Google's, Apple's, and other companies' automated assistants are increasingly serving as personal shoppers. These digital intermediaries will save us time by purchasing grocery items, transferring bank accounts, and subscribing to cable. The literature has only begun to hint at the paradigm shift needed to navigate the legal risks and rewards of this coming era of automated commerce. This Article begins to fill that gap by surveying legal battles related to contract exit, data access, and deception that will determine the extent to which automated assistants are able to help consumers to search and switch, potentially bringing tremendous societal benefits. Whereas observers have largely focused on protecting consumers and sellers from digital intermediaries' market power, sellers like Amazon, Comcast, and Wells Fargo can also harm consumers by obstructing automated assistants. Advancing consumer welfare in the automated era requires not just consumer protection, but digital intermediary protection.
The Article also shows the unpredictable side of eliminating switching costs. If digital assistants become pervasive, they could gain the ability to rapidly direct millions of consumers to new purchases whenever a lower price or new innovation becomes available. Significantly accelerated consumer switching--what I call hyperswitching--does not inevitably harm society. But in the extreme it could make some large markets more volatile, raising unemployment costs or financial stability concerns as more firms fail. This new kind of disruption could pose challenges for commercial and banking regulators akin to those familiar to securities regulators, who deploy idiosyncratic tools such as a pause button for the stock market. Even if sellers prevent extreme hyperswitching, managers may strategically prepare for hyperswitching with economically costly behavior such as hoarding liquid assets or forming conglomerates to provide insurance against a sudden exodus of customers. The transaction-cost-focused literature has missed macro-level drawbacks.
The regulatory architecture reflects these scholarly gaps. One set of agencies regulates automated assistants for consumer protection and antitrust violations but does not go beyond those microeconomic inquiries. Nor do they prioritize strengthening digital intermediaries. Regulators with more macroeconomic missions lack jurisdiction over automated assistants. The intellectual framework and regulatory architecture should expand to encompass both the upsides and downsides of digital consumer sovereignty.
TABLE OF CONTENTS INTRODUCTION I. AN OVERVIEW OF AUTOMATED COMMERCE MARKET DYNAMICS A. The AI Business Model B. Common Features II. THE UPSIDES OF AI PROTECTION A. The Theory Supporting Digitally Perfected Competition 1. The Policy Push Toward Perfect Competition 2. AIs as Agents of Perfect Competition B. Legal and Market Battlegrounds 1. Data Obstruction 2. Exit Prevention 3. Obfuscation 4. Collusion C. Summary III. THE RISKS OF HYPERSWITCHING A. Hyperswitching as a New Form of Disruption 1. Lasting Disruption 2. Faster Disruption 3. Larger-Scale Disruption B. Managerial Responses 1. Capitalization 2. Conglomeration 3. Consolidation and Collusion C. Market Volatility 1. The Structure of Past Financial Instability 2. Financial Product Instability 3. Real Economy Turbulence D. Limits to Hyperswitching E. Summary IV. IMPLICATIONS A. Shifting the Paradigm for Consumer Switching 1. Recognizing the Upsides of Automated Switching 2. Recognizing the Full Downsides of Automated Commerce B. Redesigning the Regulatory Structure 1. Adding More Micro to Financial Regulation 2. Adding More Macro to Trade Regulation CONCLUSION INTRODUCTION
The world's largest companies, including Apple, Google, and Microsoft, are racing to develop artificially intelligent butlers (AIs). They aim to allow consumers to outsource the tasks of opening a new bank account, locating cheaper laundry detergent, and finding the highest quality groceries. (1) By way of illustration, a consumer might at any moment receive a phone alert from her AI. The subsequent conversation could unfold as follows:
SIRI. As part of my regular monitoring of your spending, I have located an opportunity to save money on your phone bill and on your grocery bill each month. Would you like to hear more?
CONSUMER Tell me about the phone bill.
SIRI. Based on your monthly data usage and the performance of the networks where you spend most of your time, you can receive comparable service through Sprint at $140 less per year. Let me know if you want to hear more. Or, if you would like me to switch your account, place your thumbprint on the phone.
The imminent technological possibility that machines could take over most of the consumer's purchase process calls for a reexamination of the framework for market intervention. This Article expands on the literature beginning that undertaking in three main ways. The first is to show the broader legal reforms and intellectual shifts that would help AIs to reduce transaction costs. (2) Today, even when competing products are available at the click of a button, shoppers regularly fail to locate the best deal. (3) People simply may not want to spend the time clicking on various sites and calculating the differences. Sellers also make shopping more difficult by burying the lowest price option in the second page of search results, where few look, or by hiding costs through fees or add-on products, such as expensive printer ink. (4) The recent scholarship on digital intermediaries has largely focused on preventing technology firms from adding harm--which they might do by directly manipulating users or indirectly exercising market power over sellers. (5) Consumer harms from sellers' behavior toward AIs have received less attention. (6) But consumers are also harmed when, for instance, sellers deceive the AI into giving bad advice to the consumer or prevent AIs from having full access to market data. Counterintuitively, consumer welfare may depend on even powerful technology companies benefitting from the types of laws traditionally deployed to protect consumers.
Part of the challenge in seeing the need for such policies may be a limited sense of the potential magnitude of efficiency gains. (7) Before Ronald Coase's contributions to law and economics, models assumed an absence of transaction costs. (8) The modern high-transaction-cost paradigm results from generations of Nobel Prize-winning work by Coase and others showing that real-world markets instead face high transaction costs that impede switching, including from information asymmetries and behavioral biases. (9) Those intellectual contributions have made models more accurate but have also normalized high transaction costs because it has proven difficult to improve markets meaningfully as long as consumers still play an active role. (10) Through that lens, an observer would see the opportunity for significant market improvements but would not be surprised by markets with minimal searching and switching. With an alternative baseline of automated markets, minimal searching and switching should prompt inquiry into potential AI barriers. In other words, the intellectual framework for automated commerce should recognize that through AI protection laws, real markets can move closer to those in discarded pre-Coasian models.
The Article's second contribution is to expand upon the downsides associated with AIs, and with hyperswitching in particular. Once they take hold, AIs could potentially constitute a new form of disruption by collapsing firms more quickly, reaching a larger portion of the economy, and extending uncertainty longer. Assume, for instance, that the above digital advice to switch cell phone carriers was given by Apple's Siri, which operates about 54% of mobile operating systems. (11) If a quarter of Siri's customers were to switch cell phone carriers upon locating savings, this would amount to a mass exodus of customers from several large Fortune 500 companies, such as Verizon and AT&T. But the cell phone carrier benefitting from that advice could lose in the next round of advice. Mass departures are more likely if the AI market follows that of other highly concentrated digital markets such as smartphones, in which Apple and Google together provide 99% of mobile operating systems. (12)
To be clear, extreme hyperswitching may never occur in many industries. Besides the possibility that entrenched firms will strive to prevent it, markets also have some inherent limits, such as capacity constraints on sellers' ability to service large numbers of new customers on short notice. (13) But given the stakes, as AIs gain influence, other companies like Citibank, AT&T, and Clorox may need to prepare strategically for the possibility of hyperswitching--and some have already begun to do so. (14) Predictable moves include forming conglomerates to diversify revenues; hoarding cash or other liquid assets to provide insurance in case revenues drop; and colluding with other firms on price so that AIs have little basis to redirect consumers. Those changes are important to consider because they implicate policy decisions, and they could significantly lessen the efficiency gains that motivate regulators and scholars to design laws empowering digital intermediaries. (15) Thus, even if hyperswitching never arrives, large companies' fears of it could reshape industrial organization, capital markets, and the economy.
Another set of downsides relates to the volatility that could result if extreme hyperswitching materializes. Numerous scholars have concluded that financial technology innovation poses a threat to economic stability. (16) That literature has extensively analyzed automated stock trading but has yet to explore the related risks of consumer AIs in any sustained manner. (17) Yet byproducts of consumer financial activity, like stock market volatility...