Book Read Free

The Economics of Artificial Intelligence

Page 75

by Ajay Agrawal


  user profi le pictures and likenesses without user consent).25

  Similarly, a privacy scare prompted Samsung to change its privacy policy.

  In February 2015, CNN quoted a paragraph of Samsung’s privacy policy,

  22. https:// www .washingtonpost .com/ business/ economy/ 2012/ 06/ 26/ gJQATDUB5V_story

  .html?utm_term=.1ab4fedd7683, accessed October 19, 2017.

  23. Matt McKeon gives a graphical account of how Facebook privacy evolves from 2005 to 2010, at http:// mattmckeon .com/ facebook- privacy/ , accessed on October 24, 2017.

  24. http:// 60secondmarketer .com/ blog/ 2014/ 09/ 21/ facebook- tightens- privacy- controls

  - aff ect- marketing/ , accessed October 24, 3017.

  25. https:// www .wired .com/ 2013/ 08/ judge- approves- 20-million- facebook- sponsored

  - stories- settlement/ , accessed October 24, 2017.

  452 Ginger Zhe Jin

  which stated that words spoken in front of a Samsung Smart TV are cap-

  tured and transmitted to a third party through use of voice recognition.26 In

  response to the intense fear that smart TVs “spy” in a private living room,

  Samsung later changed its privacy policy.27 Samsung also clarifi ed that voice

  recognition can be disabled and it uses industry standard encryption to

  secure the data.

  The privacy competition in the smartphone market is even more inter-

  esting. In 2015, Google launched Android Marshmallow in Android 6.0,28

  which prompts users to grant or deny individual permissions (e.g., access to

  the camera) to a mobile app when it is needed for the fi rst time, rather than

  automatically grant apps all of their specifi ed permissions at installation.

  It also allows users to change the permissions at any time. Similar features

  were made available earlier in Apple iOS 8.29 Apple’s commitment to privacy

  protection was also highlighted when Apple refused to unlock the iPhone

  from one of the shooters in the December 2015 terrorist attack in San Ber-

  nardino, California.

  As a pioneer in biometric authentication, Apple recently announced Face

  ID in its next smartphone launch (iPhone X). Using infrared cameras, Face

  ID uniquely identifi es a user’s face and utilizes that information to unlock

  the smartphone and authorize Apple Pay. Though it is meant to enhance

  convenience and security, Face ID has stirred a number of privacy concerns

  including exposing consumer privacy to Apple employees and allowing the

  police to forcefully unlock a phone using the owner’s face. Whether this AI-

  powered technology will reduce or enhance privacy protection is an open

  question.

  Note that market mechanisms can also work against consumer privacy

  and data security. Dina Florêncio and Cormac Herley (2010) examined the

  password policy of seventy- fi ve websites and found that password strength

  is weaker for some of the largest, most attacked sites that should have greater incentives to protect their valuable database. Compared to security demand,

  it seems that competition is more likely to drive websites to adopt a weaker

  password requirement as they need to compete for users, traffi

  c, and adver-

  tising. The sample size of this study is too small to represent the whole mar-

  ket, but the message is concerning: consumer demand in privacy and data

  26. According to CNN (http:// money.cnn .com/ 2015/ 02/ 09/ technology/ security/ samsung

  - smart- tv- privacy/ index .html), Samsung’s privacy policy said “Please be aware that if your spoken words include personal or other sensitive information, that information will be among the data captured and transmitted to a third party through your use of Voice Recognition.”

  The article further points out that, Samsung SmartTV has a set of pre- programmed commands that it recognizes even if you opt out of voice recognition.

  27. https:// www .cnet .com/ news/ samsung- changes- smarttv- privacy- policy- in-wake- of

  -spying- fears/ , accessed on October 24, 2017.

  28. Android Mashmellow was fi rst released as a beta on May 28, 2015, followed by the offi cial

  release on October 5, 2015. Its new model of app permission was received positively: https:// fpf

  .org/ 2015/ 06/ 23/ android- m- and- privacy- giving- users- control- over- app- permissions/.

  29. https:// fpf .org/ 2014/ 09/ 12/ ios8privacy/ , accessed on October 24, 2017.

  Artifi cial Intelligence and Consumer Privacy 453

  security may compete with the same consumers’ demand for convenience,

  usability, and other attributes (such as lower price). When these demands

  confl ict with each other, fi rms may have a stronger incentive to accommo-

  date the attributes that are more visible and easier to evaluate. Probably the

  same reason explains why only a small fraction of fi rms adopt multifactor

  authentication,30 despite its ability to reduce data risk.

  So far, we have considered AI as an external factor that potentially

  increases the risk of privacy violation and data breach. It is important to

  recognize that AI could also serve as a tool to mitigate the risk. Recently, AI has demonstrated super intelligence in games such as Go, even without the

  help of any human knowledge (Silver et al. 2017). Imagine what data risk

  would look like if the same AI power is used to grant data access to autho-

  rized personnel, to detect data attack when (or even before) it materializes,

  and to precisely predict whether a user- generated posting is authentic or

  fake. In fact, the technology frontier is moving this direction, though its net

  benefi ts remain to be seen.

  Take diff erential privacy as an example. It was invented more than ten

  years ago (Dwork et al. 2006) and claimed by Apple as a key feature to

  protect consumer identity in some of its data collection since 2016. The

  basic logic goes as follows: the data collecting fi rm adds random noise to an

  individual user’s information before uploading it to the cloud. That way, the

  fi rm can still use the collected data for meaningful analysis without knowing

  each user’s secret. The eff ectiveness of this technology depends on how much

  noise to add, a parameter under the control of the data- collecting fi rm.

  To evaluate how Apple implements diff erential privacy in practice, Tang

  et al. (2017) reverse- engineered Apple’s MacOS and iOS operating systems.

  They fi nd that the daily privacy loss permitted by Apple’s diff erential pri-

  vacy algorithm exceeds values acceptable by the theoretical community (Hsu

  et al. 2014), and the overall privacy loss per device may be unbounded. Apple

  disputes the results and argues that its diff erential privacy feature is subject

  to user opt- in. Google is another user of diff erential privacy (in its web

  browser Chrome). The “noise” parameter that Google uses—as estimated

  by Erlingsson, Pihur and Korolova (2014)—seems to be more privacy-

  protective than what is claimed to be used in Apple, but still falls outside

  the most acceptable range.31 These debates cast doubt on the promise of dif-

  ferential privacy, especially on its real use relative to its theoretical potential.

  Another promising technology is blockchain. In plain English, blockchain

  is an ever- growing list of records (blocks) that are linked with timestamp

  and transaction data. Secured by cryptography, blockchain is designed to

  30. Multifactor a
uthentication is a security measure that requires two or more independent credentials to verify the identity of the user. https:// twofactorauth .org/ allows one to search whether a fi rm uses multi- factor authentication in various types of products or services.

  31. https:// www .wired .com/ story/ apple- diff erential- privacy- shortcomings/ , accessed on October 24, 2017.

  454 Ginger Zhe Jin

  be verifi able, permanent, and resistant to data modifi cation. Its successful

  application in Bitcoin suggests that similar technology could trace identi-

  ties in data trade and data use, thus reducing the risk in privacy and data

  security (Catalini and Gans 2017). Ironically, a ransomware attacker in May

  2017 demanded Bitcoin instead of traditional money, probably for a similar

  security reason.

  18.3.3 Policy

  Landscape

  Any market description is incomplete without a summary of the policy

  background. In the United States, there is no overarching legislation on con-

  sumer privacy or data security. So far, the policy landscape is a patchwork

  of federal and local regulations.

  Only a few federal laws are explicit on privacy protection and they all

  tend to be industry specifi c. For example, the Gramm- Leach- Bliley Act

  (GLBA) controls the ways that fi nancial institutions deal with personal data,

  the Health Insurance Portability and Accountability Act of 1996 (HIPAA)

  provides data privacy and security provisions for medical records, and the

  Children’s Online Privacy Protection Act of 1998 (COPPA) disciplines

  online services directed to children under the age of thirteen. In accor-

  dance, privacy is subject to federal regulation by sectors: the Department

  of Health & Human Resources (DHHS) enforces HIPAA in health care, the

  Federal Communication Commission (FCC) regulates telecommunication

  services, the federal reserve systems monitors the fi nancial sector, the Secu-

  rity and Exchange Commission (SEC) focuses on public fi rms and fi nancial

  exchanges, and the Department of Homeland Security (DHS) deals with

  terrorism and cybercrimes related to national security.

  Two exceptions are worth mentioning. First, the Federal Trade Commis-

  sion (FTC) can address privacy violations and inadequate data security as

  deceptive and unfair practice, following the 1914 FTC Act. This enforce-

  ment authority covers almost every industry and overlaps with many sector-

  specifi c regulators.

  More specifi cally, FTC’s privacy enforcement focuses on “notice and

  choice,” which emphasizes how fi rms’ actual data practice deviates from

  the privacy notice they disclose to the public. For industries not subject to

  GLBA, HIPAA, or COPPA, there is no legislation that mandates privacy

  notice, but many fi rms provide it voluntarily and seek consumer consent

  before purchase or consumption. Some industries also adopt self- regulatory

  programs to encourage certain privacy practices.32 This background allows

  the FTC to obtain privacy notice of the targeted fi rm and enforce it under

  the FTC Act.

  32. For example, Digital Advertising Alliance (DAA), a nonprofi t organization led by advertising and marketing trade associations, establishes and enforces privacy practices for digital advertising.

  Artifi cial Intelligence and Consumer Privacy 455

  The FTC has published a number of guidelines on privacy,33 but the best

  way to understand its enforcement is through cases. For example, the FTC

  alleged that Practice Fusion misled consumers by fi rst soliciting reviews

  for their doctors and then publicly posting these reviews on the internet

  without adequate consumer notice. The case eventually settled in June

  2016.34 In another case against Vizio, FTC alleged Vizio captured second-

  by- second information about video displayed on its smart TV, appended

  specifi c demographic information to the viewing data, and sold this infor-

  mation to third parties for targeted ads and other purposes. According to

  the complaint, VIZIO touted its “Smart Interactivity” feature that “enables

  program off ers and suggestions,” but failed to inform consumers that the

  settings also enabled the collection of consumers’ viewing data.35 The case is

  joint with New Jersey Attorney General and settled for $2.2 million in Feb-

  ruary 2017. The third case is against Turn, a digital advertising company

  that tracks consumers in online browser and mobile devices and uses that

  information to target digital advertisements. The FTC alleged that Turn

  used unique identifi ers to track millions of Verizon consumers even after

  they choose to block or delete cookies from websites, which is inconsistent

  with Turn’s privacy policy. Turn settled with FTC in December 2016.36

  While privacy notice is something that consumers can access, read

  (whether they read them is another question) and consent to, most data

  security practices are not visible until someone exposes the data vulnerability

  (via data breach or white- hat discovery). Accordingly, FTC enforcement

  on data security focuses on whether a fi rm has adequate data security, not

  whether the fi rm has provided suffi

  cient information to consumers. Follow-

  ing this logic, the FTC has settled with Ashley Madison, Uber, Wyndham

  Hotel and Resorts, Lenovo, and TaxSlayer, but is engaged in litigation with

  LabMD and D-Link.37

  The second exception relates to government access to personal data.

  Arguably, the US Constitution, in particular the First and Fourth Amend-

  ments, has already covered individual rights in free speech and limited gov-

  ernment ability to access and acquire personal belongings. However, exactly

  how the Constitution applies to electronic data is subject to legal debate

  (Solove 2013).

  33. The most comprehensive FTC guideline is its 2012 privacy report (FTC 2012). A list

  of privacy- related press releases can be found at https:// www .ftc .gov/ news- events/ media

  - resources/ protecting- consumer- privacy/ ftc- privacy- report.

  34. https:// www .ftc .gov/ news- events/ press- releases/ 2016/ 06/ electronic- health- records

  - company- settles- ftc- charges- it- deceived, accessed on October 24, 2017.

  35. https:// www .ftc .gov/ news- events/ press- releases/ 2017/ 02/ vizio- pay- 22-million- ftc- state

  - new- jersey- settle- charges- it, accessed on October 25, 2017.

  36. https:// www .ftc .gov/ news- events/ press- releases/ 2016/ 12/ digital- advertising- company

  - settles- ftc- charges- it- deceptively, accessed on October 25, 2017.

  37. For a list of FTC cases in data security, see https:// www .ftc .gov/ enforcement/ cases

  - proceedings/ terms/ 249.

  456 Ginger Zhe Jin

  Beyond the Constitution, a series of federal laws—the Electronic Com-

  munications Privacy Act of 1986 (ECPA), the Stored Communications Act

  (1986), the Pen Register Act (1986), and the 2001 USA Patriot Act—stipu-

  late when and how the government can collect and process electronic infor-

  mation of individuals. But many of these laws were enacted in the wake of

  the Watergate scandal, long before the use of the internet, email, search

  engines, and social media. It is unclear how they apply to real cases. The

  legal ambiguity is highlighted in t
hree events: fi rst, as exposed by Edward

  Snowden, the NSA has secretly harvested tons of personal information for

  its global surveillance programs. The exposure generates an outcry for pri-

  vacy and a hot debate in the balance between individual privacy and na-

  tional security. Second, the Microsoft email case, regarding whether the US

  government has the right to access emails stored by Microsoft overseas, has

  reached the US Supreme Court. In March 2018, the CLOUD Act clarifi ed

  how US law enforcement orders issued under the Stored Communication

  Act may reach data in other countries and how data hosting companies may

  challenge such law enforcement requests.38 Third, Apple refused to unlock

  the iPhone of one of the shooters in the 2015 San Bernardino terrorist at tack.

  Since the FBI was able to unlock the phone before the court hearing, it

  re mains unknown whether Apple has the legal obligation to help the FBI.39

  At the local level, all fi fty states have enacted data breach notifi cation laws,

  but no federal law has been passed on this topic.40 According to the National

  Conference of State Legislatures, at least seventeen states have also passed

  some law on privacy. For example, the California Consumer Privacy Act

  was enacted in June 2018 and set to be eff ective on January 1, 2020. These

  local laws tend to vary greatly in content, coverage, and remedy.41 From the

  38. http://techcrunch.com/2018/04/17/supreme-court-dismisses-warrant-case-against

  -microsoft-after-cloud-act-renders-it-moot/, accessed January 13, 2019. The CLOUD Act was enacted in March 2018 and stands for the Clarifying Lawful Overseas Use of Data Act.

  39. http:// www .latimes .com/ local/ lanow/ la- me- ln- fbi- drops- fight- to-force- apple- to

  -unlock- san- bernardino- terrorist- iphone- 20160328-story .html, accessed October 25, 2017.

  40. There have been multiple eff orts towards a federal data breach notifi cation law. In 2012, Senator Jay Rockefeller advocated for a cyber security legislation that strengthens the requirement to report cybercrimes. In January 2014, the Senate Commerce, Science, and Transportation Committee (led by Senator Rockefeller) introduced a bill to create a federal requirement for data breach notifi cation (S. 1976 Data Security and Breach Notifi cation Act of 2014). In his 2015 State of the Union Speech, President Obama proposed new legislation to create a national data breach standard with a thirty- day notifi cation requirement for data breach. A related bill was later introduced by the US House of Representatives (H.R. 1770L Data Security and Breach Notifi cation Act of 2015). All of them failed. In the wake of the mega breaches in 2017, Congress has introduced Personal Data Notifi cation and Protection Act of 2017 (H.R.

 

‹ Prev