The Economics of Artificial Intelligence

Home > Other > The Economics of Artificial Intelligence > Page 98
The Economics of Artificial Intelligence Page 98

by Ajay Agrawal


  (RFF) feature through which sellers can set a rebate value for any item

  they sell (cash back or store coupon) as a reward for a buyer’s feedback.

  If a seller chooses this option, then Taobao guarantees that the rebate will

  be transferred from the seller’s account to a buyer who leaves high- quality

  feedback. Importantly, feedback quality only depends on how informative it

  is, rather than whether the feedback is positive or negative. Taobao measures

  the quality of feedback with a NLP algorithm that examines the comment’s

  content and length and fi nds out whether key features of the item are men-

  tioned. Hence, the marketplace manages the market for feedback by forcing

  the seller to deposit at Taobao a certain amount for a chosen period, so that

  funds are guaranteed for buyers who meet the rebate criterion, which itself

  is determined by Taobao.6

  6. According to a Taobao survey (published in March 2012), 64.8 percent of buyers believed that they will be more willing to buy items that have the RFF feature, and 84.2 percent of buyers believed that the RFF option will make them more likely to write detailed comments.

  How AI and Machine Learning Can Impact Market Design 581

  Taobao’s motivation behind the RFF mechanism was to promote more

  informative feedback, but as Li, Tadelis, and Zhow (2016) noted, economic

  theory off ers some insights into how the RFF feature can act as a potent

  signaling mechanism that will further separate higher- from lower- quality

  sellers and products. To see this, recall the literature launched by Nelson

  (1970) who suggested that advertising acts as a signal of quality. According

  to the theory, advertising—which is a form of burning money—acts as a

  signal that attracts buyers who correctly believe that only high- quality sellers

  will choose to advertise. Incentive compatibility is achieved through repeat

  purchases: buyers who purchase and experience the products of advertisers

  will return in the future only if the goods sold are of high enough quality.

  The cost of advertising can be high enough to deter low- quality sellers from

  being willing to spend the money and sell only once because those sellers

  will not attract repeat customers, and still low enough to leave profi ts for

  higher- quality sellers. Hence, ads act as signals that separate high- quality

  sellers, and in turn attract buyers to their products.

  As Li, Tadelis, and Zhow (2016) argue, the RFF mechanism plays a

  similar signaling role as ads do. Assuming that consumers express their ex-

  periences truthfully in written feedback, any consumer who buys a product

  and is given incentives to leave feedback will leave positive feedback only

  if the buying experience was satisfactory. Hence, a seller will off er RFF

  incentives to buyers only if the seller expects to receive positive feedback,

  and this will happen only if the seller will provide high quality. If a seller

  knows that their goods and services are unsatisfactory, then paying for feed-

  back will generate negative feedback that will harm the low- quality seller.

  Equilibrium behavior then implies that RFF, as a signal of high quality, will

  attract more buyers and result in more sales. The role of AI was precisely

  to reward buyers for information, not for positive feedback.

  Li, Tadelis, and Zhou (2016) proceeded to analyze data from the period

  where the RFF mechanism was featured and confi rmed that fi rst, as ex-

  pected, more feedback was left in response to the incentives provided by the

  RFF feature. More important, the additional feedback did not exhibit any

  biases, suggesting that the NLP algorithms used were able to create the kind

  of screening needed to select informative feedback. Also, the predictions of

  the simple signaling story were borne out in the data, suggesting that using

  NLP to support a novel market for feedback did indeed solve both the free-

  rider problem and the cold- start problem that can hamper the growth of

  online marketplaces.

  23.4 Using AI to Reduce Search Frictions

  An important application of AI and machine learning in online market-

  places is the way in which potential buyers engage with the site and proceed

  to search for products or services. Search engines that power the search of

  582 Paul R. Milgrom and Steven Tadelis

  products online are based on a variety of AI algorithms that are trained

  to maximize what the provider believes to be the right objective. Often this

  boils down to conversion, under the belief that the sooner a consumer con-

  verts a search to a purchase, the happier the consumer is both in the short

  and the long run. The rationale is simply that search itself is a friction, and

  hence, maximizing the successful conversion of search activity to a purchase

  reduces this friction.

  This is not inconsistent with economic theory that has modeled search

  as an inevitable costly process that separates consumers from the products

  they want. The canonical search models in economics either build on the

  seminal work of Stigler (1961), who assumes that consumers sample a fi xed

  number of stores and choose to buy the lowest priced item, or more often,

  on the models of McCall (1970) and Mortensen (1970), who posit that

  a model of sequential search is a better description of consumer search

  behavior. In both modeling approaches consumers know exactly what they

  wish to buy.

  However, it turns out that unlike the simplistic models of search employed

  in economic theory, where consumers know what they are looking for and

  the activity of search is just a costly friction, in reality, people’s search behav-

  ior is rich and varied. A recent paper by Blake, Nosko, and Tadelis (2016)

  uses comprehensive data from eBay to shed light on the search process with

  minimal modeling assumptions. Their data show that consumers search

  signifi cantly more than other studies—which had limited access to search

  behavior over time—have suggested.

  Furthermore, search often proceeds from the vague to the specifi c. For

  example, early in a search a user may use the query “watch,” then refi ne it to

  “men’s watch,” and later add further qualifying words such as color, shape,

  strap type, and more. This suggests that consumers often learn about their

  own tastes, and what product characteristics exist, as part of the search

  process. Indeed, Blake et al. (2016) show that the average number of terms

  in the query rises over time, and the propensity to use the default- ranking

  algorithm declines over time as users move to more focused searches like

  price sorting.

  These observations suggest that marketplaces and retailers alike could

  design their online search algorithms to understand search intent so as to

  better serve their consumers. If a consumer is in the earlier, exploratory

  phases of the search process, then off ering some breadth will help the con-

  sumer better learn their tastes and the options available in the market. But

  when the consumer is driven to purchase something particular, off ering a

  narrower set of products that match the consumer’s preferences would be

  better. Hence, machine learning and AI can play an instrumental role in

>   recognizing customer intent.

  Artifi cial intelligence and machine learning cannot only help predict a

  customer’s intent, but given the large heterogeneity on consumer tastes, AI

  How AI and Machine Learning Can Impact Market Design 583

  can help a marketplace or retailer better segment the many customers into

  groups that can be better served with tailored information. Of course, the

  idea of using AI for more refi ned customer segmentation, or even personal-

  ized experiences, also raises concerns about price discrimination. For ex-

  ample, in 2012 the Wall Street Journal reported that “Orbitz Worldwide Inc.

  has found that people who use . . . Mac computers spend as much as 30%

  more a night on hotels, so the online travel agency is starting to show them

  diff erent, and sometimes costlier, travel options than Windows visitors see.

  The Orbitz eff ort, which is in its early stages, demonstrates how tracking

  people’s online activities can use even seemingly innocuous information—in

  this case, the fact that customers are visiting Orbitz .com from a Mac—to

  start predicting their tastes and spending habits.”7

  Whether these practices of employing consumer data and AI will help or

  harm consumers is not obvious, as it is well known from economic theory

  that price discrimination can either increase or reduce consumer welfare. If,

  on average, Mac users prefer staying at fancier and more expensive hotels

  because owning a Mac is correlated with higher income and tastes for luxury,

  then the Orbitz practice is benefi cial because it shows people what they want

  to see and reduces search frictions. However, if this is just a way to extract

  more surplus from consumers who are less price sensitive, but do not neces-

  sarily care for the snazzier hotel rooms, then it harms these consumers.

  There is currently a lot of interest in policy circles regarding the poten-

  tial harms to consumers from AI- based price discrimination and market

  segmentation. McSweeny and O’Dea (2017) suggest that once AI is used to

  create more targeted market segments, this may not only have implications

  only for consumer welfare, but for antitrust policy and market defi nitions for

  mergers. But, as Gal and Elkin- Koren (2017) suggest, the same AI- targeting

  tools used by retailers and marketplaces to better segment consumers may

  be developed into tools for consumers that will help them shop for better

  deals and limit the ways in which marketplaces and retailers can engage in

  price discrimination.

  23.5 Concluding

  Remarks

  In its early years, classical economic theory paid little attention to market

  frictions and treated information and computation as free. That theory led

  to conclusions about effi

  ciency, competitive prices for most goods, and full

  employment of valuable resources. To address the failures of that theory,

  economists began to study models with search frictions, which predicted that

  price competition would be attenuated, that some workers and resources

  7. See “On Orbitz, Mac Users Steered to Pricier Hotels,” Dana Mattioli, The Wall Street Journal, Aug. 23, 2012. https:// www .wsj .com/ articles/ SB1000142405270230445860457748882

  2667325882.

  584 Paul R. Milgrom and Steven Tadelis

  could remain unemployed, and that it could be costly to distinguish reliable

  trading partners from others. They also built markets for complex resource-

  allocation problems in which computations and some communications were

  centralized, lifting the burden of coordination from individual market par-

  ticipants.

  With these as the key frictions in the traditional economy, AI holds enor-

  mous potential to improve effi

  ciency. In this chapter, we have described

  some of the ways that AI can overcome computational barriers, reduce

  search frictions, and distinguish reliable partners. These are among the most

  important causes of ineffi

  ciency in traditional economies, and there is no

  longer any question that AI is helping to overcome them, with the promise

  of widespread benefi ts for all of us. As Roth (2002) noted, market designers

  “cannot work only with the simple conceptual models used for theoretical

  insights into the general working of markets. Instead, market design calls

  for an engineering approach.” Artifi cial intelligence has already proven to

  be a valuable tool in the economist- as-engineer tool box.

  References

  Blake, T., C. Nosko, and S. Tadelis. 2016. “Returns to Consumer Search: Evidence

  from eBay.” 17th ACM Conference on Electronic Commerce (EC 2016), 531– 45.

  Filippas, A., J. J. Horton, and J. M. Golden. 2017. “Reputation in the Long- Run.”

  Working paper, Stern School of Business, New York University.

  Gal, M. S., and N. Elkin- Koren. 2017. “Algorithmic Consumers.” Harvard Journal

  of Law & Technology 30:1– 45.

  Hutter, F., H. Hoos, and K. Leyton- Brown. 2011. “Sequential Model- Based Opti-

  mization for General Algorithm Confi guration.” Conference paper, International

  Conference on Learning and Intelligent Optimization. https:// link.springer .com

  / chapter/ 10.1007/ 978-3-642-25566-3_40.

  Leyton- Brown, K., P. R. Milgrom, and I. Segal. 2017. “Economics and Computer

  Science of a Radio Spectrum Reallocation.” Proceedings of the National Academy

  of Sciences 114 (28): 7202– 09. www .pnas .org/ cgi/ doi/ 10.1073/ pnas.1701997114.

  Li, L. I., S. Tadelis, and X. Zhou. 2016. “Buying Reputation as a Signal of Qual-

  ity: Evidence from an Online Marketplace.” NBER Working Paper no. 22584,

  Cambridge, MA.

  Masterov, D. V., U. F. Mayer, and S. Tadelis. 2015. “Canary in the E-commerce

  Coal Mine: Detecting and Predicting Poor Experiences Using Buyer- to-Seller

  Messages.” In Proceedings of the Sixteenth ACM Conference on Economics and

  Computation, EC ‘15, 81– 93.

  Mayzlin, D., Y. Dover, and J. Chevalier. 2014. “Promotional Reviews: An Empirical

  Investigation of Online Review Manipulation.” American Economic Review 104

  (8): 2421– 55.

  McCall, J. J. 1970. “Economics of Information and Job Search.” Quarterly Journal

  of Economics 84 (1): 113– 26.

  McSweeny, T., and B. O’Dea. 2017. “The Implications of Algorithmic Pricing for

  How AI and Machine Learning Can Impact Market Design 585

  Coordinated Eff ects, Analysis and Price Discrimination Markets in Antitrust

  Enforcement.” Antitrust 32 (1): 75– 81.

  Milgrom, P. R. 2017. Discovering Prices: Auction Design in Markets with Complex

  Constraints. New York: Columbia University Press.

  Milgrom, P. R., D. C. North, and B. R. Weingast. 1990. “The Role of Institutions

  in the Revival of Trade: The Law Merchant, Private Judges, and the Champagne

  Fairs.” Economics and Politics 2 (1): 1– 23.

  Milgrom, P. R., and I. Segal. 2017. “Deferred Acceptance Auctions and Radio Spec-

  trum Reallocation.” Working paper.

  Mortensen, D. T. 1970. “Job Search, the Duration of Unemployment and the Phil-

  lips Curve.” American Economic Review 60 (5): 847– 62.

  Myerson, R. B. 1981. “Optimal Auction Design.” Mathematics of Operations

  Research 6 (1)
: 58– 73.

  Nelson, P. 1970. “Information and Consumer Behavior.” Journal of Political

  Economy 78 (2): 311– 29.

  Newman, N., A. Fréchette, and K. Leyton- Brown. Forthcoming. “Deep Optimiza-

  tion for Spectrum Repacking.” Communications of the ACM (CACM).

  Nosko, C., and S. Tadelis. 2015. “The Limits of Reputation in Platform Markets:

  An Empirical Analysis and Field Experiment.” NBER Working Paper no. 20830,

  Cambridge, MA.

  Ostrovsky, M., and M. Schwartz. 2017. “Reserve Prices in Internet Advertising: A

  Field Experiment.” Working paper.

  Stigler, G. J. 1961. “The Economics of Information.” Journal of Political Economy

  69 (3): 213– 25.

  Roth, A. E. 2002. “The Economist as Engineer: Game Theory, Experimentation,

  and Computation as Tools for Design Economics.” Econometrica 70 (4): 1341– 78.

  Roth, A. E. 2015. Who Gets What—and Why: The New Economics of Matchmaking

  and Market Design. New York: Houghton Miffl

  in Harcourt.

  Vickrey, W. 1961. “Counterspeculation, Auctions, and Competitive Sealed Tenders.”

  Journal of Finance, American Finance Association 16 (1): 8– 37.

  Xu, H., D. Liu, H. Wang, and A. Stavrou. 2015. “E- commerce Reputation Manipu-

  lation: The Emergence of Reputation- Escalation- as-a- Service.” Proceedings of

  24th World Wide Web Conference (WWW 2015):1296– 306.

  Zervas, G., D. Proserpio, and J. W. Byers. 2015. “A First Look at Online Reputa-

  tion on Airbnb, Where Every Stay Is Above Average.” Working paper, Boston

  University.

  24

  Artifi cial Intelligence and

  Behavioral Economics

  Colin F. Camerer

  24.1 Introduction

  This chapter describes three highly speculative ideas about how artifi cial

  intelligence (AI) and behavioral economics may interact, particular in future

  developments in the economy and in research frontiers. First note that I will

  use the terms AI and machine learning (ML) interchangeably (although AI

  is broader) because the examples I have in mind all involve ML and predic-

  tion. A good introduction to ML for economists is Mullainathan and Spiess

  (2017), and other chapters in this volume.

  The fi rst idea is that ML can be used in the search for new “behavioral”-

  type variables that aff ect choice. Two examples are given, from experimen-

 

‹ Prev