by ME Stucke · 2017 · Cited by 10 — algorithmic collusion, behavioural discrimination and abuses by dominant super- We show how network effects, big data and big analytics will likely.

62 KB – 34 Pages

PAGE – 1 ============
Research Paper # 324 July 2017 How Your Digital Helper May Undermine Your Welfare, and Our Democracy Maurice E. Stucke & Ariel Ezrachi Berkeley Technology Law Journal (Forthcoming) This paper may be downloaded without charge from the Social Science Resear ch Network Electronic library at : Learn more about the University of Tennessee College of Law:

PAGE – 2 ============
1 How Your Digital Helper May Undermine Your Welfare, and Our Democracy Maurice E. Stucke * & Ariel Ezrachi ** ÒAll you need to do is say,Ó a recent article proclaimed, ÒÕI want a beerÕ and Alexa will oblige. The future is now.Ó 1 Advances in technology have seemingly increased our choices and opened markets to competition. As we migrate from brick -and-mortar shops to online commerce, we seemingly are get ting more of what we desire at better prices and quality. And yet, behind the competitive fa“ade, a more complex reality exists. We explore in VIRTUAL COMPETITION several emerging threats, namely algorithm ic collusion, behavioural discrimination and abuses by dominant super -platforms. 2 But the harm is not just economic. The potential anticompet itive consequences go beyond our pocketbooks. The toll will likely be on our privacy, well -being and democracy. To see why, this Essay examines the emerging frontier of digital personal assistants. These helpers are being developed by the leading online platforms: Google Assistant, AppleÕs Siri, FacebookÕs M, and AmazonÕs Alexa -powered Echo. 3 These super -platforms are heavily investing to improve their offerings. For those of us who grew up watching The Jetsons , the prospect of our own personal helper m ight seem marvelous. With its increas ing sophistication, a digital butler promise s to transform the way we access information , communicate, shop, are entertained, control our smart household appliances, and raise our children. As the artificial intelligenc e and communication interface advance, digital assistants will offer an unparalleled personalized experience. These digital assistants Ð or Òdigital butlers Ó Ð can provide us not just with information and services , but can anticipate our needs and requests. They can do so, based on our connections, data profile, behavior, and so forth. The digital butler , with our trust and consent, will likely * Professor, University of Tennessee College of Law; Co -founder, The Konkurrenz Group. ** Slaughter and May Professor of Competition Law, The University of Oxford. Director, Oxford University Centre for Competition Law and Policy. 1 -can -now -tell -amazons -alexa -to-bring -you -a-beer -amazon -echo 2 For further discussion, see: Ariel Ezrachi and Maurice E Stucke, Virtual Competition: The Promise and Perils of the Algorithm Driven Economy (Harvard 2016). 3 10-K, Alphabet Inc., CIK 0001652044, 001 -37580, February 03, 2017, 2017 WL 00448915 (identifying as competitors, digital assistant providers, Òsuch as Apple, Amazon, Facebook, and MicrosoftÓ).

PAGE – 3 ============
2 become our key gateway to the World Wide Web. With this unique position of power, it will act as a gate-keeper in a multi-sided market. Our digital butler will also undertake mundane tasks and free our time. As AlphabetÕs CEO noted, ÒYour phone should proactively bring up the right documents, schedule and map your meetings, let people know if you are late, suggest responses to messages, handle your payments and expenses, etc.Ó 4 As the digital butler, powered by sophisticated algorithms, learns more about us, our routine, wants and communications, it can excel in its role. In a human -like manner, it can be funny Ñ at just the appropriate level Ñ and trustworthy. After all, being p rivy to so many of our activities, it will become our digital shadow. This is unsurprising. Many of us already rely on GoogleÕs search engine to find relevant results, Facebook to identify relevant news stories, Amazon for book recommendations, and Siri to place phone calls, send text messages, and find a good Chinese restaurant nearby. So, we will happily relinquish other less personal and useful interfaces, and increasingly rely on our butler to anticipate and fulfill our needs, and alert us. And yet, de spite their promise, can personalized digital assistants actually reduce our welfare? Might their rise reduce the number of gateways to the digital world, increase the market power of a few firms, and limit competition? And if so, what are the potential so cial, political, and economic concerns? Our Essay seeks to address these questions. We show how network effects, big data and big analytics will likely undermine attempts to curtail a digital assistantÕs power, and will likely allow it to operate below the regulatory and antitrust radar screens. As a result, rather than advance our overall welfare, these digital assistants – if left to their own devices – can undermine our welfare . I.!The Rise of Digital Personal Assistants Many of us already benefit from basic digital assistants. Apple iPhones users may have Siri call their mo ther on speaker. Siri can ÒpredictÓ what app they might want to use, which music they would like to listen to. Navigation apps can anticipate where we are heading throughout the day and provide traffic updates and time estimates. Other application s encourage use by ranking us in comparison to others and updating us during the day. Even your favourite coffee outlet may send you a notification and prepare your loyalty card on your device whenever youÕre near an outlet. 4 nternet/google -ceo -pichai -sees -the -end -of-computers -as-physical -devices/articleshow/52040890.cms

PAGE – 4 ============
3 Now digital personal assistants are seeking to interact with us in a human -like way. AmazonÕs voice recognition personal assistant, Alexa , for example, can shop for you (knowing everything you previously bought through Amazon); plan your mornings, including upcoming meetings, traffic, weather, etc.; entertain you with music; suggest movies, shows, or au diobooks; and control your houseÕs smart appliances. 5 In 2016, Google showed a video of a suburban family undergoing its morning wakeup routine: ÒThe dad made French press coffee while telling Google to turn on the lights and start playing music in his kid sÕ rooms. The mom asked if Ômy packageÕ had shipped. It did, Google said. The daughter asked for help with her Spanish homework.Ó 6 A positive feedback loop can be observed . As more people use a particular digital platform , the greater the demand for products and services that can connect to the digital platform , the m ore likely other manufacturers and developers will develop applications for that platform, the more appealing the platform becomes to both consumers, manufacturers and software d evelopers . We will consider the competitive implications of these network effects later. For now, our focus is how these network effects can increase demand for the product. We are starting to see thi s network effect for digital personal assi stants. Sales for these digital assistants are accelerating. Alexa -enabled devices, Amazon reported in 2017, Òwere the top -selling products across all categories on this holiday season. Customers purchased and gifted a record -setting number of devices from the Amazon Echo family with sales up over 9x compared to last holiday season.Ó 7 To increase sales of Alexa, Amazon in 2015 opened its Alexa Voice Service to third -party hardware makers, Ògiving them the tools to inte grate Alexa into internet -connected devices .Ó8 The aim is to get more ÒsmartÓ appliances, like lights, fans, switches, thermostats, garage doors, sprinklers, locks, and more , connected to Alexa. Amazons announced in early 2017 that Ò[t]ens of thousands of developers Ó were using the Alexa Voice Service to integrate Alexa into their products, including ÒDish DVRs, Ford and Volkswagen vehicles, GE C Lamp, Huawei Mate 9, LG Smart Instaview 5 -echo -assistant/ . 6 Danny Y adron, ÔGoogle Assistant takes on Amazon and Apple to be the ultimate digital butlerÕ; -home -assistant -amazon -echo -apple -siri 7 8-K | EX -99.1, Amazon Com Inc., CIK 0001018724, 000 -22513, February 02, 2017. 8 8-K | EX -99.1, AMAZON COM INC, CIK 0001018724, 000 -22513, July 23, 2015.

PAGE – 5 ============
4 fridge, and Whirlpool appliances.Ó 9 Thus, as more people use Alexa, more manufacturers will make smart -products which Alexa can control, and the more appealing Alexa becomes to prospective purchasers and manufacturers . A second feedback loop is to teach the digital assistant new skills. Amazon offers a free Alexa Skills Kit, which Òmakes it fast and easy for developers to create new voice -driven capabilities for Alexa. Ó10 As more people purchas e Alexa, more companies will develop new skills for Alexa . In early 2016, for example, Alexa could directly order a pizza from DominoÕs or a car from Uber, check your credit balance with Capital One, get fitness information from Fitbit, alert you with election updates from NBC News , play Jeopardy!, get stock quotes wit h Fidelity, hear headlines from The Huffington Post, provide you a seven -minute workout, and test your Star Wars knowledge with a trivia quiz from Disney. 11 Indeed, AlexaÕs skills selection tripled in three months in 2016 alone , with over Ò3,000 skills available, including Food Network, GE A ppliances, Yahoo Sports Fantasy Football, and more.Ó 12 By mid -2016, Amazon had Ò tens of thousands of developers building new skills for Alexa.Ó 13 Also in 2016 Amazon announced Òthe Alexa Prize, an annual university competition with $2.5 million dedicated to accelerating the field of conversational artificial intelligence. Ó14 The competition Õs aim is Òto build a ÔsocialbotÕ on Alexa that will converse with people about popular topics and news events.Ó 15 Thus, as more people use a particular digital butler, more companies will develop new skills for that digital butler (like ordering beer and pizza) , which makes the butler more appealing to prospective purchasers and developers . As the digital butler seamlessly converses with us, provid ing more of what interests us and less of what doesnÕt, we will grow to like and trust it. ÒAlexa may be AmazonÕs most loved invention yet Ñ literally Ñ with over 250,000 marriage proposals from customers and counting,Ó said Jeff Bezos, AmazonÕs founder and CE O. ÒAnd sheÕs just getting better. Because AlexaÕs brain is in the cloud, we can easily and continuously add to her capabilities and make her more useful Ñ wait until you see some of the surprises the team is working on now.Ó 16 9 8-K | EX -99.1, AMAZON COM INC, CIK 0001018724, 000 -22513, February 02, 2017. 10 8-K | EX -99.1, AMAZON COM INC, CIK 0001018724, 000 -22513, July 23, 2015. 11 8-K | EX -99.1, AMA ZON COM INC, CIK 0001018724, 000 -22513, April 28, 2016; 8 -K | EX -99.1, AMAZON COM INC, CIK 0001018724, 000 -22513, January 28, 2016. 12 8-K | EX -99.1, AMAZON COM INC, CIK 0001018724, 000 -22513, October 27, 2016. 13 Id. 14 Id. 15 Id. 16 Id.

PAGE – 6 ============
5 As one reviewer in early 201 6 noted, ÒWith a rapidly growing slate of features and integrations, it’s easy to get excited about the Echo’s potential. . . . More than a year after its debut, the Echo is smarter than ever, and one of the best connected home products money can currently buy.Ó17 Over the n ext few year s, as more skills are developed, more features are added, and more trial -and-error learning (as we discuss below), our digital assistants will be even smarter and in many more homes. II.!The Race to Become Our Head Butler Currently Microsoft and the four online super -platforms ÑGoogle, Apple, Facebook, and Amazon Ñare jockeying for their digital assistant to become our head butler. Each is jockeying as to Òwho gets to control the primary interface of mobile devices.Ó 18 The stakes are great and go beyond the mere use of the digital assistant. In this competitive race, each super -platform wants its personal assistant to become our key gateway. Let us see why. As we shift from a mobile -dominated world to an AI -dominated pl atform, we will converse primarily with our head butler, who will increasingly predict and fulfil our needs. We will less frequently search the web, look at price -comparison websites, or download apps. As Alphabet CEO Sundar Pichai wrote shareholders in an April 2016 letter, ÒThe next big step will be for the very concept of the ÔdeviceÕ to fade away. Over time, the computer itself Ñwhatever its form factor Ñwill be an intelligent assistant helping you through your day.Ó 19 Google Õs search engine used Òto show just ten blue links in [its] results, which you had to click through to find your answers. Ó20 Now Google is Ò increasingly able to provide direct answers — even if you’re speaking your question using Voice Search — which makes it quicker, easier an d more natural to find what you’re looking for. Ó21 Rather than searching online for information, you can now talk with Google Assistant Òin a natural conversational way to help you get things done. Ó22 Thus, Google Assistant forms part of the companyÕs Òeffor t to further 17 -echo -review/ 18 Mims, ÒAsk M for Help.Ó 19 Microsoft, Other Tech Giants Race to Develop Machine Intelligence, -giants -race -to-develop -machine -intelligence -1465941959 20 10-K, Alphabet Inc., CIK 0001652044, 001 -37580, February 03, 2017, 2017 WL 00448915 21 Id. 22 Id.

PAGE – 8 ============
7 Firms compete to dominate markets characterized by ne twork effects. As one product or standard increases in popularity, it trends toward dominance . The big get bigger, until they dominate the industry. As one U.S. court observed, Òonce dominance is achieved, threats come largely from outside the dominated market, because the degree of dominance of such a mark et tends to become so extreme.Ó 28 The stakes are huge with digital personal assistant s because several network effects are at play. One network effect is the positive feedback loop in attracting manufac turers and developers. 29 It is inefficient for them to develop apps, hardware and software for every digital butler. Instead they will likely focus on the top -selling digital butlers. So, if more people primarily use AmazonÕs Alexa, its operating platformÕs applications and functions will likely attract more developers and smart-appliance manufacturers . Alexa will learn more skills compared to rivals , making it more attractive than rival digital butlers. This type of network effect helped Microsoft maintain its dominance in personal computer operating systems for decades. In Microsoft , the government plaintiffs argued that network effects acted as structural barriers for those seeking to ent er the market for Intel -compatible personal computer operati ng systems. 30 The Court of Appeals agreed . An Òapplications barrier to entryÓ prot ected MicrosoftÕs dominance. Thi s is because Ò(1) most consumers prefer operating systems for which a large number of applications have already been written; and (2) most developers prefer to write for operating systems that already have a substantial consumer base.Ó This Òchicken -and-eggÓ situation Òensures that applications will continue to be written for the already dominant Windows, which in turn ensures that consumers will continue to prefer it over other operating systems.Ó The court also noted that this applications barrier to entry led consumers to prefer the dominant operating system, even if they did not need all the available applications : The consumer wants an o perating system that runs not only types of applications that he knows he will want to use, but also those types in which 28 Novell, Inc. v. Microsoft Corp., 505 F.3d 302, 308 (4th Cir. 2007). 29 Indirect network effects arise when people increasingly use a product or technology (for example, software platforms). T he more people that use the platform, Òthe more there will be invested in developing products compatible with that platform, which, in turn reinforces the popularity of that platform with users.Ó Case T -201/04, Microsoft Corp. v. CommÕn, 2007 E.C.R. II-3601 (Ct. First Instance), para 1061. 30 United States v. Microsoft Corp., 253 F.3d 34, 49 Ð50, 55 (D.C. Cir. 2001).

PAGE – 9 ============
8 he might develop an interest later. Also, the consumer knows that if he chooses an operating system with enough demand to support mult iple applications in each product category, he will be less likely to find himself straitened later by having to use an application whose features disappoint him. Finally, the average user knows that, generally speaking, applications improve through succes sive versions. He thus wants an operating system for which successive generations of his favorite applications will be released Ñpromptly at that. The fact that a vastly larger number of applications are written for Windows than for other PC operating syste ms attracts consumers to Windows, because it reassures them that their interests will be met as long as they use Microsoft’s product. 31 This network effect also helped solidify GoogleÕs and AppleÕs dominance over the mobile phone operating system. Besides this traditional network effect , an additional data -driven network effect involves learning -by-doing. We see this data -driven network effect with search engines. 32 Each personÕs utility from using the search engine increases as others use it as well. As more people use the search engine, the more trial -and-error experiments, the more likely the search engine can learn consumersÕ preferences, the more relevant the sea rch results will likely be, which in turn will likely attract others to use the search engine, and the positive feedback continues. Interestingly, with this network effect, as more people use the service or product, its quality improves. The underlying c ode and algorithms of FacebookÕs M, for example, are largely open source. The key assets are not the algorithms . (Otherwise why share them?) Key are the scale and scope of data and the algorithmÕs ability to learn by trial -by-error. As the Wall Street Jour nal reported, ÒFacebook Messenger already has more than 700 million users,Ó which yields it the following advantage: Òwith access to so many users, Facebook has a plausible way to get the gigantic quantity of conversational data required to make a chat -based assistant sufficiently automated.Ó 33 With more users making more requests, M can quickly process more tasks easily. 31 United States v. Microsoft Corp., 253 F.3d 34, 55 (D.C. Cir. 2001). 32 Maurice E. Stucke and Allen P. Grunes, Big Data and Competition Policy ( OUP 2016). 33 Christopher Mims, ÒAsk M for Help: Facebook Tests New Digital Assistant: Single Interface Could Replace Web Searches and Apps on Mobile Devices,Ó Wall Street Journal, November 9, 2015, -m -for -help -facebook -tests -new -digital -assistant -1447045202.

PAGE – 10 ============
9 By learning through servicing us, digital butlers can take a pro -active role Ñ anticipating our needs and wants Ñ rather than following instructions. This requires the platform to have enough users, data and opportunities to experiment to train the algorithms A third data -driven network effect involves the scope of personal data collected on us and predicting our unique needs . The super -platforms already expend a lot of effort to better track us, collect ou r personal data, and profile us. So t he feedback loop adds a dimension: it is no longer the trial -and-error, learning -by-doing from earlier queries by millions of other users , but an additional layer of trial-and-error in predicting individual tastes and preferences from the variety of personal data the super -platform collects about you . The more time you converse with your digital butler, and the mo re data it collects about you, the more opportunities the butler can anticipate your particular needs. Ultimately these data-driven network effects will further weed out the five large platforms. We do not want five butlers, each asking us about movies tonight or food to order. So e ach super -platform wants its digital butler to primarily undertake our tasks and make decisions for us. In discussing its digital personal assistant, GoogleÕs CEO said, ÒWe want users to have an ongoing two -way dialogue with Google.Ó 34 The more we converse with, and delegate to, the head butler, the better it can predict our tastes, and the more likely we are to rely on it for our daily activities. As our butler accumulates information over time, the switching costs (and quali ty gap) between butlers will become higher. We could therefore be willingly locked into our comfort zone. III.!Several Privacy Concerns So what are the implications of this winner -take-all contest to be our head butler? First is the toll these digital butlers can have on our privacy and peace of mind. A recent criminal case gives us a glimpse. The Bentonville Police Department in Arkansas was investigating a death at the defendantÕs residence. The defendant was charged with first -degree murder. While searching the defendantÕs residence, the police seized an Echo device. The police next served Amazon a warrant seeking any audio recordings and transcripts that were 34 Jack Nicas, ÒGoogle Touts New AI -Powered Tools,Ó Wall Street Journal, May 19, 2016, B1, B4.

PAGE – 11 ============
10 created as a result of interactions with defendantÕs Amazon Echo. Citing Òimportant First Amendment and privacy implications at stake ,Ó Amazon sought to quash the search warrant Òunless the Court finds that the State has met its heightened burden for compelled production of such materials.Ó 35 As Amazon told the court, the pri vacy concerns were significant. Its digital butler Òcan be commanded to, among other things, play music, stream podcasts, play audio books, request information about various subjects, or request Òreal -time information,Ó including news, weather, and traffic conditions related to the userÕs or any other location.Ó 36 As one example, Òusers may ask for information about a sensitive health condition or a controversial political figure.Ó 37 Users can also use their digital assistant to order products from Amazon, in cluding books and other expressive materials. Thus, the digital personal assistant sweeps in a lot of data that can Òreveal much more in combination than any isola ted record. Ó 38 Those with access to the data can reconstruct Ò[t]he sum of an individualÕs pr ivate life.Ó 39 Amazon was concerned with governmental invasions of its usersÕ privacy and First Amendment interest s. As Amazon cautioned, Òthe knowledge that government agents are seeking records concerning customer purchases of expressive material from Amazon Ôwould frost keyboards across America.ÕÓ 40 Indeed, Ò Õrumors of an Orwellian federal criminal investigation into the reading habits of AmazonÕs customers could frighten countless potential customers Õ into cancelling their on line purchases through Amazon, Ô now and perhaps forever, Õ resulting in a chilling effect on the publicÕs willingness to purchase expressive materials. Ó41 Eventually, after the defendant consented, Amazon disclosed the information to the State. 42 But, as Amazon identified, government surveillance remains a concern. It is questionable whether the accused can challenge under the Fourth Amendment any 35 Memorandum of Law in Support of AmazonÕs Motion to Quash Search Warrant, in State v. James A. Bates, Case No. Cr -2016-370-2 (Circuit Court of Benton County Arkansas filed Feb. 17, 2017), at 1. Amazon argued that the State must demonstrate: (1) a compellin g need for the information sought, including that it is not available from other sources; and (2) a sufficient nexus between the information and the subject of the criminal investigation. Id. at 2. 36 Id. at 5. 37 Id. 38 Id. at 9 (quoting Riley v. California , 134 S. Ct. 2473, 2489 (2014)). 39 Id. (quoting Riley , 134 S. Ct. at 2489). 40 Id. at 14 (quoting In re Grand Jury Subpoena to dated August 7, 2006, 246 F.R.D. at 573) . 41 Id. (quoting Grand Jury Subpoena , 246 F.R.D. at 573) . 42 Amazon gives up Ale xa data sought in murder probe, 2017 WLNR 7410625.

62 KB – 34 Pages