As AI systems grow more and more capable, a lurking question looms over us, as oppressive as a dark cloud creeping over Mordor: Will there still be work for humans in an AI-saturated world?
I understand the concern. Generative AI can write, design, analyze, and automate tasks that once required specialized expertise. But when we step back to look at the broader economic picture, once we take into account historical precedent, the outlook is far more nuanced and hopeful.
To understand why, we need to distinguish among three different types of work.
Three Kinds of Work
Human labor can be divided into three broad categories:
- “Do” work — Physical labor: digging, lifting, assembling.
- “How” work — Knowledge labor: calculations, analysis, problem-solving.
- “Why” work — Goal-setting labor: deciding what should be done and why.
For centuries, we’ve been automating away do work. The plow automated the digging of furrows; industrial machines now automate manufacturing. These tools displaced certain forms of manual labor, but they also dramatically increased productivity and created entirely new economic sectors.
We are now entering a phase where we will not only automate do work, but how work. Because generative AI systems can draft reports, generate code, compose images, and assist in research, a number of white-collar jobs will inevitably be lost.
Economic displacement is real and painful. But history suggests it is not the final chapter.
Lessons from History: Displacement and Creation
Consider the word computer.
Before “computer” was a machine it was a job title. Human computers, often women, performed mathematical calculations by hand for scientific organizations like NASA before the advent of electronic processors. The 2016 film Hidden Figures tells the story of a group of human computers facing job reassignment courtesy of their digital computer counterparts.
When electronic computers emerged, the job of “human computer” largely disappeared. Humans were displaced.
But what followed? Entirely new industries: software development, information technology, digital media, and the global tech economy.
The same pattern occurred with the printing press, another culturally disruptive technology. Before Johannes Gutenberg’s invention, scribes painstakingly copied manuscripts by hand. An individual book the size of the Bible cost the equivalent of $60,000 to produce, providing months of work for a copyist. Then the printing press automated the profession of “scribe” out of existence.
Yet printing unleashed an explosion of literacy, publishing, journalism, science, and mass media. It didn’t just create jobs, it created new industries.
Disruptive technologies follow this pattern. Automation eliminates specific roles while expanding the overall space of economic opportunity.
Generative AI will likely do the same.
The Combinatorial Explosion of Opportunity
Why does the space of opportunity grow?
Here’s the key insight: when you encapsulate capabilities as deployable components, the number of possible systems you can build from them grows exponentially.
Imagine we can package human-like abilities into “capability widgets.” For example, consider two widgets:
- A system that turns text into images.
- An AI agent that can autonomously perform tasks.
If you have just one widget, you can build one type of system (or no system). With two, you can combine them in four ways, either including or excluding each particular component. With three, the number of possible systems doubles again.
The pattern is exponential:
- 1 capability → 2 possibilities
- 2 capabilities → 4 possibilities
- 3 capabilities → 8 possibilities
- 4 capabilities → 16 possibilities
And that’s only counting whether components are included or excluded. If we take into account ordering among the components, as is the case for processing pipelines, we introduce even more permutations.
Therefore, the space of things we can build expands exponentially as the number of components we can build from increases.
How does this affect labor? Our ability to explore these possibilities only grows linearly.
Crucially, AI agents also face deployment constraints. They require ample compute, torrential energy, and massive storage. At best we can deploy them linearly.
Thus, we cannot investigate every possible combination of systems. Nor can our AI surrogates. There will always be more potential innovations than there are agents (human or artificial) to explore them.
Similarly, in science there will always be more interesting questions than there are researchers to investigate them. The frontier expands faster than we can scout it.
This creates a persistent gap between opportunity and exploitation.
What AI Lacks
Even in a future where AI systems equal or surpass humans on all knowledge tasks, there remains something uniquely human, beyond their cybernetic grasp: our ability to define meaningful goals.
AI systems lack desire. They do not wake up wanting to cure cancer, write a symphony, or explore Mars. They do not want anything. They respond to our prompts and optimize the objectives we define for them.
But they can still be useful. A large language model is arguably a lot like a lazy genius: a colleague who is extraordinarily capable but never initiates any work on his own. He can still contribute to a team and provide value, given the right leadership and direction.
A more charitable analogy is a brilliant undergraduate. As a professor, you may have exceptionally talented students. Talented students still require guidance. They need someone to frame the research question, define the problem, and determine what is worth investigating.
That direction — that why — is uniquely human.
We set the goals. We determine what questions matter and we decide which systems are ultimately worth building.
The Shift Toward “Why”
Historically, the economy has shifted from do work (manual labor) to how work (knowledge labor). Now we are seeing another shift: from how work to why work.
In the future, many of us will become project managers and directors, metaphorically and literally. Our value will lie increasingly in defining objectives, orchestrating AI tools, and deciding what should be pursued.
There will be disruption. In the short term, generative AI will almost certainly displace humans. But displacement is not replacement.
The exponential growth of opportunity ensures there will always be systems to design, questions to ask, and projects to lead. We will remain indispensable in defining direction.
The Long View
Economically, AI systems may change the nature of work, but they cannot fully eliminate the need for human participation even in the most dystopian scenarios. There are simply too many opportunities to investigate.
The frontier of possibility expands faster than any man or automaton can explore. And as long as we are the ones who ask why, setting the goals to determine the ends toward which technology is aimed, there will always be meaningful work.