The conversation about AI and the future of work tends to run at two extremes.
On
the one hand, you have utopian visions of liberated humans in post-scarcity
abundance. On the other, dystopian forecasts of mass unemployment and even outright
doom.
This polarised debate has regrettably become typical of our public discourse today.
Which means we’re neglecting the most interesting insights lurking in the centre
ground.
As AI progress continues, a dramatic shift in the nature of human work is
inevitable regardless of p(doom). This shift has already begun and will speed up
dramatically in the years ahead.
With AI moving up the stack of work, what remains for humans isn't nothing – it's
everything that matters most.
This is the first instalment of a five-part blog series where I’ll explore
humanity's evolving role in an era of increasingly capable AI. Below, I'll discuss
why human work will shift more toward goal-setting than goal execution.
How should we understand intelligence?
In ‘Human Compatible’ Stuart Russell defines intelligence as "the ability to
achieve goals in a wide range of environments." This emphasises that intelligence
isn't just about having knowledge or smarts. It’s about using what you know in the
right way to accomplish your objectives.
Another influential definition comes from psychologists and AI researchers such as
Robert Sternberg. He describes intelligence as "mental activity directed toward
purposive adaptation to, selection of, and shaping of real-world environments
relevant to one's life."
Both definitions share a common perspective. Intelligence is about the successful
pursuit of goals.
AI doesn’t decide what you want to do
What's often lost in discussions about AI capabilities is a fundamental truth:
these systems are “just” pursuing goals that people have set.
When you type a prompt like "Write me a business plan for a sustainable coffee
shop" into ChatGPT, you're setting the goal. The AI doesn't have its own agenda or
desires – it's trained through millions of iterations to predict what text should
follow your prompt.
Behind the scenes, the developers have shaped the system with a process called
gradient descent. This is essentially a mathematical method that gradually adjusts
the system's parameters to better achieve its assigned objective of producing
helpful, accurate text. That is, to train in its intelligence.
This pattern holds across all AI systems today. They don't decide what to do;
people do. They don't choose what goals to pursue; people do. Assuming we can solve
alignment (discussion of which I’ll reserve for a future blog), this should remain
true.
As AI becomes more capable of performing human labour – including not just routine
tasks but increasingly creative and analytical work – this distinction becomes
crucial.
AI systems need people to specify the goal they should pursue in training and at
inference time.
Philosopher kings
This isn't as abstract as it might initially sound. In fact, these dynamics are
already visible in today's society.
Consider some of the roles that command some of the highest status and compensation
in our society: CEOs, investors, policymakers, and other leaders. What is their
primary function? It isn't to personally execute tasks – it's to determine which
tasks are worth doing at all. They form judgments about objectives, trade-offs, and
values. They decide what problems their organisations should solve and what goals
they should pursue.
Consider great musicians, artists, and creators. They imagine new things they could
bring into the world and decide whether it will resonate with others. That
resonance is in some sense a proxy for what will be considered valuable.
These are fundamentally philosophical questions: What is valuable? What is the good
we should aim for? How should we balance competing values? What kind of world do we
want to create? What goals should we set? Which should we pursue at the expense of
others?
In the end, what is left for us, is to determine who wants what, who gets what, and
why. We’re all philosophers now.
The new dividing line
This suggests a profound change in how we should think about the future of work:
not as a competition between humans and AI, but as a division of labour between
goal-setting (people) and goal pursuit (AI).
This partnership will be central to how resources are allocated and consumed, as
well as the value – both real and perceived – that results. As this division of
responsibility takes hold, our economic systems will necessarily transform to
reflect this new reality.
In the past, technological revolutions created dividing lines within human labour
markets. The industrial revolution drove a distinction between manual and cognitive
work. The information revolution accelerated this shift, privileging knowledge
workers over physical labourers.
The AI revolution is different. Rather than creating new divisions between types of
human work, it fundamentally redefines the boundary between human and machine
contribution.
As AI capabilities expand, the domain of uniquely human work contracts. But it
contracts toward something essential: the determination of which goals are worth
pursuing in the first place.
This has profound implications for how we prepare people for the future, how we
structure our organisations, and how we think about the distribution of power in
society.
If only a small elite is empowered to set objectives for increasingly powerful AI
systems, we risk amplifying existing inequalities and surrendering and centralising
control. The democratisation of goal-setting power becomes not just a question of
fairness, but of ensuring that the objectives pursued by increasingly powerful
systems reflect our collective values and aspirations.
Conclusion: The path forward
The future of work in an AI-enabled world isn't about humans competing with
machines at tasks machines are increasingly better suited to perform. It's about
humans focusing on what remains essentially human: determining which goals are
worth pursuing and why.
This transition won't be easy. Our economic systems, educational institutions, and
social structures aren't currently designed to distribute goal-setting power widely
or to develop this capacity at scale. And this transition will be disruptive. Jobs
will be lost. Industries will be transformed. Skills will become obsolete.
But rather than simply reacting to these changes, we have an opportunity to
proactively shape this transition. The question isn't just how to prepare for a
world where AI can do more of what people do currently. It's how to create a world
where people – all people – can do more of what truly matters – exercise judgment
and give voice as to which goals are worth pursuing and why.
In the next part of this series, we'll explore the stubborn persistence of resource
constraints, even in a world of AI abundance.
