Productizing Work
The dominant corporate narrative is that AI will seamlessly automate our drudgery and free us up for "high-level" thinking. The reality on the ground is far messier. These papers show that dropping AI into workflows doesn't magically reduce work—it often intensifies it and degrades learning. True productivity requires intentionally designing "Hybrid Intelligence" systems that elevate human capability rather than replacing it.
Research Roundup
Cyborg Superman to the Rescue
Imagine you could hear every cry for help on the planet and you had the power to help. This Superman story [1] revealed the price: every moment you took for yourself—every sip of coffee or casual conversation, every night's sleep, every private thought—was a lost life you could have saved.
An eight-month ethnographic study of a tech company suggests AI-users might be suffering a similar, if less noble, mania: workload creep.
Because AI removes the friction of starting tasks, employees didn't work less; they expanded their scope, multitasked more, and bled work into their breaks. I’ll admit to experiencing this myself; my task list is now filled with whole papers, experiments, and coding projects I now expect to finish by week’s or even day’s end.
The illusion of a "frictionless" AI partner creates a self-reinforcing cycle of acceleration, which might easily lead to cognitive fatigue and burnout. When you treat AI as an autonomous minion rather than a cognitive tool, you don't get more free time—you just get an unsustainable treadmill.
[1] Told in the amazing comic series Astro City via the homage character, The Samaritan.
It’s In The Way That You Use It [1]
Does leaning on AI degrade human reasoning? A randomized trial of law students produced an appropriately messy result.
2nd and 3rd year law school students who used AI to synthesize legal materials produced better work faster, and it didn't hurt their comprehension on subsequent unassisted tasks. However, when students used AI to revise their own reasoning memos, the already-strong writers actually regressed.
What is missing in the reported results is the usage heterogeneity I have seen in my experiments and Anthropic has identified in their studies. 100 elite law school students are likely falling much more in the “cyborg” side of the distribution of users, in which case I’m not surprised at the result.
But I wouldn’t assume that this pattern generalized to most students for the simple reason that it hasn’t in dozens of other studies on other populations.
That final finding—already-strong writers actually regressed during AI-assisted memo revisions—is the more curious one. Were they so close to the finish line that even these elite students allowed AIs to write for them? Context rules!
[1] Listen to mister Clapton speak the truth.
…or, Building Pro-Human Hybrid Intelligence
Famed economists David Autor and Daron Acemoglu argue that we are suffering from a pervasive "pro-automation ideology" and they propose a framework separating technologies that merely replace human labor from those that create new tasks and expand human expertise.
They describe “5 categories of technological change”:
- labor-augmenting,
- capital-augmenting,
- automating,
- expertise-leveling, and
- new task-creating.
And claim, “Only the last category is unambiguously pro-worker, generating demand for novel human expertise rather than commodifying it.”
I love this stuff and I’m enjoying the authors’ Brookings interview, but I still think their “everything is a skill” model of labor is limiting. I’m worried about “pro-worker” being about enabling traditionally-lower-skilled labor to do higher-skilled tasks rather than redesigning agentic AI toward hybrid intelligence. New tasks are only pro-worker if a meaningful number of workers are qualified to do them.
Media Mentions
As promised last week, my new FT op-ed came out. If you didn't see it, here's a summary and a link:
Two students, same assignment, same AI, same glowing screen. From the front of the room they looked identical. But the EEG headsets told a different story.
In most students, the brainwaves marking cognitive effort collapsed within minutes of using the AI, their neural state drifting closer to watching TV than solving problems. In a small few, those same signals lit up. The difference? This group argued with the machine instead of accepting its answers.
My new op-ed for the FT is about that divergence and why it matters far more than the benchmark scores AI labs love to announce. Those exams test the model alone, but that's never how AI is actually used. A radiologist weighs its read, a lawyer checks its brief, an engineer reviews its code. A more capable AI can produce a less capable human-AI team.
The fix isn't more "human oversight"—a bored clerk rubber-stamping fluent hallucinations is not oversight. The fix is designing for friction. When I had the AI respond with questions instead of answers, the share of actively engaged students more than doubled.
The economies that optimize for fluency risk automating away the very minds they need.
Read the full piece here: https://www.ft.com/content/d2d8f531-2833-4edc-9107-7bb73d9f0c4b
SciFi, Fantasy, & Me
𝗦𝗰𝗶𝗙𝗿𝗶𝗱𝗮𝘆: I’ve worked my way through a third of my read/watch list from last week.
- It’s an unneeded recommendation but narky, neurotic Murderbot and his favorite soap operas continue to make me happy.
- Widow’s Bay is fun and creepy enough to satisfy. Matthew Rhys and Stephen Root everything better.
- I’m past tired of characters behaving like a reader stand-in rather than as rational people embodied in a world they’ve spent their whole lives.
Stage & Screen
- June 22-30, Online: Six separate talks for Pride, because "The Tax On Being Different" can't be wished away. It's wonderful that so many companies are choosing celebration over fear.
- July 7, MIT: I'm giving the keynote for the MIT App Inventor Global Education Summit taking place this year at MIT CSAIL.
- July 8, NYC: It a book talk for Robot-Proof at the Harvard Club...how swanky!
- Maybe July 24...Maybe San Diego: Maybe....
- September 15, SF: Innovation Day with INSEAD!
- September 16, DC: AI and education–beyond dreams and dread.
- September 19, Phoenix: I'm giving the keynote for the Association of Science & Technology Centers annual conference.
- September 21, Stanford: We're still working on the details, but hopefully I'll be talking about my research on machine learning and neurodiversity for Stanford's Neurodiversity Project.
- September 24, NYC: Culture Shifting Deal Making Summit
- September 29, Cincinnati: Still baking...
- September 30, Irvine: Hybrid Intelligence for innovation!
- October 6, SF: UCSD Alumni Association
- October 6, SF: Giving a talk at the Draper Richards Kaplan Foundation
- October 21-23, Warsaw: So much good stuff is in the works for my first visit to Poland (and maybe time in Germany as well!)
- October, Toronto: The Future of Work...in the Future
- November 19, NYC: Secrets in the dark!