Key Moments
What If AI Makes Us Less Productive? (The Study Sam Altman Doesn’t Want You to See)
Deep Questions with Cal NewportKey Moments
AI may not make knowledge workers more productive, especially in deep work, according to a recent study.
Key Insights
A study found experienced developers using AI were 20% slower at deep work tasks than those without AI.
AI's benefit in programming might be limited to shallow tasks, not cognitively demanding deep work.
Cybernetic collaboration with AI can reduce focus intensity and duration, hindering productivity.
True collaborative deep work enhances focus, unlike AI-assisted 'cybernetic collaboration' which offers breaks.
The environmental impact of large AI models is a concern, but economic factors may limit their widespread use for all tasks.
Effective deep work requires cultivating intense focus, which AI-assisted interactive loops can disrupt.
THE SURPRISING STUDY ON AI AND PRODUCTIVITY
A recent study by MER (pronounced 'meter') challenged the common assumption that AI tools inherently increase knowledge worker productivity. The research focused on experienced open-source developers and measured their productivity on real-world tasks, both with and without AI assistance, primarily using advanced models like Cursor Pro. The findings revealed an unexpected outcome: developers using AI were, on average, about 20% slower in completing tasks compared to those who did not use AI.
DEEP WORK VERSUS SHALLOW TASKS
The study's results are particularly significant when considering the nature of the work performed. The tasks involved bug fixes, feature implementations, and refactoring, which require 'deep work' – cognitively demanding activities necessitating sustained, distraction-free focus. Cal Newport emphasizes that deep work is the engine of valuable output in the modern economy, distinct from supporting activities like email and meetings. AI's current limitations in enhancing this type of focused, original cognitive effort appear to be a key factor in the productivity paradox observed.
CYBERNETIC COLLABORATION AND THE FOCUS DRAIN
The study detailed how developers integrated AI: through an interactive, back-and-forth 'cybernetic collaboration.' This involved prompting the AI, reviewing its output, and debugging its generated code. While intuitively this collaborative approach with a machine should boost efficiency, it instead led to reduced programmer productivity. Developers spent more time reviewing outputs, prompting, and waiting for AI generations, and also experienced more idle time, indicating a shift away from active coding and intense focus.
THE WHITEBOARD EFFECT VERSUS AI INTERACTION
Newport contrasts this cybernetic collaboration with successful human collaboration in deep work, termed the 'whiteboard effect.' True collaboration, as seen in academic or scientific settings, uses the presence of others to increase focus intensity and duration. The social pressure and shared cognitive load drive deeper concentration. In contrast, AI-assisted cybernetic collaboration seems to reduce this intensity by providing breaks and offloading cognitive effort, making the process more pleasant but ultimately less productive, as it decreases the brain's operating gear.
THE ENVIRONMENTAL CONCERNS AND ECONOMIC REALITY OF AI
The discussion touches on AI's environmental impact, noting that large, frontier models consume significant energy. However, Newport posits that the high computational cost makes extensive querying of these massive models economically unsustainable for most everyday tasks. He suggests the future of AI likely lies in smaller, specialized models and hybrid systems running on existing hardware, rather than constant reliance on cloud-based, resource-intensive frontier models. This economic reality, he argues, will naturally limit the environmental footprint of AI in the long run.
NAVIGATING TECHNOLOGY IN EDUCATION AND WORK
Examining a case study of a West Virginia school without Wi-Fi, Newport illustrates the danger of drawing premature conclusions about technology's impact. While the op-ed suggested the lack of Wi-Fi harmed student performance, data analysis revealed a more complex picture. The school's struggles were not definitively linked to Wi-Fi limitations, highlighting the need for rigorous data analysis before accepting intuitive claims about technology's benefits or harms in education and professional life.
THE ENDURING VALUE OF INTENSE FOCUS
Ultimately, the core takeaway is that deep work thrives on the intensity and duration of focus. While AI may automate shallow tasks or assist in non-cognitively demanding areas, it can actively hinder productivity in deep work scenarios by disrupting focused effort. The allure of 'cybernetic collaboration' with AI, offering a more pleasant and less demanding experience, can be a trap that leads to reduced output quality and efficiency. The ability to sustain intense focus remains paramount for creating significant value.
Mentioned in This Episode
●Products
●Software & Apps
●Companies
●Organizations
●Books
●Studies Cited
●Concepts
●People Referenced
AI Productivity Predictions vs. Observed Results
Data extracted from this episode
| Predictor Group | Predicted Productivity Change (%) | Observed Productivity Change (%) |
|---|---|---|
| Economic Experts | +40% | N/A |
| Machine Learning Experts | +40% | N/A |
| Developers (During Study) | +20-30% | N/A |
| Developers (After Study) | +20% | N/A |
| Actual Measurement | N/A | -20% (Slower) |
Common Questions
A 2025 MER study found that experienced open-source developers were approximately 20% slower when using AI tools for coding tasks. While AI might help with shallow tasks, it can decrease productivity in deep work if it reduces focus intensity and duration.
Topics
Mentioned in this video
A pioneering academic program at Georgetown University that integrates computer science and ethics.
A field of AI development focused on ensuring AI systems operate safely and align with human values, mentioned as a less secure area of critique for some.
The game in which the AI system Plurobis achieved professional-level performance, demonstrating the power of specialized AI models.
An interdisciplinary field focused on the ethical implications of technology, a focus area for Georgetown University and Cal Newport's work.
Casual pants made from soft bamboo fabric by Cozy Earth, highlighted for their comfort and versatility.
Used as an analogy for frontier AI models, suggesting they are high-performance demonstration tools rather than practical, everyday applications.
Devices that teachers in Green Bank, WV, noted would enable individualized learning but were impractical without Wi-Fi.
Devices that students in Green Bank, WV, could not fully utilize for online curriculum and standardized testing due to the lack of Wi-Fi.
A pant by Cozy Earth described as both stylish and comfortable, made from their signature fabric.
Professor at MIT whose group Cal Newport worked with on the theory of distributed systems.
A teacher quoted in the Washington Post op-ed about Green Bank, highlighting the detriment to instruction caused by the lack of online curriculum components.
A teacher quoted in the Washington Post op-ed about Green Bank, expressing concerns about the impact of lacking Wi-Fi on individualized learning.
Professor at MIT whose network and mobile systems group Cal Newport worked with during his post-doc.
A governmental body that Cal Newport's writing for The New Yorker aims to reach with policy-relevant insights.
A nonprofit that conducts evaluations of AI capabilities, producing high-quality, unbiased studies often cited by AI companies.
National Public Radio, a platform Cal Newport uses to reach policymakers and a broader audience.
A historical center at Georgetown University that was influential in bioethics, mentioned as a precursor to tech ethics initiatives.
A town in West Virginia near a radio telescope that restricts cell phone and Wi-Fi use, studied for its impact on student test scores.
A county in West Virginia that includes Greenbank, analyzed for student test performance trends in relation to Wi-Fi availability.
More from Cal Newport
View all 286 summaries
31 minAre LLMs a Dead-End? (Investors Just Bet $1 Billion on “Yes”) | AI Reality Check | Cal Newport
68 minWhy Is AI Making My Job *Worse*? | Cal Newport
24 minDid AI Just Become Sentient? (Not Quite...) | AI Reality Check
56 minReduce Your Screentime (5 Simple Steps) | Cal Newport
Found this useful? Build your knowledge library
Get AI-powered summaries of any YouTube video, podcast, or article in seconds. Save them to your personal pods and access them anytime.
Get Started Free