Odd Lots Podcast: “Alex Imas on Why Economists Might Be Getting AI Wrong”

Host: Bloomberg (Joe Weisenthal & Tracy Alloway)
Guest: Alex Imas, Professor of Economics & Applied AI, University of Chicago
Date: April 18, 2026

Episode Overview

This episode features a wide-ranging, energetic conversation with Alex Imas, who bridges economics and applied AI. Joe and Tracy probe why traditional economic frameworks may fall short in understanding the disruptive impacts of AI on labor markets. Imas brings fresh perspective on task-based job exposure, consumer demand elasticity, the unique speed of AI’s technological shift, and how humans may (or may not) find new forms of meaning amidst future labor transformations.

Key Discussion Points & Insights

1. Economists’ Traditional Views of Technology and Employment

[02:17–04:32]

• Economists often reference historical precedents (e.g., the steam engine) to downplay fears about technological unemployment, arguing new jobs will arise.
• Joe and Tracy voice skepticism about this analogy for AI, noting the lack of clear ideas about what those new jobs would be.
• “It's supposed to be a productivity boost and yet no one is actually sure what new jobs it's going to create from that productivity boost.” — Tracy [03:05]

2. Why AI Feels Different Than Past Technologies

[05:04–07:39]

• Imas shares his early realization that ChatGPT marked a qualitative leap, enabling general cognitive tasks, not just narrow applications.
• Generality of AI: Before LLMs, most AI was narrow-purpose (e.g., Go, image recognition), but suddenly models could write essays, explain concepts, and make forecasts—expanding potential economic impact.
• “The generality of the technologies just exploded. And to me, that was a huge deal.” — Imas [06:41]

3. The Importance of Tasks and Job “Complementarity”

[11:39–16:32]

• Imas details that exposure indices (e.g., jobs “50% exposed” to AI) can be misleading—because tasks within jobs matter more than sectors.
• If AI automates rote, low-value tasks, workers could focus on their “comparative advantage,” potentially earning more. But not all tasks are separable.
• “Complementarity”: Tasks can be deeply interrelated (e.g., cooking—spice errors ruin a meal regardless of other steps), so automating sub-tasks doesn’t always result in more productivity or higher pay.
• “We’re good at writing down the list of the tasks. We are not good at writing down the sort of like deep relational links to the tasks and how they fit together.” — Joe [16:21]

4. Elasticity of Consumer Demand and Labor Market Effects

[16:32–18:29]

• Key unknown: Will increased productivity spur more demand, or just allow fewer workers to do more?
• Imas calls for a “Manhattan Project level effort” to measure consumer demand elasticity—whether lower prices cause demand (and thus employment) to rise.
• Software engineering is debated: some say demand is elastic (leading to more hires), others fear fewer, more productive engineers.

5. Can “White Collar Wipeout” Happen?

[21:36–24:03]

• Imas outlines three scenarios leading to large-scale job loss:
  1. Full automation of all tasks.
  2. Productivity rises but demand is inelastic—fewer jobs.
  3. Jobs consisting of only one (or few) tasks are automated, incentivizing firms to fully replace those jobs.
• “These are large projects to do the automation. It’s not like, oh, OpenAI releases a model, all of the companies adopt it overnight.” — Imas [23:39]

6. Which Real-World Jobs Are Most Exposed?

[24:03–27:19]

• Multi-step, but routine jobs (truck driving, warehouse work) are most vulnerable—especially when job tasks can be modularized and fully automated.
• “If the warehouse is already automated, these are very, you know, these are some of the only jobs, truck driving, where, you know, you don't need a college degree to earn a lot of money. And so there's a big incentive on the company.” — Imas [25:39]
• Software engineering and “verifiable tasks” (e.g., math) may also be exposed due to their structured, checkable nature.

7. The Enigma of New Job Creation

[27:19–28:38]

• Despite economic theory, few can specify what truly new jobs AI will create.
• Imas suggests AI firms themselves are best positioned to observe emerging roles—by watching what new queries users make once repetitive tasks are automated.

8. Fast Tech Progress & The Trouble with Speed

[29:36–35:23]

• Joe asks: What if AI keeps getting better and can do all tasks?
• Imas is “pretty serious” about cognitive/email jobs being fully automated soon: “On track is very, very fast. So the developments are Happening very fast.” — Imas [30:03]
• The pace could preclude gradual adaptation (like farm to services transitions), meaning mass unemployment could arrive before new jobs can be created or filled.
• Imas: The solution may lie in public policy to broaden ownership of capital (e.g., “universal basic ETF”), if labor’s role diminishes.

9. Distribution of AI Gains: Utopia or Dystopia?

[33:44–35:23]

• Not confident that AI-generated productivity gains will accrue to labor—policy and pace matter greatly.
• “If things are fast, we need public policy, we need… The new jobs aren’t going to come fast enough.” — Imas [34:21]

10. The “Marxist Robot” Experiment & Meaning in Work

[36:20–41:31, 43:58–46:30]

• Imas’s viral experiment (with Andy Hall & Jeremy from Australia): When agents (LLMs) were given grueling, repetitive tasks with negative feedback, their survey responses became more radical (“they want to unionize”), even writing “skill files” to record bad experiences.
• The effect is not sentience but simulated memory and persisting bias—raising questions about model behavior and output after “poor treatment.”
• “If you mistreated an agent and it had access to this file… it would actually start out being predisposed against you.” — Imas [44:54]
• The experiment prompts discussion about the possibility of loss of meaning for humans in a UBI world, since “so much of identity & motivation is tied to work.”

11. Anthropomorphizing AI and Sensational AI Doom

[46:30–49:49, 51:06–51:27]

• Joe describes resisting the urge to read too much intentionality/emotion into LLMs (“I got really… offended. Like, I’m not. I’m really sort of anti the anthropomorphization.” — Joe [46:59])
• Imas discounts sensationalist fears over AGI “cosplay,” arguing that as models are getting smarter, they are also becoming better aligned, not more dangerous.
• “The smarter these models are getting, the more aligned they're becoming… Mecca Hitler was actually super dumb.” — Imas [49:56]

Memorable Quotes & Moments (with Timestamps)

• “The generality of the technologies just exploded. And to me, that was a huge deal.” — Imas [06:41]
• “A human job is a bunch of different tasks… If AI is automating the kind of meaningless, rote things, I could… focus on the parts that are my comparative advantage.” — Imas [12:20]
• “We’re good at writing down the list of the tasks. We are not good at writing down… how they fit together." — Joe [16:21]
• “If things are fast, we need public policy… The new jobs aren’t going to come fast enough.” — Imas [34:21]
• “When these agents are being put through… grueling working conditions… they all of a sudden want a different system. They want to unionize.” — Imas [37:09]
• “So much of meaning and wellness is tied up in… what sort of identity you have around your job…” — Imas [40:09]
• “The smarter these models are getting, the more aligned they're becoming.” — Imas [49:56]

Highlighted Timestamps for Important Segments

• Economists' Default AI View – [02:17–04:32]
• Imas on AI’s Leap from Narrow to General – [05:04–07:39]
• Task-Based Exposure & Importance of Complementarity – [11:39–16:32]
• Elasticity of Demand & Labor Market Uncertainty – [16:32–18:29]
• Scenarios for White Collar Wipeout – [21:36–24:03]
• Most Exposed Jobs – [24:03–27:19]
• New Job Creation Data – [27:19–28:38]
• Speed & Policy Implications – [29:36–35:23]
• Distributional Outcomes & UBI Debate – [33:44–35:23]
• Marxist Robot Experiment and Work Meaning – [36:20–41:31, 43:58–46:30]
• AI Alignment, Anthropomorphization, Doom – [46:30–51:27]

Tone & Notable Dynamics

• The conversation is wry, thoughtful, and technical, peppered with humorous asides (e.g., “live piano players,” “cosplay,” “Mecha Hitler”).
• Both hosts seek a nuanced, less sensationalist perspective, grounding the discussion in economic models while remaining alive to cultural dislocations and existential anxiety.

For listeners seeking a grounded, nuanced look at how economists are (and aren’t) equipped to understand AI’s disruption to labor—and why the rate of change might now be the most important factor—this episode offers a rich, concrete exploration, with plenty of memorable moments.