When the tech industry talks about the future of work, the narrative is aggressively focused on automation, efficiency, and replacing human effort, creativity, and ingenuity with software. We are told that artificial intelligence will automate most of the work of humanity, and that will usher in either a utopia or dystopia.
We are told that algorithms “learn,” software “reads,” and machines “work.” Newer campaigns go further, claiming software can “create,” “explore,” “reason,” and act as a “creative partner.” There is an excitement about collapsing the boundary between human and machine when there is no need, but there is a strong marketing drive to make the two indistinguishable.
Whether or not it is intended to be, this language is deceptive. Machines can complete an expanding number of tasks, but the process is not the same. When we borrow human terminology to describe the processes of software, we risk losing a clear definition of what meaningful human work is.
It helps to define the work of humans and computers in order to compare them. What follows is not a definitive computer-science account of computation, but a simplified framing that highlights the core inputs and constraints so the differences with human work become clearer.
Computer work
Computer hardware and software can complete all sorts of tasks, many of which humans do regularly, and many more that are beyond even the most exceptional person. Computers can process vast datasets in seconds, generate reports from complex models, or simulate climate systems no individual could calculate by hand.
Computer work rests on three essential elements: energy, compute, and cryptography. Compute is the raw mechanical processing power. When you ask an AI to generate a report or run a complex spreadsheet in the cloud, you are using compute. It requires hardware and electricity to perform billions of mathematical operations.
To make that work function at scale across networks, the data must be protected. Cryptography provides the mathematical safeguards that keep information private and ensure it reaches only the intended people. Without it, cloud computing and AI would still be technically possible, but they could not become trusted, high-value work in a connected economy.
Anything that can be described mathematically can be automated using energy, compute, and cryptography. Much of what constitutes human work cannot.
Human work that people don’t enjoy doing is often described in mechanical terms. The work is described as robotic or repetitive. People start to feel like automatons. But human work requires different inputs than computers. Human work is not just the output of a task, it is defined by what goes into it.
Human work is part of a complex, embodied experience.
Physical work. The physical component of work is our embodied reality. It can be the fatigue at the end of a long shift, the stress held in our shoulders, or the physical presence required to show up to a workplace. Every part of that physicality has a direct feedback loop with our emotional and cognitive processes.
Even in knowledge work that is completed remotely, our physical state dictates our capacity. Mental exhaustion comes from physical and biochemical limitations on the brain and body. We are not servers that can be rapidly scaled up with more electricity and new hardware. We have to make difficult decisions limited by our time, attention, energy, and individual priorities.
Cognitive Work
The cognitive component involves judgement, pattern recognition shaped by lived experience, and the ability to hold competing possibilities in mind without collapsing them into a single mathematical optimum. It is slower and messier than pure computation, yet capable of leaps that current models simulate but cannot perfectly replicate or fully model.
Emotional Work
The emotional component is a unique part of human experience that cannot be replicated, partly because of human limitations. Human work requires emotional labour which involves the navigation of interpersonal dynamics and the empathy required to understand other people’s motivations and objectives.
We come to understanding based on complex networks of relationships and our perception of other people in the spaces they inhabit, with verbal and nonverbal communication. Humans have evolved over millennia to make rapid judgements based on complex social situations as well as threats and limitations that are far beyond our abilities to mathematically model, and hence to automate.
A computer has no skin in the game. It does not feel the emotional weight of a bad decision. When a language model hallucinates a dangerous piece of advice it is not risking physical safety and health in anywhere near the same ways as a human.
Capabilities and Constraints
The danger of sophisticated automation is that by praising the frictionless, tireless output of machines, we begin to view the human components of work as problems to be fixed or bugs to be patched. These components involve strengths and limitations that come from our physical limits and emotional complexities.
However, we can understand these limitations in terms of unique human characteristics that create sensitivity to time, interpersonal relationships, emotional experience, and unique ways of perceiving risk and reacting to threat.
If we too easily trust computers to do human work, and delegate our most intimate relationships, rewarding tasks, creative expression and ethical decision-making away, we risk losing the very real human strengths that are challenging and worthwhile to cultivate over time.
There is also a big risk that we try to combine the two. We expect humans to scale linearly with automation like a computer program. We expect people to churn out ever-greater amounts of work of the same level of quality without compromise.
There is only a certain amount of human work that can be scaled up by finding efficiencies; some definitely can, and it is worth finding those. If we treat humans like computers, demanding relentless output without physical or emotional resources that match the task, we face burnout, which is far more difficult to recover from than to prevent.
We face a profound choice. We are currently deciding which parts of our economy to automate, and which must remain the domain of human expertise and emotion. Automating work and understanding how to automate some tasks that humans do is a triumph of engineering.
The definition of work will change, but we have to understand work as part of the processes. Work is not just about outputs and what we get from it, but about what we become from it.