by a constant tug of war between the humans and the computers, instead of settling down to a lifelong partnership. Teams made exclusively of humans – such as Sherlock Holmes and Dr Watson – usually develop permanent hierarchies and routines that last decades. But a human detective who teams up with IBM’s Watson computer system (which became famous after winning the US TV quiz show Jeopardy! in 2011) will find that every routine is an invitation for disruption, and every hierarchy an invitation for revolution. Yesterday’s sidekick might morph into tomorrow’s superintendent, and all protocols and manuals will have to be rewritten every year.17
A closer look at the world of chess might indicate where things are heading in the long run. It is true that for several years after Deep Blue defeated Kasparov, human–computer cooperation flourished in chess. Yet in recent years computers have become so good at playing chess that their human collaborators lost their value, and might soon become utterly irrelevant.
On 7 December 2017 a critical milestone was reached, not when a computer defeated a human at chess – that’s old news – but when Google’s AlphaZero program defeated the Stockfish 8 program. Stockfish 8 was the world’s computer chess champion for 2016. It had access to centuries of accumulated human experience in chess, as well as to decades of computer experience. It was able to calculate 70 million chess positions per second. In contrast, AlphaZero performed only 80,000 such calculations per second, and its human creators never taught it any chess strategies – not even standard openings. Rather, AlphaZero used the latest machine-learning principles to self-learn chess by playing against itself. Nevertheless, out of a hundred games the novice AlphaZero played against Stockfish, AlphaZero won twenty-eight and tied seventy-two. It didn’t lose even once. Since AlphaZero learned nothing from any human, many of its winning moves and strategies seemed unconventional to human eyes. They may well be considered creative, if not downright genius.
Can you guess how long it took AlphaZero to learn chess from scratch, prepare for the match against Stockfish, and develop its genius instincts? Four hours. That’s not a typo. For centuries, chess was considered one of the crowning glories of human intelligence. AlphaZero went from utter ignorance to crea
tive mastery in four hours, without the help of any human guide.18
AlphaZero is not the only imaginative software out there. Many programs now routinely outperform human chess players not just in brute calculation, but even in ‘creativity’. In human-only chess tournaments, judges are constantly on the lookout for players who try to cheat by secretly getting help from computers. One of the ways to catch cheats is to monitor the level of originality players display. If they play an exceptionally creative move, the judges will often suspect that this cannot possibly be a human move – it must be a computer move. At least in chess, creativity is already the trademark of computers rather than humans! Hence if chess is our coal-mine canary, we are duly warned that the canary is dying. What is happening today to human–AI chess teams might happen down the road to human–AI teams in policing, medicine and banking too.19
Consequently, creating new jobs and retraining people to fill them will not be a one-off effort. The AI revolution won’t be a single watershed event after which the job market will just settle into a new equilibrium. Rather, it will be a cascade of ever-bigger disruptions. Already today few employees expect to work in the same job for their entire life.20 By 2050, not just the idea of ‘a job for life’, but even the idea of ‘a profession for life’ might seem antediluvian.