There was sleight of hand in this self-portrayal. The algorithm may be the essence of computer science – but it’s not precisely a scientific concept. An algorithm is a system, like plumbing or a military chain of command. It takes knowhow, calculation and creativity to make a system work properly. But some systems, like some armies, are much more reliable than others. A system is a human artefact, not a mathematical truism. The origins of the algorithm are unmistakably human, but human fallibility isn’t a quality that we associate with it. When algorithms reject a loan application or set the price for an airline flight, they seem impersonal and unbending. The algorithm is supposed to be devoid of bias, intuition, emotion or forgiveness.
Silicon Valley’s algorithmic enthusiasts were immodest about describing the revolutionary potential of their objects of affection. Algorithms were always interesting and valuable, but advances in computing made them infinitely more powerful. The big change was the cost of computing: it collapsed, just as the machines themselves sped up and were tied into a global network. Computers could stockpile massive piles of unsorted data – and algorithms could attack this data to find patterns and connections that would escape human analysts. In the hands of Google and Facebook, these algorithms grew ever more powerful. As they went about their searches, they accumulated more and more data. Their machines assimilated all the lessons of past searches, using these learnings to more precisely deliver the desired results.
For the entirety of human existence, the creation of knowledge was a slog of trial and error. Humans would dream up theories of how the world worked, then would examine the evidence to see whether their hypotheses survived or crashed upon their exposure to reality. Algorithms upend the scientific method – the patterns emerge from the data, from correlations, unguided by hypotheses. They remove humans from the whole process of inquiry. Writing in Wired, Chris Anderson, then editor-in-chief, argued: “We can stop looking for models. We can analyse the data without hypotheses about what it might show. We can throw the numbers into the biggest computing clusters the world has ever seen and let statistical algorithms find patterns where science cannot.”
On one level, this is undeniable. Algorithms can translate languages without understanding words, simply by uncovering the patterns that undergird the construction of sentences. They can find coincidences that humans might never even think to seek. Walmart’s algorithms found that people desperately buy strawberry Pop-Tarts as they prepare for massive storms.
Still, even as an algorithm mindlessly implements its procedures – and even as it learns to see new patterns in the data – it reflects the minds of its creators, the motives of its trainers. Amazon and Netflix use algorithms to make recommendations about books and films. (One-third of purchases on Amazon come from these recommendations.) These algorithms seek to understand our tastes, and the tastes of like-minded consumers of culture. Yet the algorithms make fundamentally different recommendations. Amazon steers you to the sorts of books that you’ve seen before. Netflix directs users to the unfamiliar. There’s a business reason for this difference. Blockbuster movies cost Netflix more to stream. Greater profit arrives when you decide to watch more obscure fare. Computer scientists have an aphorism that describes how algorithms relentlessly hunt for patterns: they talk about torturing the data until it confesses. Yet this metaphor contains unexamined implications. Data, like victims of torture, tells its interrogator what it wants to hear.
Like economics, computer science has its preferred models and implicit assumptions about the world. When programmers are taught algorithmic thinking, they are told to venerate efficiency as a paramount consideration. This is perfectly understandable. An algorithm with an ungainly number of steps will gum up the machinery, and a molasses-like server is a useless one. But efficiency is also a value. When we speed things up, we’re necessarily cutting corners; we’re generalising.
Algorithms can be gorgeous expressions of logical thinking, not to mention a source of ease and wonder. They can track down copies of obscure 19th-century tomes in a few milliseconds; they put us in touch with long-lost elementary school friends; they enable retailers to deliver packages to our doors in a flash. Very soon, they will guide self-driving cars and pinpoint cancers growing in our innards. But to do all these things, algorithms are constantly taking our measure. They make decisions about us and on our behalf. The problem is that when we outsource thinking to machines, we are really outsourcing thinking to the organisations that run the machines.
Mark Zuckerberg disingenuously poses as a friendly critic of algorithms. That’s how he implicitly contrasts Facebook with his rivals across the way at Google. Over in Larry Page’s shop, the algorithm is king – a cold, pulseless ruler. There’s not a trace of life force in its recommendations, and very little apparent understanding of the person keying a query into its engine. Facebook, in his flattering self-portrait, is a respite from this increasingly automated, atomistic world. “Every product you use is better off with your friends,” he says.
What he is referring to is Facebook’s news feed. Here’s a brief explanation for the sliver of humanity who have apparently resisted Facebook: the news feed provides a reverse chronological index of all the status updates, articles and photos that your friends have posted to Facebook. The news feed is meant to be fun, but also geared to solve one of the essential problems of modernity – our inability to sift through the ever-growing, always-looming mounds of information. Who better, the theory goes, to recommend what we should read and watch than our friends? Zuckerberg has boasted that the News Feed turned Facebook into a “personalised newspaper”.
Unfortunately, our friends can do only so much to winnow things for us. Turns out, they like to share a lot. If we just read their musings and followed links to articles, we might be only a little less overwhelmed than before, or perhaps even deeper underwater. So Facebook makes its own choices about what should be read. The company’s algorithms sort the thousands of things a Facebook user could possibly see down to a smaller batch of choice items. And then within those few dozen items, it decides what we might like to read first.
Algorithms are, by definition, invisibilia. But we can usually sense their presence – that somewhere in the distance, we’re interacting with a machine. That’s what makes Facebook’s algorithm so powerful. Many users – 60%, according to the best research – are completely unaware of its existence. But even if they know of its influence, it wouldn’t really matter. Facebook’s algorithm couldn’t be more opaque. It has grown into an almost unknowable tangle of sprawl. The algorithm interprets more than 100,000 “signals” to make its decisions about what users see. Some of these signals apply to all Facebook users; some reflect users’ particular habits and the habits of their friends. Perhaps Facebook no longer fully understands its own tangle of algorithms – the code, all 60m lines of it, is a palimpsest, where engineers add layer upon layer of new commands.
Pondering the abstraction of this algorithm, imagine one of those earliest computers with its nervously blinking lights and long rows of dials. To tweak the algorithm, the engineers turn the knob a click or two. The engineers are constantly making small adjustments here and there, so that the machine performs to their satisfaction. With even the gentlest caress of the metaphorical dial, Facebook changes what its users see and read. It can make our friends’ photos more or less ubiquitous; it can punish posts filled with self-congratulatory musings and banish what it deems to be hoaxes; it can promote video rather than text; it can favour articles from the likes of the New York Times or BuzzFeed, if it so desires. Or if we want to be melodramatic about it, we could say Facebook is constantly tinkering with how its users view the world – always tinkering with the quality of news and opinion that it allows to break through the din, adjusting the quality of political and cultural discourse in order to hold the attention of users for a few more beats.
But how do the engineers know which dial to twist and how hard? There’s a whole discipline, data science, to guide the writing and revision of algorithms. Facebook has a team, poached from academia, to conduct experiments on users. It’s a statistician’s sexiest dream – some of the largest data sets in human history, the ability to run trials on mathematically meaningful cohorts. When Cameron Marlow, the former head of Facebook’s data science team, described the opportunity, he began twitching with ecstatic joy. “For the first time,” Marlow said, “we have a microscope that not only lets us examine social behaviour at a very fine level that we’ve never been able to see before, but allows us to run experiments that millions of users are exposed to.”
Facebook likes to boast about the fact of its experimentation more than the details of the actual experiments themselves. But there are examples that have escaped the confines of its laboratories. We know, for example, that Facebook sought to discover whether emotions are contagious. To conduct this trial, Facebook attempted to manipulate the mental state of its users. For one group, Facebook excised the positive words from the posts in the news feed; for another group, it removed the negative words. Each group, it concluded, wrote posts that echoed the mood of the posts it had reworded. This study was roundly condemned as invasive, but it is not so unusual. As one member of Facebook’s data science team confessed: “Anyone on that team could run a test. They’re always trying to alter people’s behaviour.”
There’s no doubting the emotional and psychological power possessed by Facebook – or, at least, Facebook doesn’t doubt it. It has bragged about how it increased voter turnout (and organ donation) by subtly amping up the social pressures that compel virtuous behaviour. Facebook has even touted the results from these experiments in peer-reviewed journals: “It is possible that more of the 0.60% growth in turnout between 2006 and 2010 might have been caused by a single message on Facebook,” said one study published in Nature in 2012. No other company has made such claims about its ability to shape democracy like this – and for good reason. It’s too much power to entrust to a corporation.
The many Facebook experiments add up. The company believes that it has unlocked social psychology and acquired a deeper understanding of its users than they possess of themselves. Facebook can predict users’ race, sexual orientation, relationship status and drug use on the basis of their “likes” alone. It’s Zuckerberg’s fantasy that this data might be analysed to uncover the mother of all revelations, “a fundamental mathematical law underlying human social relationships that governs the balance of who and what we all care about”. That is, of course, a goal in the distance. In the meantime, Facebook will keep probing – constantly testing to see what we crave and what we ignore, a never-ending campaign to improve Facebook’s capacity to give us the things that we want and things we don’t even know we want. Whether the information is true or concocted, authoritative reporting or conspiratorial opinion, doesn’t really seem to matter much to Facebook. The crowd gets what it wants and deserves.
The automation of thinking: we’re in the earliest days of this revolution, of course. But we can see where it’s heading. Algorithms have retired many of the bureaucratic, clerical duties once performed by humans – and they will soon begin to replace more creative tasks. At Netflix, algorithms suggest the genres of movies to commission. Some news wires use algorithms to write stories about crime, baseball games and earthquakes – the most rote journalistic tasks. Algorithms have produced fine art and composed symphonic music, or at least approximations of them.
It’s a terrifying trajectory, especially for those of us in these lines of work. If algorithms can replicate the process of creativity, then there’s little reason to nurture human creativity. Why bother with the tortuous, inefficient process of writing or painting if a computer can produce something seemingly as good and in a painless flash? Why nurture the overinflated market for high culture when it could be so abundant and cheap? No human endeavour has resisted automation, so why should creative endeavours be any different?
How algorithms rule our working lives
The engineering mindset has little patience for the fetishisation of words and images, for the mystique of art, for moral complexity or emotional expression. It views humans as data, components of systems, abstractions. That’s why Facebook has so few qualms about performing rampant experiments on its users. The whole effort is to make human beings predictable – to anticipate their behaviour, which makes them easier to manipulate. With this sort of cold-blooded thinking, so divorced from the contingency and mystery of human life, it’s easy to see how long-standing values begin to seem like an annoyance – why a concept such as privacy would carry so little weight in the engineer’s calculus, why the inefficiencies of publishing and journalism seem so imminently disruptable.
Facebook would never put it this way, but algorithms are meant to erode free will, to relieve humans of the burden of choosing, to nudge them in the right direction. Algorithms fuel a sense of omnipotence, the condescending belief that our behaviour can be altered, without our even being aware of the hand guiding us, in a superior direction. That’s always been a danger of the engineering mindset, as it moves beyond its roots in building inanimate stuff and begins to design a more perfect social world. We are the screws and rivets in the grand design.