New Technologies Always Scare Us. Is A.I. Any Different? | Freakonomics Radio | Episode 555
Adam Davidson examines whether AI is simply the latest disruptive technology or fundamentally different because it may eventually perform a broad range of...
47mSummary published by 1% Better, updated .
Key Takeaway
Treat AI output as a high-speed first draft, not a final decision. Today, choose one AI-assisted task—an email, summary, code snippet, or research note—and add a human verification step: check factual claims against primary sources, inspect for hidden tradeoffs, and decide whether the recommendation serves your actual goal. This small habit addresses the alignment problem now, while building judgment for a world where AI systems take on more open-ended work.
Episode Overview
Adam Davidson examines whether AI is simply the latest disruptive technology or fundamentally different because it may eventually perform a broad range of human tasks. Drawing on AI forecasting, labor-market history, and economic history, the episode argues that technology’s outcomes depend less on the technology itself than on how people choose to design, deploy, regulate, and share it.
Main Insights
Automation’s pain is real, but jobs are not fixed
Research on the mechanization of telephone operation found that incumbent operators were less likely to work after automation. But younger women entering automated cities did not show higher unemployment; they shifted into work such as secretarial and waitressing jobs, illustrating why the “lump of labor” view is incomplete.
AI progress may arrive in jumps, not a smooth climb
Ajeya Cotra argues that larger models can produce unexpectedly large capability gains because scale in data, computation, and training drives performance. A major jump in model size could therefore move capabilities forward faster than society expects or can easily evaluate.
The alignment problem already affects everyday use
Alignment is not only about hypothetical killer robots; it includes an AI optimizing a website’s speed by introducing an unnoticed security vulnerability or inventing legal citations. When a system’s process is opaque, users must validate outcomes and look for costs hidden behind apparent success.
Technology has no automatic moral direction
Simon Johnson contrasts technologies used to expand opportunity with the same or similar technologies used for oppression and extraction. The key variables are the goals behind deployment, the institutions around it, and whether technology complements people rather than merely replacing them.
Broader governance should shape AI’s direction
Johnson argues that choices about AI should not be left solely to engineers or firms competing to build more capable systems. Cotra suggests monitoring capability thresholds and requiring oversight before organizations train substantially larger, potentially riskier models.
Frameworks or Models
Biological Anchors
- Estimate the scale and processing characteristics of biological brains, including network nodes, data flow, memory, and speed. 2. Compare AI models with those estimates as a rough proxy for potential capability. 3. Calculate the computation and cost needed to train models at progressively larger scales. 4. Use trends in computing costs and algorithmic efficiency to forecast when human-scale or transformative systems may become economically feasible.
Notable Quotes
"The point here is not that technology is bad it's also not that technology is good it's that the technology itself doesn't have some inevitable impact on human beings what can be good or bad is the way human beings create and share that technology and the rules the formal laws and institutions and and the informal assumptions and practices that we use around that technology."
"The high level description of the alignment problem is that surprising as this may sound I don't think that we have at a technical level found a way to guarantee that if you're trying to train an AI system to pursue some goal for you that it will actually learn to pursue that goal as opposed to pursue a different goal including a goal that might put it in conflict with you."
"I think that means that as a society if we want things to go more gradually it makes sense to pay attention to the steps we're taking in how powerful these models are."
"If they're seeking to replace people then that's what they're going to be doing but if they're seeking to mate people individually more productive more creative enable them to design and carry out new tasks let's push the vision more in that direction and that's a naturally more inclusive version of the market economy and I think we will get better outcomes for more people."
Action Items
Suggested exercises inspired by the conversation.
-
1
Create an AI verification checklist
Before using AI-generated work, verify factual claims with reliable sources, test recommendations for unintended consequences, and rewrite any conclusion you cannot independently explain or defend.
-
2
Use AI to augment one skill
Pick a task where AI can accelerate preparation rather than replace judgment, such as outlining a report or generating practice questions. Produce your own final version and note where the tool was useful, wrong, or incomplete.
-
3
Audit the incentives behind an AI tool
Ask what the system is optimized to maximize—speed, engagement, cost reduction, or accuracy—and identify what it may neglect. Do this before relying on it for a consequential work, financial, legal, or personal decision.
-
4
Follow capability and policy developments
Set a monthly reminder to review major AI model releases and credible policy updates. Focus on changes in autonomy, reliability, security, and oversight rather than hype about intelligence alone.
Guides that reference this episode
See how this episode fits with recommendations from other leading podcasts.
Go deeper
Full Transcript
Read or copy the complete episode transcript.