Artificial Intelligence and the Future of Competition Policy with Catherine Tucker

Episode Summary

Executive Summary: Catherine Tucker argues that AI should be understood economically as falling prediction costs, but current generative AI differs from prior digital tech because it combines high fixed and marginal costs. She says this may be temporary, yet it creates major uncertainty for competition policy, enforcement, pricing, and evidentiary standards as antitrust agencies confront AI-generated documents and algorithmic decision-making.

Main Topics: AI as falling prediction costs (Priority: 5/5): Tucker explains the field’s shift from viewing AI as robots to viewing it as prediction, which makes economics central because the key question is what happens when prediction becomes cheaper. AI’s cost structure differs from traditional digital economics (Priority: 5/5): Unlike earlier digital technologies built around near-zero marginal costs, current AI involves expensive training, processing, and talent, creating a hybrid of high fixed and marginal costs that may be temporary. Monetization and metering remain unsettled (Priority: 4/5): Tucker argues that AI firms have not yet discovered durable business models, and current usage limits and metering feel inconsistent with the abundance model associated with digital tech. Competition structure may be less winner-take-all than feared (Priority: 4/5): She cautions against assuming AI will produce permanent concentration; disruption and shifting costs may allow new firms or architectures to emerge, especially as firms adapt with mini-models and on-device processing. AI will change antitrust evidence and enforcement (Priority: 5/5): A major theme is that AI-generated text may weaken the traditional reliance on “hot documents,” making intent harder to prove and requiring antitrust agencies to rely more on technical auditing and expert analysis. Algorithmic pricing complicates collusion analysis (Priority: 4/5): She says existing academic models of algorithmic collusion are too detached from real industry practice, and enforcement against AI-driven pricing will require deeper understanding of how pricing algorithms function. Need for more economists and technical expertise (Priority: 5/5): Tucker worries the policy world is advancing faster than the research base and that competition authorities lack enough economists, computer scientists, and data science capacity to evaluate AI robustly.

Key Arguments: Economists ignored AI for years because it was misframed as robotics, but AI is fundamentally about prediction and therefore about changes in cost structures. Current AI differs from classic digital economics because it appears to have both high fixed costs and high marginal costs, especially for model training and processing. These high costs are likely temporary because computing costs tend to fall quickly and labor markets adjust through upskilling. Usage limits in chatbots illustrate metering and marginal cost pressures, which are foreign to the abundance logic of prior digital goods. AI firms still have not solved monetization; advertising is possible but not yet a clear or stable model. The short-run industry may look concentrated, but long-run winner-take-all dynamics are not inevitable given rapid technological change and uncertainty about winning architectures. In antitrust, AI will reduce the usefulness of employee emails and other “hot documents” as intent evidence because much text will be machine-generated. Competition agencies will need more technical staff and better auditing tools to understand how algorithms behave in practice, not just in theory. Algorithmic pricing and collusion analysis cannot rely only on lab simulations; enforcement must engage with real-world algorithm behavior. A major risk is confusion between data as an input and output, and between digital-economy assumptions and generative-AI realities, which may lead to inconsistent policy decisions.

Data Points: Time since economists shifted attention to AI: about 7 years - Tucker says the profession only recently realized AI is about prediction rather than robots. Podcast date: Thursday, April 25th, 2024 - The episode introduction states the recording date. Paper publication venue: Oxford Review of Economic Policy - Tucker mentions her forthcoming paper on AI and competition policy. Years leading digital economics and AI group: 10 years - She says she has headed the NBER digital economics and AI group for a decade. Historical benchmark for digital-economy thinking: 1998 to 2002 - She compares current AI monetization uncertainty to the early internet monetization debate. Reference period for platform predictions: 2005 - She notes that in 2005 few would have predicted YouTube or Facebook’s eventual dominance. Reference to current chatbot access limits: 7 uses left until 1pm - Used as an example of AI metering and marginal-cost-based rationing. Reference to user age in example: 16 year old daughter - Used in an example contrasting narrow predictive AI with broad generative uses.

Pivotal Quotes: "AI wasn’t about robots. It wasn’t about Judgment Day. Instead, AI was about prediction." — Catherine Tucker: She defines the modern economic framework for understanding AI. "We seem to have both high fixed costs and potentially high marginal costs." — Catherine Tucker: She explains why current AI cost structure differs from earlier digital technologies. "What scares me most about this sort of change is it’s a completely big shift in process. It’s not a shift in tools, it’s a shift in process." — Catherine Tucker: She describes how AI will alter antitrust enforcement and institutional practice.

Implications: AI may reshape competition policy more than market structure in the short run, forcing regulators to build technical capacity, rethink evidence standards, and avoid overconfident assumptions about concentration, collusion, or data scarcity.