Let’s look at how AI is slowly but fundamentally reshaping the landscape of professional work. Forget the simplistic debate over whether robots will destroy more jobs than they create and ask instead, “What will the new jobs look like and who will be qualified to fill them?” The reality emerging from cutting-edge research, including my own, is far more nuanced than automation or “co-pilots”. It offers a future where AI acts less as a replacement or advisor, and more as cybernetics, integrally augmenting human capacity…but only if we make that choice.

Studying fields as diverse as accounting and material science, emerging research documented significant productivity gains for professionals using AI. Unlike previous research, however, these studies show complementarity with experience and mastery. In other words, the best accountants showed the biggest boost. What is important is understanding the mechanism: meta-uncertainty. These elite knew when to rely on the AI and when to push back. AI substituted statistical learning, a core feature of human intelligence, but augmented other more valuable forms of human intelligence, such as meta-uncertainty, creative problem solving, and meta-learning.

Another powerful story of AI complementarity comes from medicine. The trend extends even into high-stakes domains like medicine. While advanced models are achieving state-of-the-art performance, often surpassing human experts across benchmarks, the biggest benefits are coming from the interplay of human and AI intelligence. Medical models, such as Med-Gemini, perform better when they ingest not just “facts” but also the messy and diverse observation of human doctors and other medical staff. And doctors improve their diagnostic and treatment planning performance when they subject their plans to AAI evaluation. (btw, evaluation is the dramatically underused superpower of AI-human collaboration.)

These studies collectively reveal a powerful shift: even as AI rapidly masters complex, high-value tasks previously exclusive to skilled professionals, the messy and ill-posed facets of this work remain human domains. The future of professional work isn't about humans competing against these machines, but learning how to blend our distinct forms of intelligence together. (And that is not, “AI does the work while I sip margaritas.”) The AI-augmented professional will be someone who understands how to selectively leverage AI's analytical power, interpret its outputs critically (especially when uncertainty is flagged), and focus unique human intelligence—contextual assessment, communication, perspective taking, complex problem-solving—on the ill-posed problems where human expertise remains not just indispensable but transformative. (If you’re a true fan of why-why-why, you find that in reality every problem is imposed). This future requires adaptation, meta-learning, and a willingness to become a cyborg.

In the spirit of being grounded, here are a few concrete lessons for working professionals:

Become an "AI Advisor": Your value is not in competing with AI answering well-posed questions. Instead, much like an advisor to a graduate student, focus on your ability to guide, interpret, and validate its outputs. Develop your meta-uncertainty to know when an AI-generated solution is sound and when it requires your unique human insight.

Collaborative Mindset Requires Trust: AI is a tool, but it’s also an economy in which collaboration (humana-human-AI) flourish. Learn its strengths and weaknesses, and proactively seek opportunities to integrate it into your team’s workflows to augment your own capabilities. Meta-uncertainty gives you trust in AI when it is strong, but its also a spidy sense for when it will break.

Invest in Hybrid Systems, Not Just Technology: The goal shouldn't be to simply implement the latest AI, but to design intelligent systems that seamlessly integrate human guidance and creativity. Do not obsess over performance and efficiency, but instead leverage the gains for increased exploration and discovery.

Redefine Roles and Training: Think through every job in your organization: if the routine (well-posed) parts of the job are consumed by AI, what should that new job look like? How can each role be transformed to leverage uniquely human capabilities to explore the unknown. Rebuild each role for depth and creativity. Retrain each employee for meta-learning. Focus on AI to augment those capabilities, not just for automation.

Foster a Culture of Critical Evaluation: Encourage a healthy skepticism of AI-generated outputs. Famously cantankerous leaders tell their employees, “This isn’t good enough.” Now you need every employee to be ready to do the same with their AI tools. Create an environment where employees feel empowered to question, challenge, and ultimately improve upon the recommendations of any automated system.

Follow me on LinkedIn or join my growing Bluesky!

Research Roundup

AI Mastery=Facts; Human Mastery=Uncertainty

Accounting rarely makes the headlines*, but accounting plus AI reveals the fundamental rule for human+AI collaboration.

*(Unless you're Enron…or the inevitable “shocking” reveal of dodgy crypo accounting every 6 months.)

When an LLM was used in a major accounting firm, it produced a whopping “55% increase in weekly client support” and reallocated “8.5% of accountant time from routine data entry toward high-value tasks”. [yawn] It also improved financial reporting quality and reduced monthly close times. [double yawn]

So boring accounting+AI = exactly the productivity boost you'd predict…why is this with sharing? 

Because unlike other papers it was the elite accountants that showed the biggest effects. Crucially, they didn't just blindly accept AI suggestions; they selectively leveraged automation and increased their intervention when AI signaled uncertainty.

Foundation models have mastered facts; we humans have mastered uncertainty…poorly, but still better than anyone else.

Uncertainty Guidance

Despite what some have claimed, LLMs are intelligent. Importantly, though, humans and LLM intelligence overlap but are not the same. New research reveals that these differences lead to different kinds of mistakes…but how collaboration can leverage the best of both.

The new study identified "complementary error patterns" when diagnoses from over 40,000 physicians were combined with those from five leading LLMs. “Hybrid collectives of physicians and LLMs outperform both single physicians and physician collectives, as well as single LLMs and LLM ensembles”.

The gain from hybrid systems “holds across a range of medical specialties and professional experience”. It mirrors earlier findings hybrid decision-making in medicine, product development, and creative problem solving.

For too long, the AI debate has been between a zero-sum game of human versus machine or an effort-free ascendance of superintelligence. Instead the most effective outcomes emerge from an effort-fueled engine built from the unique strengths of both.

I Speak Eloi

Online communication is getting simpler…but not in a good way. A huge study of 300 million comments across 8 platforms over 34 years found a “nearly universal reduction in text length, diminished lexical richness, and decreased repetitiveness”. While people still introduce new words at the same historical rate, the structure and density of online language are consistently shrinking. Online language is becoming cognitively shallow…and our minds will follow. Our digital ecosystem, now aided by an increasing number of “efficicency”-focused AI tools, are optimized for ease and speed. This erosion of cognitive complexity and productive friction in public discourse will have consequences for generations to come. It reminds me of the research showing that people who grow up in spatially complex cities (Prague) have measurable cognitive advantages over those whose brains develop in cities with simple layouts (Chicago). When technology lessens our cognitive load it can also reduce our cognitive capacity.

<<Support my work: book a keynote or briefing!>>

Want to support my work but don't need a keynote from a mad scientist? Become a paid subscriber to this newsletter and recommend to friends!

SciFi, Fantasy, & Me

The shambling endstate of all children’s fantasy is Gombles the Clown.

Hurray for anti-prophesy spider people!

Also another name for biomedical AI could be “Poop & Boners: A Statistical Approach”

Stage & Screen

• September 18, Oakland: Reactive Conference
• Sep 28-Oct 3, South Africa: Finally I can return. Are you in SA? Book me!
• October 6, UK: More med school education
• October 29, Baltimore: The amazing future of libraries!
• November 4, Warsaw: More fun with Singularity University
• November 21, Warsaw: The Human Tech Summit
• December 8, San Francisco: Fortune Brainstorm AI SF talking about build a foundation model for human development

If your company, university, or conference just happen to be in one of the above locations and want the "best keynote I've ever heard" (shockingly spoken by multiple audiences last year)?

Vivienne L'Ecuyer Ming