Book Read Free

The Economics of Artificial Intelligence

Page 69

by Ajay Agrawal


  cating their workforce about security practices, and so on, they have a safe

  harbor with respect to liability for costs associated with security incidents.

  But where does the due care standard come from? One possibility is from

  the government, particularly from military or law enforcement practices.

  The Orange Book and its successor, the Common Criteria standard, are

  good examples. Another possibility is that insurance agencies off er insur-

  ance to parties that implement good practices security. Just as an insurer

  may require a sprinkler system to off er fi re insurance, cyber insurance may

  only be off ered to those companies that engage in best practices (see Varian

  2000 for more discussion).

  This model is an appealing approach to the problem. However, we know

  that there are many issues involving insurance such as adverse selection and

  moral hazard that need to be addressed. See the archives of the Workshop

  on the Economics of Information Security for more work in this area, and

  Anderson (2017) for an overview.

  16.5.2 Privacy

  Privacy policy is a large and sprawling area. Acquisti, Taylor, and Wag-

  man (2016) provide a comprehensive review of the economic literature.

  There are several policy questions that arise in the machine- learning area.

  For example, do fi rms have appropriate incentives to provide appropriate

  levels of privacy? What is the trade- off between privacy and economic per-

  formance? It is widely recognized that privacy regulations may limit ability

  of ML vendors to combine data from multiple sources and there may be

  Artifi cial Intelligence, Economics, and Industrial Organization 417

  limits on transfer of data across corporate boundaries and/or sale of data.

  There is a tendency to promulgate regulation in this area that leads to unin-

  tended consequences. An example is the Health Insurance Portability and

  Accountability Act of 1996, commonly known as HIPAA. The original

  intent of the legislation was to stimulate competition among insurers by

  establishing standards for medical record keeping. However, many research-

  ers argue that it has had a signifi cant negative impact on the quantity and

  quality of medical research.

  16.5.3 Explanations

  European regulators are examining the idea of a “right to an explana-

  tion.” Suppose information about a consumer is fed into a model to predict

  whether or not he or she will default on a loan. If the consumer is refused

  the loan, are they owed an “explanation” of why? If so, what would count as

  an explanation? Can an organization keep a predictive model secret because

  if it were revealed it could be manipulated? A notable example is the Dis-

  criminant Inventory Function. better known as the DIF function that the

  IRS uses to trigger audits. Is it legitimate to reverse engineer the DIF func-

  tion? See CAvQM (2011) for a collection of links on the DIF function.

  Can we demand more of an ML model than we can of a person? Sup-

  pose we show you a photo and that you correctly identify it as a picture of

  your spouse. Now we ask, “how do you know?” The best answer might be

  “because I’ve seen a lot of pictures that I know are pictures of my spouse,

  and that photo looks a lot like those pictures!” Would this explanation be

  satis factory coming from a computer?

  16.6 Summary

  This chapter has only scratched the surface of how AI and ML might

  impact industrial structure. The technology is advancing rapidly, with the

  main bottleneck now being analysts who can implement these machine-

  learning systems. Given the huge popularity of college classes in this area

  and the wealth of online tutorials, we expect this bottleneck will be alleviated

  in the next few years.

  References

  Acquisti, Alessandro, Curtis R. Taylor, and Liad Wagman. 2016. “The Economics

  of Privacy.” Journal of Economic Literature 52 (2).

  Acquisti, Alessandro, and Hal Varian. 2004. “Conditioning Prices on Purchase His-

  tory.” Marketing Science 24 (4): 367– 81.

  418 Hal Varian

  Anderson, Ross. 1993. “Why Cryptosystems Fail.” Proceedings of the 1st ACM Con-

  ference on Computer and Communtications Security. https:// dl.acm .org/ citation

  .cfm?id=168615.

  ———. 2017. “Economics and Security Resource Page.” Working paper, Cambridge

  University. http:// www .cl .cam.ac .uk/ ~rja14/ econsec .html.

  Arrow, Kenneth J. 1962. “The Economic Implications of Learning by Doing.”

  Review of Economic Studies 29 (3): 155– 73.

  Axelrod, Robert. 1984. The Evolution of Cooperation. New York: Basic Books.

  Bessen, James. 2016. Learning by Doing: The Real Connection between Innovation,

  Wages, and Wealth. New Haven, CT: Yale University Press.

  ———. 2017. “Information Technology and Industry.” Law and Economics

  Research Paper no. 17-41, Boston University School of Law.

  Borenstein, Severin. 1997. “Rapid Communication and Price Fixing: The Airline

  Tariff Publishing Company Case.” Working paper. http:// faculty.haas.berkeley

  .edu/ borenste/ download/ atpcase1 .pdf.

  Bughin, Jacques, and Erik Hazan. 2017. “The New Spring of Artifi cial Intelli-

  gence.” Vox CEPR Policy Portal. https:// voxeu .org/ article/ new- spring- artifi cial

  - intelligence- few- early- economics.

  CavQM. 2011. “Reverse Engineering The IRS DIF- Score. ” Comparative Advan-

  tage via Quantitative Methods blog, July 10. http:// cavqm.blogspot .com/ 2011/ 07

  / reverse- engineering- irs- dif- score .html.

  Christie, William G., and Paul H. Schultz. 1995. “Did Nasdaq Market Makers

  Implicitly Collude?” Journal of Economic Perspectives 9 (3): 199– 208.

  DellaVigna, Stefano, and Matthew Gentzkow. 2017. “Uniform Pricing in US Retail

  Chains.” NBER Working Paper no. 23996, Cambridge, MA.

  Dubé, Jean- Pierre, and Sanjog Misra. 2017. “Scalable Price Targeting.” NBER

  Working Paper no. 23775, Cambridge, MA.

  Eckersley, Peter, and Yomna Nassar. 2017. “Measuring the Progress of AI Research.”

  Electronic Frontier Foundation. https:// eff .org/ ai/ metrics.

  Etzioni, Oren, Rattapoom Tuchinda, Craig Knoblock, and Alexander Yates. 2003.

  “To Buy or Not to Buy: Mining Airfare Data to Minimize Ticket Purchase Price.”

  Proceedings of the Ninth ACM SIGKDD International Conference on Knowledge

  Discovery and Data Mining. www .doi .org/ 10.1145/ 956750.956767.

  Ezrachi, A., and M. E. Stucke. 2017. “Algorithmic Collusion: Problems and

  Counter- Measures—Note.” OECD Roundtable on Algorithms and Collusion.

  https:// www .oecd .org/ offi

  cialdocuments/ publicdisplaydocumentpdf/ ?cote=DAF

  / COMP/ WD%282017%2925&docLanguage=En.

  Fudenberg, Drew. 1992. “Explaining Cooperation and Commitment in Repeated

  Games.” In Advances in Economic Theory: Sixth World Congress, Econometric

  Society Monographs, edited by Jean- Jacques Laff ont. Cambridge, MA: Cam-

  bridge University Press.

  Goodfellow, Ian, Nicolas Papernot, Sandy Huang, Yan Duan, Pieter Abbeel, and

  Jack Clark. 2017. “Attacking Machine Learning with Adversarial Examples.”

  OpenAI blog, Feb. 26. https:// blog.openai .com/ a
dversarial- example- research/.

  Hutson, Matthew. 2017. “Will Make AI Smarter for Cash.” Bloomberg Business

  Week, Sept. 11.

  Kurakin, Alexy, Ian Goodfellow, and Samy Bengio. 2016. “Adversarial Examples in

  the Physical World.” Cornell University Library, ArXiv 1607.02533. https:// arxiv

  .org/ abs/ 1607.02533.

  Organisation for Economic Co- operation and Development (OECD). 2017. “Algo-

  rithms and Collusion: Competition Policy in the Digital Age.” www .oecd .org

  / competition/ algorithms- collusion- competition- policy- in-the- digital- age .htm.

  Comment 419

  Pierce, David G. 1992 “Repeated Games: Cooperation and Rationality.” In Advances

  in Economic Theory: Sixth World Congress, Econometric Society Monographs,

  edited by Jean- Jacques Laff ont. Cambridge, MA: Cambridge University Press.

  Rubinstein, Arial. 1986. “Finite Automata Play the Repeated Prisoner’s Dilemma.”

  Journal of Economic Theory 39:83– 96.

  Segal, Ilya. 2003. “Optimal Pricing Mechanisms with Unknown Demand.” American

  Economic Review 93 (3): 509– 29.

  Shiller, Benjamin Reed. 2013. “First Degree Price Discrimination Using Big Data.”

  Working Paper no. 58, Department of Economics and International Business

  School, Brandeis University.

  Spiegel, Yossi, and Igal Hendel. 2014. “Small Steps for Workers, A Giant Leap for

  Productivity.” American Economic Journal: Applied Economics 6 (1): 73– 90.

  Sreevallabh, Chivukula, and Wei Liu. 2017. “Adversarial Learning Games with Deep

  Learning Models.” International Joint Conference in Neural Networks. www .doi

  .org/ 10.1109/ IJCNN.2017.7966196.

  Stiglitz, Joseph E., and Bruce C. Greenwald. 2014. Creating a Learning Society. New York: Columbia University Press.

  Varian, H. 2000. “Managing Online Security Risks.” New York Times, June 1.

  Comment Judith Chevalier

  Varian provides an excellent overview of industrial organization issues aris-

  ing out of the adoption of machine learning and artifi cial intelligence. A

  number of these issues have potential competition policy implications. For

  example, exploitation of AI technologies may either increase or decrease

  economies of scale, leading potentially to situations of market power. Own-

  ership of data, if crucial to competition in a specifi c industry, may create

  barriers to entry. The potential for algorithmic collusion clearly leads to

  antitrust enforcement concerns. Here, I briefl y address one of these issues,

  data ownership, and highlight some potential antitrust policy responses.

  While I focus here on data ownership as a barrier to entry, some of the policy

  trade- off s I discuss are germane to the other potential market structure

  changes highlighted in Varian.

  Artifi cial intelligence and machine- learning processes often use raw data

  as an input. As Varian points out, it is not at all clear that data defi es our

  usual expectation that a scarce asset or resource will eventually face decreas-

  ing returns to scale. Nonetheless, one can certainly imagine circumstances

  where exclusive ownership of a body of data will create a nearly insurmount-

  able advantage to a market incumbent. While the concern that access to a

  Judith Chevalier is the William S. Beinecke Professor of Finance and Economics at the

  Yale School of Management and a research associate of the National Bureau of Economic Research.

  For acknowledgments, sources of research support, and disclosure of the author’s material fi nancial relationships, if any, please see http:// www .nber .org/ chapters/ c14033.ack.

  420 Judith Chevalier

  scarce asset creates entry barriers may be relatively new as it applies to data,

  the underlying fundamental economic issue is not new. Antitrust authori-

  ties in all jurisdictions have long wrestled with optimal policy toward fi rms

  for which the ownership of scarce assets creates barriers to entry. In the

  United States, analysis of this issue dates back at least to United States v.

  Terminal Railroad Assocation (224 US 383 (1912), a case in which consortia

  of railroads denied rival access to the only railroad bridges traversing the

  St. Louis River. In that case and subsequent ones, courts have occasion-

  ally articulated a duty to deal for a fi rm with market power that controls

  access to an asset (or facility) that is essential to competition and for which

  it is impractical for rivals to duplicate the asset. However, determining the

  precise circumstances under which a monopolist has an affi

  rmative duty to

  deal with a rival remains an unsettled area of antitrust law.

  In principle, this very kind of antitrust essential facilities doctrine could

  be applied to data ownership. Indeed, while Varian remains silent on the

  issue of remedies, recent legal literature in the United States has shown

  some enthusiasm for essential facilities doctrine as applied to data (see, e.g.,

  Meadows 2015; Abrahamson 2014). Further, European antitrust authorities

  have begun to articulate principles for the control of big data that suggest

  an essential facilities doctrine. For example, Margrethe Vesteger (2016), the

  EU Commissioner for Competition, recently stated in a speech “it’s true

  that we shouldn’t be suspicious of every company which holds a valuable

  set of data. But we do need to keep a close eye on whether companies con-

  trol unique data, which no one else can get hold of, and can use it to shut

  their rivals out of the market.” In the speech, she highlighted a 2014 case in

  which the French competition authority required a French energy producer,

  GDF Suez, to share a customer list with industry rivals.

  Despite enthusiasm in some quarters, the application of essential facilities

  doctrine to data sharing creates both important trade- off s and important

  practical concerns. I begin with the trade- off s. In evaluating antitrust poli-

  cies in innovative industries, it is important to recognize that consumer bene-

  fi ts from new technologies arise not just from obtaining goods and services at

  competitive prices, but also from the fl ow of new and improved products and

  services that arise from innovation. Thus, antitrust policy should be evalu-

  ated not just in terms of its eff ect on prices and outputs, but also on its eff ect

  on the speed of innovation. Indeed, in high- technology industries, it seems

  likely that these dynamic effi

  ciency considerations dwarf the static effi

  ciency

  considerations. In the case of an application of the essential facilities doc-

  trine to data, the trade- off s are numerous and they are directionally unclear.

  An often- cited criticism of essential facilities doctrine is that creating an

  ex post duty to share diminishes the incentive to invest in the essential facil-

  ity in the fi rst place (see, e.g., Pate 2006). In this case, creating an ex post

  duty to share data could diminish the incumbent incentive to invest in data

  creation, thus slowing the pace of innovation. However, the overall incentive

  Comment 421

  trade- off s are not as simple as that. In circumstances in which new entrants

  are an important source of potential innovation, exclusionary conduct by

  incumbents that reduces t
he incentive of entrants to invest in R&D can

  slow the pace of innovation. That is, in the case of data, if particular data

  is an essential complement to an AI innovation, exclusive ownership of the

  data by an incumbent can slow the pace of innovation by entrants. Issues

  of the impact of antitrust enforcement on the pace of innovation remains a

  nascent area of research, but is explored theoretically in, for example, Segal

  and Whinston (2007). Thus, in sum, while a broad application of the essen-

  tial facilities doctrine to proprietary data may be tempting from an ex post

  static effi

  ciency perspective, caution about ex ante incentives is warranted.

  In addition to the trade- off s already discussed, any application of an

  essential facilities doctrine to data sharing also implies a host of practical

  considerations. As in any essential facilities scenario, once a court or anti-

  trust authority establishes a duty to deal, it must also articulate terms of

  trade. Clearly, absent some articulation of terms, an incumbent can de facto

  refuse to deal by establishing transaction terms that are unattractive to any

  potential rival user of the data. Given that market conditions are continually

  changing, an ongoing regulation of the terms of trade will become unavoid-

  able. There are certainly instances in which US courts have become ongoing

  regulators of the transactions of companies for which a court has imposed a

  duty to deal. The continuing oversight of the contracts of the music licens-

  ing fi rms ASCAP and BMI are good examples of a duty to deal leading to

  de facto regulation by the courts. However, the creation of such an ongoing

  regulatory structure brings with it costs to both the regulatory entity and

  the regulated fi rms. Essential facilities is not a quick fi x.

  Finally, while essential facilities doctrine may not always be the best

  tool for addressing data whose ownership has become concentrated, the

  potential for mergers to create importantly concentrated data should be

  considered in merger analysis, just as merger analysis considers the poten-

  tial for mergers to substantially concentrate some other element of produc-

  tive capacity.

  Clearly, there are important trade- off s in implementing antitrust solu-

  tions to the problems potentially created by exclusive ownership of key data.

 

‹ Prev