by Luke Dormehl
While not all errors are as egregious as those denying treatment to cancer patients on the basis of income, a number of coded laws can lack the subtlety of their written counterparts: the result of either laziness or ignorance on the part of programmers. For instance, the United States’ Food Stamp Act limits unemployed childless adults to three months of food stamps. However, it also provides six exceptions to this rule, which cross-reference other exceptions, which then refer to yet more exceptions. Since these exceptions are statistically rare, programmers may be tempted to write code that employs just the simplified three-month version of the rule, leaving out the complicated and potentially confusing exceptions.
While at present this is a problem that relates to the retrofitting of existing laws into code, longer term it presents other possibilities. It might be, for instance, that agencies will be increasingly inclined to adopt policies that favor simple questions and answers over more nuanced approaches—even when this would be the preferable option, since the former is easier to translate into algorithm than the latter. Going forward, this could result in complex laws being unnecessarily simplified so as to allow them to be better automated. Evidence of this is already being seen. In Massachusetts, IT specialists persuaded agency decision-makers to avoid adopting public benefits policies that would prove both challenging and expensive to automate. “Sometimes it feels like the IT department is running the policy,” said one person close to the situation.61
Judge, Jury and Executable Code
Judge Richard Posner is the most cited legal scholar of the 20th century. Years ago, when Posner was a younger, less experienced judge, he presided over a patent case involving an early application of the kind of targeted recommendation technology Amazon would later carry out using algorithms. In the nascent days of satellite television, American television viewers experienced a seismic leap from having access to around 5 or 6 channels to up to 500. For many people, this was the birth of the so-called paradox of choice. With so many options available, how could they possibly be expected to pick the channel they most wanted to watch? One company came up with an answer. Asking for a single channel in each home, they promised to send questionnaires to everyone who received the channel, asking them to list the type of programs they watched most regularly. On the basis of this questionnaire, the company then claimed it would pick the programming they predicted individual customers would most enjoy, thereby giving them their own personalized television channels. “In the morning they could work out that you might like to watch news,” Posner recalls. “In the afternoon they might see that you like to watch soap operas, and in the evening that you enjoy watching horror movies.” Why worry about the other 499 channels when there was one that would have everything you wanted on it?
Posner thought the concept was “ingenious.” The idea stuck with him, and over the years he considered how it could affect his own profession. “You could imagine a similar thing being done with judges,” he reflects. “You take their opinions, their public statements, and whatever other information you have about them and use it to create a profile that could function as a predictive device.” Posner elaborated on the notion in a recent paper entitled “The Role of the Judge in the Twenty-First Century,” in which he postulates on the ways in which judicial practices will likely be altered by the arrival of The Formula. “We are all familiar,” he writes, “with how Amazon.com creates and modifies reader profiles, and some of us are familiar with data-mining, which is the same procedure writ large—the computer identifies patterns and updates them as new data are received.”
I look forward to a time when computers will create profiles of judges’ philosophies from their opinions and their public statements, and will update these profiles continuously as the judges issue additional opinions. [These] profiles will enable lawyers and judges to predict judicial behavior more accurately, and will assist judges in maintaining consistency with their previous decisions—when they want to.62
Judges can be prickly when it comes to the complexity of their decision-making process. “Our business is prophecy, and if prophecy were certain, there would not be much credit in prophesying,” wrote Judge Max Radin in a 1925 essay entitled “The Theory of Judicial Decision: Or How Judges Think.”63 Since he was writing close to a century ago, Radin cannot entirely be blamed for underestimating the power of computation—and yet it is possible to take him to task for his judicial arrogance. After all, if a judge is objective in his thought processes and therefore not subject to hunches, biases or any other “processes . . . alien to good judges”64 (as federal judge Joseph C. Hutcheson Jr. would note four years after Radin’s essay was published), it stands to reason that their decision-making process should be apparent to anyone with the proper legal grounding.
As it turns out, even sufficient legal training may not be necessary, as was suggested in a 2004 study in which an algorithm competed with a team of legal experts to see which could predict the greater number of Supreme Court verdicts. The algorithm surprised the study’s authors by correctly predicting 75 percent of the verdicts (based on only a handful of different metrics), as compared to the team of legal scholars, who guessed just 59 percent, despite having access to far more specialized information.65 In its own way, the “Supreme Court Forecasting Project” was the legal profession’s equivalent of IBM’s Watson supercomputer winning $1 million on Jeopardy! in 2011—marking, as it did, the culmination of a long-held techno dream first proposed by the jurimetrics movement. In 1897, Oliver Wendell Holmes Jr. wrote enthusiastically of his belief that the legal system, as with science’s natural laws, should be quantifiably predictable. “The object of our study . . . is prediction,” he observed, “the prediction of the incidence of the public force through the instrumentality of the courts.”66
But if anything, the Supreme Court Forecasting Project was a bastardization of the jurimetrics’ utopian vision. Rather than demonstrating that the legal system’s innate objectivity made it predictable, the 2004 study showed that the ability to correctly predict judges’ verdicts resided in the fact that they were not objective, but rather filtered in each case through a combination of ideological preferences. For instance, one of the metrics the algorithm used to predict verdicts came down to whether judges voted Democrat or Republican, a factor that, objectively speaking, should no more influence whether a judge finds a defendant guilty or innocent than the length of time since they last ate.
It is these “attitudinal” biases that data-mining algorithms could be used to discover. “This could be a useful tool for judges, to help them be more self-aware when it comes to bias,” Judge Posner says. A judge might, for instance, be soft on criminals, but tough on business fraud. “When they receive their profile they may become aware that they have certain unconscious biases that push them in certain directions,” Posner continues.
Harry Surden, associate professor at the University of Colorado Law School and formerly a software engineer for Cisco Systems, agrees with Posner’s proposal, but says that the revelations about the biases that affect judicial decision-making could shock the general public. “As a democratic society, we might not like what we see,” he says.
One area where such algorithms would likely be embraced would be in the tool kit of lawyers, who could use them to accurately model a judge’s behavior to shape witnesses and arguments for maximum impact in a courtroom. “This is probably going on today in some of the more sophisticated legal entities,” Surden speculates. He points to the likes of hedge funds that make large bets whose outcomes are predicated on legal results. Using sophisticated legal modeling algorithms to predict verdicts in such a scenario could directly result in vast quantities of money being made.
Other legal scholars have also suggested that data mining could be used to reveal a granularity of wrongfulness in criminal trials, using machine-readable criteria to calculate the exact extent of a person’s guilt. One imagines that it will not be too long before neurosc
ience (currently in the throes of an algorithmic turn) seeks to establish the exact degree to which a person is free to comply with the law, arguing over questions like determinism versus voluntarism. Data-mining tools may additionally have an application when it comes to meting out punishments: perhaps using algorithms to compare datasets of verdicts with datasets showing the postconviction behaviors of offenders. Could it be, for example, that there are precise lengths of ideal sentence to avoid wrongdoers relapsing into crime?
Pushing the idea yet further, what about the concept of automated judges? If we could construct a recommender system able to predict with 99 percent accuracy how a certain judge might rule on a particular case (perhaps even with greater levels of consistency than the human judge it is based on), would this help to make trials fairer? “In principle, it would be possible, although it’s still a way away,” Judge Posner says. “The main thing that would be left out would be how the judge’s views changed with new information. Any change that may affect the way judges think would somehow have to be entered into the computer program, and weighed in order to decide how he would decide a case.”
At least at present it takes the creativity of a (human) judge to resolve multiple parties’ grievances, while also reconciling differing interpretations of the law. In this way, the judicial process is less about a mechanical objectivity than it is about a high level of intersubjective agreement: a seemingly minor, but crucial difference. Algorithms may have plenty of applications in the courtroom—and could even be used effectively to make the existing system fairer—but they’re unlikely to start handing down sentences.
For now, at least.
CHAPTER 4
The Machine That Made Art
Quite possibly the most famous statement ever made about Hollywood is the one that screenwriter William Goldman laid out in his 1983 memoir, Adventures in the Screen Trade. Combing through his decades in the film industry for something approaching a profundity, all Goldman was able to muster was the idea that, when it comes to the tricky business of moviemaking, “Nobody knows anything. Did you know,” he asks,
that Raiders of the Lost Ark was offered to every single studio in town and they all turned it down? All except Paramount. Why did Paramount say yes? Because nobody knows anything. And why did all the other studios say no? Because nobody knows anything. And why did Universal, the mightiest studio of them all, pass on Star Wars . . . ? Because nobody . . . knows the least goddam [sic] thing about what is or isn’t going to work at the box office.1
Goldman is hardly alone in his protestations. In the autobiography of studio executive Mike Medavoy, the legendary film executive who had a hand in Apocalypse Now, One Flew Over the Cuckoo’s Nest and The Silence of the Lambs, opines, “The movie business is probably the most irrational business in the world . . . [It] is governed by a set of rules that are absolutely irrational.”2 Hollywood lore is littered with stories of surefire winners that become flops, and surefire flops that become winners. There are niche films that appeal to everyone, and mainstream films that appeal to no one. Practically no Hollywood decision-maker has an unblemished record and, when one considers the facts, it’s difficult to entirely blame them.
• • •
Take the ballad of the two A-list directors, for example. In the mid-2000s, James Cameron announced that he was busy working on a mysterious new script, called “Project 880.” Cameron had previously directed a string of hit movies, including The Terminator, Terminator 2, Aliens, The Abyss, True Lies and Titanic, the latter of which won 11 Academy Awards and became the first film in history to earn more than a billion dollars. For the past ten years, however, he had been sidelined making documentaries. Moreover, his proposed new film didn’t have any major stars, he was planning to shoot it in the experimental 3-D format, and he was asking for $237 million to do so. Perhaps against more rational judgment, the film was nonetheless made, and when it arrived in cinemas it was instantly labeled in the words of one reviewer, “The Most Expensive American Film Ever . . . And Possibly the Most Anti-American One Too.”3
Would you have given Cameron the money to make it? The correct answer, as many cinema-goers will be aware, is yes. Retitled Avatar, Project 880 proceeded to smash the record set by Titanic, becoming the first film in history to earn more than $2 billion at the box office.
At around the same time that Avatar was gaining momentum, a second project was doing the rounds in Hollywood. This was another science-fiction film, also in 3-D, based upon a classic children’s story, with a script cowritten by a Pulitzer Prize–winning author, and was to be made by Andrew Stanton, a director with an unimpeachable record who had previously helped create the highly successful Pixar films WALL-E and Finding Nemo, along with every entry in the acclaimed Toy Story series. Stanton’s film (let’s call it “Project X”) came with a proposed $250 million asking price to bring to the screen—a shade more than Project 880. Project X received the go-ahead too, only this didn’t turn out to be the next Avatar, but rather the first John Carter, a film that lost almost $200 million for its studio and resulted in the resignation of the head of Walt Disney Studios (the company who bankrolled it), despite the fact that he had only taken the job after the project was already in development. As per Goldman’s Law, nobody knows anything.
Patterns Everywhere
This is, of course, exactly the type of conclusion that challenge-seeking technologists love to hear about. The idea that there should be something (the entertainment industry at that) that is entirely unpredictable is catnip to the formulaic mind. As it happens, blockbuster movies and high-tech start-ups do have a fair amount in common. Aside from the fact that most are flops, and investors are therefore reliant on the winners being sufficiently big that they more than offset the losers, there are few other industries where the power of the “elevator pitch” holds more sway. The elevator pitch is the idea that any popular concept should be sufficiently simple that it can be explained over the course of a single elevator ride. Perhaps not coincidentally, this is roughly the time frame of a 30-second commercial: the tool most commonly used for selling movies to a wide audience.
Like almost every popular idea, the elevator pitch (also known in Hollywood as the “high concept” phenomenon) has been attributed to a number of industry players, although the closest thing to an agreed-upon dictionary definition still belongs to Steven Spielberg. “If a person can tell me the idea in twenty-five words or less, it’s going to make a pretty good movie,” the director of Jurassic Park and E.T. the Extra-Terrestrial has said. “I like ideas . . . that you can hold in your hand.”4 Would you be in the least surprised to hear that Spielberg’s father, Arnold Spielberg, was a pioneering computer scientist who designed and patented the first electronic library system that allowed the searching of what was then considered to be vast amounts of data? Steven Spielberg might have been a flop as a science student, but The Formula was in his blood.5
The Formula also runs through the veins of another Hollywood figure, whom Time magazine once praised for his “machine-like drive,” and who originally planned to study engineering at MIT, before instead taking the acting route and going on to star in an almost unblemished run of box office smashes. Early in Will Smith’s career, when he was little more than a fad pop star appearing in his first television show, The Fresh Prince of Bel-Air, the aspiring thespian sat down with his manager and attempted to work out a formula that would transform him from a nobody into “the biggest movie star in the world.” Smith describes himself as a “student of the patterns of the universe.” At a time when he was struggling to be seen by a single casting director, Smith spent his days scrutinizing industry trade papers for trends that appeared in what global audiences wanted to see. “We looked at them and said, ‘Okay, what are the patterns?’” he later recalled in an interview with Time magazine.6 “We realized that ten out of ten had special effects. Nine out of ten had special effects with creatures. Eight out of ten had special eff
ects with creatures and a love story . . .” Two decades later, and with his films having grossed in excess of $6.36 billion worldwide, Smith’s methodology hasn’t changed a great deal. “Every Monday morning, we sit down [and say], ‘Okay, what happened this weekend, and what are the things that resemble things that have happened the last ten, twenty, thirty weekends?’” he noted.
Of course, as big a box-office attraction as Smith undoubtedly is, when it comes to universal pattern spotting, he is still strictly small-fry.
The Future of Movies
In the United Kingdom, there is a company called Epagogix that takes the Will Smith approach to movie prediction, only several orders of magnitude greater and without the movie star good looks. Operating out of a cramped office in Kennington, where a handful of data analysts sit hunched over their computers and the walls are covered with old film posters, Epagogix is the unlikely secret weapon employed by some of the biggest studios in Hollywood. Named after the Greek word for the path that leads from experience to knowledge, Epagogix carries a bold claim for itself: it can—so CEO and cofounder Nick Meaney claims—accurately forecast how much money a particular film is going to earn at the box office, before the film in question is even made.
