Three years previously, on 6 May 2010, the New York stock exchange underwent an even sharper shock. Within five minutes – from 14:42 to 14:47 – the Dow Jones dropped by 1,000 points, wiping out $1 trillion. It then bounced back, returning to its pre-crash level in a little over three minutes. That’s what happens when super-fast computer programs are in charge of our money. Experts have been trying ever since to understand what happened in this so-called ‘Flash Crash’. We know algorithms were to blame, but we are still not sure exactly what went wrong. Some traders in the USA have already filed lawsuits against algorithmic trading, arguing that it unfairly discriminates against human beings, who simply cannot react fast enough to compete. Quibbling whether this really constitutes a violation of rights might provide lots of work and lots of fees for lawyers.5
And these lawyers won’t necessarily be human. Movies and TV series give the impression that lawyers spend their days in court shouting ‘Objection!’ and making impassioned speeches. Yet most run-of-the-mill lawyers spend their time going over endless files, looking for precedents, loopholes and tiny pieces of potentially relevant evidence. Some are busy trying to figure out what happened on the night John Doe got killed, or formulating a gargantuan business contract that will protect their client against every conceivable eventuality. What will be the fate of all these lawyers once sophisticated search algorithms can locate more precedents in a day than a human can in a lifetime, and once brain scans can reveal lies and deceptions at the press of a button? Even highly experienced lawyers and detectives cannot easily spot deceptions merely by observing people’s facial expressions and tone of voice. However, lying involves different brain areas to those used when we tell the truth. We’re not there yet, but it is conceivable that in the not too distant future fMRI scanners could function as almost infallible truth machines. Where will that leave millions of lawyers, judges, cops and detectives? They might need to go back to school and learn a new profession.6
When they get in the classroom, however, they may well discover that the algorithms have got there first. Companies such as Mindojo are developing interactive algorithms that not only teach me maths, physics and history, but also simultaneously study me and get to know exactly who I am. Digital teachers will closely monitor every answer I give, and how long it took me to give it. Over time, they will discern my unique weaknesses as well as my strengths. They will identify what gets me excited, and what makes my eyelids droop. They could teach me thermodynamics or geometry in a way that suits my personality type, even if that particular way doesn’t suit 99 per cent of the other pupils. And these digital teachers will never lose their patience, never shout at me, and never go on strike. It is unclear, however, why on earth I would need to know thermodynamics or geometry in a world containing such intelligent computer programs.7
Even doctors are fair game for the algorithms. The first and foremost task of most doctors is to diagnose diseases correctly, and then suggest the best available treatment. If I arrive at the clinic complaining about fever and diarrhoea, I might be suffering from food poisoning. Then again, the same symptoms might result from a stomach virus, cholera, dysentery, malaria, cancer or some unknown new disease. My doctor has only five minutes to make a correct diagnosis, because this is what my health insurance pays for. This allows for no more than a few questions and perhaps a quick medical examination. The doctor then cross-references this meagre information with my medical history, and with the vast world of human maladies. Alas, not even the most diligent doctor can remember all my previous ailments and check-ups. Similarly, no doctor can be familiar with every illness and drug, or read every new article published in every medical journal. To top it all, the doctor is sometimes tired or hungry or perhaps even sick, which affects her judgement. No wonder that doctors often err in their diagnoses, or recommend a less-than-optimal treatment.
Now consider IBM’s famous Watson – an artificial intelligence system that won the Jeopardy! television game show in 2011, beating human former champions. Watson is currently groomed to do more serious work, particularly in diagnosing diseases. An AI such as Watson has enormous potential advantages over human doctors. Firstly, an AI can hold in its databanks information about every known illness and medicine in history. It can then update these databanks every day, not only with the findings of new researches, but also with medical statistics gathered from every clinic and hospital in the world.
Credit 1.44
44. IBM’s Watson defeating its two humans opponents in Jeopardy! in 2011.
Secondly, Watson can be intimately familiar not only with my entire genome and my day-to-day medical history, but also with the genomes and medical histories of my parents, siblings, cousins, neighbours and friends. Watson will know instantly whether I visited a tropical country recently, whether I have recurring stomach infections, whether there have been cases of intestinal cancer in my family or whether people all over town are complaining this morning about diarrhoea.
Thirdly, Watson will never be tired, hungry or sick, and will have all the time in the world for me. I could sit comfortably on my sofa at home and answer hundreds of questions, telling Watson exactly how I feel. This is good news for most patients (except perhaps hypochondriacs). But if you enter medical school today in the expectation of still being a family doctor in twenty years, maybe you should think again. With such a Watson around, there is not much need for Sherlocks.
This threat hovers over the heads not only of general practitioners, but also of experts. Indeed, it might prove easier to replace doctors specialising in a relatively narrow field such as cancer diagnosis. For example, in a recent experiment a computer algorithm diagnosed correctly 90 per cent of lung cancer cases presented to it, while human doctors had a success rate of only 50 per cent.8 In fact, the future is already here. CT scans and mammography tests are routinely checked by specialised algorithms, which provide doctors with a second opinion, and sometimes detect tumours that the doctors missed.9
A host of tough technical problems still prevent Watson and its ilk from replacing most doctors tomorrow morning. Yet these technical problems – however difficult – need only be solved once. The training of a human doctor is a complicated and expensive process that lasts years. When the process is complete, after ten years of studies and internships, all you get is one doctor. If you want two doctors, you have to repeat the entire process from scratch. In contrast, if and when you solve the technical problems hampering Watson, you will get not one, but an infinite number of doctors, available 24/7 in every corner of the world. So even if it costs $100 billion to make it work, in the long run it would be much cheaper than training human doctors.
And what’s true of doctors is doubly true of pharmacists. In 2011 a pharmacy opened in San Francisco manned by a single robot. When a human comes to the pharmacy, within seconds the robot receives all of the customer’s prescriptions, as well as detailed information about other medicines taken by them, and their suspected allergies. The robot makes sure the new prescriptions don’t combine adversely with any other medicine or allergy, and then provides the customer with the required drug. In its first year of operation the robotic pharmacist provided 2 million prescriptions, without making a single mistake. On average, flesh-and-blood pharmacists get wrong 1.7 per cent of prescriptions. In the United States alone this amounts to more than 50 million prescription errors every year!10
Some people argue that even if an algorithm could outperform doctors and pharmacists in the technical aspects of their professions, it could never replace their human touch. If your CT indicates you have cancer, would you like to receive the news from a caring and empathetic human doctor, or from a machine? Well, how about receiving the news from a caring and empathetic machine that tailors its words to your personality type? Remember that organisms are algorithms, and Watson could detect your emotional state with the same accuracy that it detects your tumours.
This idea has already been implemented by some customer-services departments,
such as those pioneered by the Chicago-based Mattersight Corporation. Mattersight publishes its wares with the following advert: ‘Have you ever spoken with someone and felt as though you just clicked? The magical feeling you get is the result of a personality connection. Mattersight creates that feeling every day, in call centers around the world.’11 When you call customer services with a request or complaint, it usually takes a few seconds to route your call to a representative. In Mattersight systems, your call is routed by a clever algorithm. You first state the reason for your call. The algorithm listens to your request, analyses the words you have chosen and your tone of voice, and deduces not only your present emotional state but also your personality type – whether you are introverted, extroverted, rebellious or dependent. Based on this information, the algorithm links you to the representative that best matches your mood and personality. The algorithm knows whether you need an empathetic person to patiently listen to your complaints, or you prefer a no-nonsense rational type who will give you the quickest technical solution. A good match means both happier customers and less time and money wasted by the customer-services department.12
—
The most important question in twenty-first-century economics may well be what to do with all the superfluous people. What will conscious humans do, once we have highly intelligent non-conscious algorithms that can do almost everything better?
Throughout history the job market was divided into three main sectors: agriculture, industry and services. Until about 1800, the vast majority of people worked in agriculture, and only a small minority worked in industry and services. During the Industrial Revolution people in developed countries left the fields and herds. Most began working in industry, but growing numbers also took up jobs in the services sector. In recent decades developed countries underwent another revolution, as industrial jobs vanished, whereas the services sector expanded. In 2010 only 2 per cent of Americans worked in agriculture, 20 per cent worked in industry, 78 per cent worked as teachers, doctors, webpage designers and so forth. When mindless algorithms are able to teach, diagnose and design better than humans, what will we do?
This is not an entirely new question. Ever since the Industrial Revolution erupted, people feared that mechanisation might cause mass unemployment. This never happened, because as old professions became obsolete, new professions evolved, and there was always something humans could do better than machines. Yet this is not a law of nature, and nothing guarantees it will continue to be like that in the future. Humans have two basic types of abilities: physical abilities and cognitive abilities. As long as machines competed with us merely in physical abilities, you could always find cognitive tasks that humans do better. So machines took over purely manual jobs, while humans focused on jobs requiring at least some cognitive skills. Yet what will happen once algorithms outperform us in remembering, analysing and recognising patterns?
The idea that humans will always have a unique ability beyond the reach of non-conscious algorithms is just wishful thinking. The current scientific answer to this pipe dream can be summarised in three simple principles:
1. Organisms are algorithms. Every animal – including Homo sapiens – is an assemblage of organic algorithms shaped by natural selection over millions of years of evolution.
2. Algorithmic calculations are not affected by the materials from which you build the calculator. Whether you build an abacus from wood, iron or plastic, two beads plus two beads equals four beads.
3. Hence there is no reason to think that organic algorithms can do things that non-organic algorithms will never be able to replicate or surpass. As long as the calculations remain valid, what does it matter whether the algorithms are manifested in carbon or silicon?
True, at present there are numerous things that organic algorithms do better than non-organic ones, and experts have repeatedly declared that something will ‘for ever’ remain beyond the reach of non-organic algorithms. But it turns out that ‘for ever’ often means no more than a decade or two. Until a short time ago, facial recognition was a favourite example of something which even babies accomplish easily but which escaped even the most powerful computers on earth. Today facial-recognition programs are able to recognise people far more efficiently and quickly than humans can. Police forces and intelligence services now use such programs to scan countless hours of video footage from surveillance cameras, tracking down suspects and criminals.
In the 1980s when people discussed the unique nature of humanity, they habitually used chess as primary proof of human superiority. They believed that computers would never beat humans at chess. On 10 February 1996, IBM’s Deep Blue defeated world chess champion Garry Kasparov, laying to rest that particular claim for human pre-eminence.
Deep Blue was given a head start by its creators, who preprogrammed it not only with the basic rules of chess, but also with detailed instructions regarding chess strategies. A new generation of AI uses machine learning to do even more remarkable and elegant things. In February 2015 a program developed by Google DeepMind learned by itself how to play forty-nine classic Atari games. One of the developers, Dr Demis Hassabis, explained that ‘the only information we gave the system was the raw pixels on the screen and the idea that it had to get a high score. And everything else it had to figure out by itself.’ The program managed to learn the rules of all the games it was presented with, from Pac-Man and Space Invaders to car racing and tennis games. It then played most of them as well as or better than humans, sometimes coming up with strategies that never occur to human players.13
Credit 1.45
45. Deep Blue defeating Garry Kasparov.
Computer algorithms have recently proven their worth in ball games, too. For many decades, baseball teams used the wisdom, experience and gut instincts of professional scouts and managers to pick players. The best players fetched millions of dollars, and naturally enough the rich teams got the cream of the market, whereas poorer teams had to settle for the scraps. In 2002 Billy Beane, the manager of the low-budget Oakland Athletics, decided to beat the system. He relied on an arcane computer algorithm developed by economists and computer geeks to create a winning team from players that human scouts overlooked or undervalued. The old-timers were incensed by Beane’s algorithm transgressing into the hallowed halls of baseball. They said that picking baseball players is an art, and that only humans with an intimate and long-standing experience of the game can master it. A computer program could never do it, because it could never decipher the secrets and the spirit of baseball.
They soon had to eat their baseball caps. Beane’s shoestring-budget algorithmic team ($44 million) not only held its own against baseball giants such as the New York Yankees ($125 million), but became the first team ever in American League baseball to win twenty consecutive games. Not that Beane and Oakland could enjoy their success for long. Soon enough, many other baseball teams adopted the same algorithmic approach, and since the Yankees and Red Sox could pay far more for both baseball players and computer software, low-budget teams such as the Oakland Athletics now had an even smaller chance of beating the system than before.14
In 2004 Professor Frank Levy from MIT and Professor Richard Murnane from Harvard published a thorough research of the job market, listing those professions most likely to undergo automation. Truck drivers were given as an example of a job that could not possibly be automated in the foreseeable future. It is hard to imagine, they wrote, that algorithms could safely drive trucks on a busy road. A mere ten years later, Google and Tesla not only imagine this, but are actually making it happen.15
In fact, as time goes by, it becomes easier and easier to replace humans with computer algorithms, not merely because the algorithms are getting smarter, but also because humans are professionalising. Ancient hunter-gatherers mastered a very wide variety of skills in order to survive, which is why it would be immensely difficult to design a robotic hunter-gatherer. Such a robot would have to know how to prepare spear points from flint stones, how to
find edible mushrooms in a forest, how to use medicinal herbs to bandage a wound, how to track down a mammoth and how to coordinate a charge with a dozen other hunters. However, over the last few thousand years we humans have been specialising. A taxi driver or a cardiologist specialises in a much narrower niche than a hunter-gatherer, which makes it easier to replace them with AI.
Even the managers in charge of all these activities can be replaced. Thanks to its powerful algorithms, Uber can manage millions of taxi drivers with only a handful of humans. Most of the commands are given by the algorithms without any need of human supervision.16 In May 2014 Deep Knowledge Ventures – a Hong Kong venture-capital firm specialising in regenerative medicine – broke new ground by appointing an algorithm called VITAL to its board. VITAL makes investment recommendations by analysing huge amounts of data on the financial situation, clinical trials and intellectual property of prospective companies. Like the other five board members, the algorithm gets to vote on whether the firm makes an investment in a specific company or not.
Examining VITAL’s record so far, it seems that it has already picked up one managerial vice: nepotism. It has recommended investing in companies that grant algorithms more authority. With VITAL’s blessing, Deep Knowledge Ventures has recently invested in Silico Medicine, which develops computer-assisted methods for drug research, and in Pathway Pharmaceuticals, which employs a platform called OncoFinder to select and rate personalised cancer therapies.17
As algorithms push humans out of the job market, wealth might become concentrated in the hands of the tiny elite that owns the all-powerful algorithms, creating unprecedented social inequality. Alternatively, the algorithms might not only manage businesses, but actually come to own them. At present, human law already recognises intersubjective entities like corporations and nations as ‘legal persons’. Though Toyota or Argentina has neither a body nor a mind, they are subject to international laws, they can own land and money, and they can sue and be sued in court. We might soon grant similar status to algorithms. An algorithm could then own a venture-capital fund without having to obey the wishes of any human master.