Avi Goldfarb - The Economic Impact of AI - [Invest Like the Best, EP.321]

Episode Summary

Executive Summary: Patrick O’Shaughnessy interviews Avi Goldfarb about how falling prediction costs from AI will reshape business models, workflows, and labor. They argue AI’s biggest value comes from upskilling millions, enabling new system designs, and shifting power toward firms that control data, compute, and complementary assets.

Main Topics: AI as falling-cost prediction (Priority: 10/5): Goldfarb frames AI as prediction that becomes cheaper, so humans use it in more places. Point, application, and system solutions (Priority: 10/5): He distinguishes simple replacements from workflow changes and full business redesigns. Rule-based to decision-based organizations (Priority: 9/5): Better predictions let firms replace blanket rules with probabilistic decisions. Disruption and incumbent advantage (Priority: 9/5): AI creates openings where incumbents are constrained by old processes, incentives, and structure. Data, simulation, and digital twins (Priority: 8/5): Winning AI products need purpose-built data, often collected through products or simulation. Labor-market upskilling and inequality (Priority: 9/5): AI may broaden capability for millions while concentrating power among a few owners of capital.

Key Arguments: AI is cheapest where it automates high-skill tasks; that can upskill millions, not just replace workers. When prediction gets cheaper, organizations should redesign workflows, not just swap in software. Point solutions keep existing workflows; system solutions create entirely new business models. Rules dominate when information is poor; prediction enables decisions tailored to state and context. Good AI often requires complementary data, compute, and distribution—not just better models. Incumbents can fail even when they see disruption if their organizational structure blocks change. The biggest risk is not only job loss, but concentration of AI ownership and economic power.

Data Points: large language model example: GPT-3 - Goldfarb says he had seen earlier OpenAI models before ChatGPT transcripts library: 55,000 transcripts - Tegas ad read describing its research platform financial models: over 4,000 fully drivable financial models - Tegas ad read investment in Tegas: $20 million - Patrick’s firm Positive Sum invested to support Tegas percentage of U.S. workplaces and homes electrified: half - Electricity adoption took decades to reach widespread use time to broad electrification: 40 years - It took about four decades from patent to widespread household/workplace adoption taxi driver training in London: about three years - The Knowledge required for licensing in London

Pivotal Quotes: "My first reaction was, wow." — Avi Goldfarb: His reaction to the release of ChatGPT "Once someone writes that interface and figures it out, then the potential to upskill millions of people is amazing for the economy." — Avi Goldfarb: On ChatGPT-style tools helping non-writers "If you know what the right answer is, these models are incredibly helpful." — Avi Goldfarb: On why generative AI is most useful as an accelerator for experts

Implications: The open question is who captures the gains: entrepreneurs should build workflows around prediction, while policymakers and managers must address concentration and worker transition.