Through the technology embedded in almost every major tech platform and every web-enabled device, algorithms and the artificial intelligence that underlies them make a staggering number of everyday decisions for us, from what products we buy, to where we decide to eat, to how we consume our news, to whom we date, and how we find a job. We've even delegated life-and-death decisions to algorithmsdecisions once made by doctors, pilots, and judges. In his new book, Kartik Hosanagar surveys the brave new world of algorithmic decision-making and reveals the potentially dangerous biases they can give rise to as they increasingly run our lives. He makes the compelling case that we need to arm ourselves with a better, deeper, more nuanced understanding of the phenomenon of algorithmic thinking. And he gives us a route in, pointing out that algorithms often think a lot like their creatorsthat is, like you and me.
Hosanagar draws on his experiences designing algorithms professionallyas well as on history, computer science, and psychologyto explore how algorithms work and why they occasionally go rogue, what drives our trust in them, and the many ramifications of algorithmic decision-making. He examines episodes like Microsoft's chatbot Tay, which was designed to converse on social media like a teenage girl, but instead turned sexist and racist; the fatal accidents of self-driving cars; and even our own common, and often frustrating, experiences on services like Netflix and Amazon. A Human's Guide to Machine Intelligence is an entertaining and provocative look at one of the most important developments of our time and a practical user's guide to this first wave of practical artificial intelligence.
Related collections and offers
|Publisher:||Penguin Publishing Group|
|Product dimensions:||5.40(w) x 8.30(h) x 0.80(d)|
About the Author
Read an Excerpt
Free Will in an Algorithmic World
If you did consider your choices, you'd be confronted with a truth you cannot comprehend: that no choice you ever made has been your own. You've always been a prisoner. What if I told you I'm here to set you free?
"The Man in Black", Westworld, season 1, episode 4
Tai, a senior at the University of Pennsylvania, wakes up at the perfect time every morning-well rested, but not late for classes or appointments. Today that meant rising at 7:18 a.m. He did not set his alarm for that time. Rather, it was chosen for him. His phone's sleep-tracker app had been following his sleep patterns over the past few months, monitoring his REM cycles and periods of lighter rest. Using this information, it set a smart alarm that wakes him during a light stage of sleep, while also trying to maintain some level of consistency over time. The theory is that this schedule will prime Tai for greater energy and concentration throughout the day.
Tai needs to be sharp. He's at a turning point in his life, about to step away from the relatively safe world of college-of information-gathering, homework, and exams-into the "real" world of practical problem solving: finding a job, choosing a place to live, and negotiating the wonderful but complicated details of a romantic relationship that's getting more serious by the day.
Although Tai is open to advice from his professors, his friends, and his family, he also wants to go his own way. He considers himself to be an independent thinker, capable of weighing lots of different options and then choosing the right one himself. He needs a good mind for that-and a good night's sleep.
Tai rolls over in bed and with one eye open grabs his phone and checks his notifications: fourteen likes on his latest Insta, seven Facebook notifications, and three comments on his new Facebook profile picture. Not bad for a Monday night. He scrolls down his Facebook feed. An article shared by his friend Harry grabs his attention with its headline, "The Wealth of New Choices With Robot Vacuum Cleaners." He clicks and, liking what he reads about the Eufy RoboVac cleaner, forwards the article to his girlfriend, Kate.
There's an email from his mom, too, with a link to a New York Times article, "What I Wish I'd Known Before Moving in Together." Tai groans. Mention even a possibility to his mother, and she sets it in stone. The picture accompanying the article shows an attractive couple in their thirties sitting on an unblemished white staircase, smiling into each other's eyes. He types, "Haha thanks. that middle-aged couple looks happy, see. How did you find this?" Calling them middle-aged will definitely get on his mom's nerves. But there's no time for more needling: it's already 7:28 a.m.
Tai rolls out of bed and, walking across his dusty carpet, opens his dresser, pulling out a pair of stretch washed chinos from Bonobos (he follows the online clothing retailer on Instagram), blue-and-gray argyle socks (top rated on Amazon), and a dress shirt and tie. He has a job interview today.
As he sits down for breakfast, Tai thinks of the fortuitous circumstances that led to the interview. He had found the job posting through his friend Samantha, whom LinkedIn's algorithms had reminded him to congratulate on her six-month work anniversary. Their conversation had been a little awkward, as Tai and Samantha had matched on Tinder a few years earlier. She was an artsy girl with a bubbling self-confidence; lots to like about her, but neither of them felt any sparks. And although they became friends, it had been hard for Tai to keep up with her since she graduated, especially since Kate wasn't Samantha's biggest fan.
Tai's friendship with Samantha is hardly the only thing that's been getting on Kate's nerves lately. Their discussion about possibly moving in together seems to be stressing her out. Over the weekend, Tai had sent Kate a Huffington Post recommended article: "15 Things Couples Should Do Before Moving in Together," which she read with great interest-especially point number 15, "Have an exit strategy." Tai had suggested that if they did split up, it would make sense for her to be the one to move out-after all, he had found the new apartment for the two of them. But it was only a contingency plan. Her angry texts on the subject were still awaiting his reply.
After dressing, Tai checks his phone again to see if there are more texts. Nothing new from Kate, but there is a reply from his mom about the Times article: "Oh, I was looking for housewarming gifts for you and Kate and it popped up on Google. Why don't you send it to her, sweetie? And good luck on your interview this morning!"
Tai can hear Chance the Rapper, chosen for him by Spotify Discover, rapping on the other side of his bedroom wall, which is now glowing with the light of the rising sun from the east window. It's time to head out for the interview. He looks for an Uber to take him to campus. The price is $11.23, which feels a bit steep; yesterday it had been $9.34 for the same route. He closes the app and relaunches it. The price is now $10.82. It's not clear to Tai why it changed, but he confirms the booking this time and waits at his door for the Toyota Corolla to pull up.
As he exchanges pleasantries with the driver, Tai opens a notebook to work on his case interviews, the part of business school job applications where students are asked to think through a challenging business scenario and present a solution. The case prep document shared by another student includes the question: What is Root Cause Analysis? Tai jots down some notes, applies that technique to analyze his day today, and produces a diagram:
It all seems kind of random at one level. But he can't help but wonder about the degree to which the algorithms employed by Facebook, Google, Tinder, and Amazon have a role to play in his present circumstances. Will he have some cooked-up equation from a programmer to thank for his next job? And is this job really the best next step for his life and career, or just the accidental result of inconsequential past decisions-clicks of a mouse and swipes on a screen? Tai likes to think of himself as being in the driver's seat. But this Uber ride suggests he's not-both figuratively and literally.
Or maybe he's just overthinking things-the aftereffect of an in-class discussion I led on personalization algorithms just a few days earlier. He sends me an email: "Have something interesting to show you. Do you have ten minutes after class?"
Tai sighs and shuts his notebook. Maybe all he and Kate need is to get away for a bit to reconsider this moving-in idea. He pulls out his phone and opens Expedia's app. It might have some good hotel recommendations.
Since 2004 I've been teaching a class at Wharton called "Enabling Technologies." In hindsight I should have named it "What's Going On in Tech," because that's a more accurate and descriptive name. The course examines the technologies that are shaping entire industries as well as the daily lives of countless individuals-including students like Tai. In 2004, we covered broadband technologies, online shopping, and Voice over IP (the ability to make voice calls over the internet using services such as Skype). Today, those topics seem almost mundane, as we now discuss the Internet of Things, virtual reality, and space tech. One topic that has remained a constant in the course through the years is algorithmic decision making. Although we once discussed Amazon's product recommendations, we now consider algorithms in such applications as driverless cars and robo-advisers. An additional and subtler change is that the sort of question that Tai asked-to what extent are we in control of our own actions?-is coming up in the class more and more often.
All of us realize how much of our lives are shaped by the decisions we make online, whether through searches on Google, connecting with friends on Facebook, or shopping on Amazon. Many of us are aware that the companies running these sites are guiding our choices, often by customizing our experience on their websites and apps. Personalization algorithms help us choose the optimal products to buy on Amazon, the best movies to watch on Netflix, the ideal person to date through Tinder and Match.com, the most useful contacts on LinkedIn, and the most compelling posts and articles to read on Facebook. But in our imagining, we generally nod politely at these recommendations and make our own choices. After all, we are in charge here.
And yet consider these facts: 80 percent of viewing hours streamed on Netflix originate from automated recommendations. By some estimates nearly 35 percent of sales at Amazon originate from automated recommendations. And the vast majority of matches on dating apps such as Tinder and OkCupid are initiated by algorithms. Given these numbers, many of us clearly do not have quite the freedom of choice that we believe we do.
One reason is that products are often designed in ways that make us act impulsively and against our better judgment. For example, suppose you have a big meeting at work tomorrow. Ideally, you want to spend some time preparing for it in the evening and then get a good night's rest. But before you can do either, a notification pops up on your phone indicating that a friend tagged you on Facebook. This will take a minute, you tell yourself as you click on it. But after logging in you discover a long feed of posts by friends. A few clicks later you find yourself watching a YouTube video that one of them shared. As soon as the video ends, YouTube suggests other related and interesting videos. Before you know it, it's 1:00 a.m., and it's clear that you will need an all-nighter to get ready for the following morning's meeting. This has happened to most of us. The reason this behavior is so common, as some product designers have noted, is that popular design approaches-such as the use of notifications and gamification to increase user engagement-exploit and amplify human vulnerabilities, such as our need for social approval, or our inability to resist immediate gratification even when we recognize that it comes with long-term costs. While we might feel as if we are making our own choices, we're often nudged or even tricked into making them.
Another reason that we aren't truly in control of our choices is that when we search for a hotel on Expedia, browse online dating profiles, or shop for a book, we're seeing only a small fraction of all the potentially relevant information available. Although we experience a clear sense of free will by making the final decision regarding what we see, read, or buy, the fact is that 99 percent of all possible alternatives were excluded.
You probably don't mind saving all the time you might have wasted in sifting through inferior options to arrive at a final choice. But algorithms do not simply help us find products or information more quickly that we might have found eventually without their assistance. In truth, they exert a significant influence on precisely what and how much we consume.
Consider the role of search algorithms. Given millions of possible search results, thousands of which are likely to be highly relevant to a particular query, Google's algorithms determine which ones are featured at the very top of the results page. This ranking exerts a powerful influence on our responses. About 33 percent of clicks go to the number-one result in Google searches; fewer than 10 percent go to links outside of the top ten results.
Automated recommendations are also a major driver of choice online. More than any individual or organization -including Oprah, the National Book Awards, or The New York Times- Amazon's recommendation algorithms have the biggest influence on which books people are reading. Automated recommendations drive purchase decisions across a wide variety of product categories, from kitchenware and perfumes to electronics and artwork. Beyond retailers such as Amazon and Walmart.com, online media companies such as Netflix, Spotify, Apple's iTunes, and Google's YouTube all employ algorithmic recommendations to gently nudge us in specific directions.
The impact of algorithms are also experienced on social media websites, where we are likely to believe that our friends are the chief drivers of the content we see. In reality, algorithms play an equally important role. Every time a user opens Facebook's app or website, there are on average about 1,500 potential stories or posts that Facebook can show. Its algorithms determine which ones you should read first, which can be read later, and even which you don't need to read at all. The algorithms consider a number of factors to determine the posts they show us-how often we interact with the friend who posted; the number of likes, comments, and shares that the post received in aggregate and from our friends; how recently the post appeared; whether any users tried to hide it; and so on. Instagram and Twitter have also recently adopted algorithmic feeds. Given how social networks have become the gateway for online news and media discovery for so many of us, news-feed algorithms are crucial to determining the reportage we read and the opinions we form about the world around us.
Perhaps inevitably, algorithms are not just selecting the media we see on social networks but are also silently determining the network itself-that is, whom we allow into our personal and professional lives. LinkedIn's algorithms will magically remind you of the people you met last week or emailed yesterday so they can be added to your professional network. Interested in reconnecting with childhood friends? Facebook's algorithms will recommend whom to add as friends on Facebook. And why stop at friends? Algorithms built by companies such as Tinder, Match.com, and eHarmony will even determine whom you date or marry. Algorithms are the primary drivers of matches on online dating apps and websites, which, by some estimates, are used by as much as 40 percent of the singles population in the United States.
Opinions are one force in determining the actions we take as individuals; our feelings are another. Consider the case of Match.com, launched in 1995 and now the country's most popular dating website.
Gary Kremen, the man who conceived of Match.com, was inspired in part by newspaper classified ads. If you're old enough, you might remember that placing a personal ad usually involved giving a few details about yourself and a few about the person you were looking to meet: "Single male, 35, avid reader, seeks single female, 20-30, fit and fun." The job of finding a match was left to the reader of the ad. Match.com's earliest algorithms sought to replicate this model, but also to step in as the actual matchmaker, noticing the single male whom the busy single woman might have missed and putting the two in touch.
Table of Contents
Part 1 The Rogue Code
1 Free Will in an Algorithmic World 21
2 The Law of Unanticipated Consequences 39
Part 2 Algorithmic Thinking
3 Omelet Recipes for Computers: How Algorithms Are Programmed 59
4 Algorithms Become Intelligent: A Brief History of Al 83
5 Machine Learning and the Predictability-Resilience Paradox 101
6 The Psychology of Algorithms 125
Part 3 Taming the Code
7 In Algorithms We Trust 145
8 Which Is to Be Master-Algorithm or User? 165
9 Inside the Black Box 181
10 An Algorithmic Bill of Rights 205
Conclusion: The Games Algorithms Play 225