Labour as Fungible Compute: Why White Collar Work Will Die

You won't have a job soon (but neither will anyone else)

I’ve talked with a lot of white collar workers about AI recently, and one thing has become clear.

They still don’t get it.

They don’t see what I see.

They don’t see how obvious it is that, eventually, their job will be replaced by AI.

So I’m going to break down exactly why this is obvious to me, and by the end of this piece it should also be obvious to you too.

Let’s start with the basics.

What is human labour?

Human labour is the exertion of energy across space and time to achieve a goal in exchange for money.

To date, the make up of our economy has focused on the concept of specialisation:

You go to Uni, study a specific degree, then usually get a job that relates to that degree.

The premise is that people can produce more output for an economy from specialisation, then generalisation.

In other words, they are more productive in the labour force because of it.

The deeper they go into their profession, the more they can produce with less effort.

But the value of specialisation of labour only holds true in a world where capital cannot be used to do the same task that labour can.

If you can instead allocate capital to do the task an employee previously did, the labour of that employee becomes redundant, provided the cost of capital is lower than the cost of labour.

By this point you might be thinking:

‘But RA, how can capital in and of itself replace labour? What are you allocating the capital to?’

In the future, the majority of capital relating to white collar work will be allocated to compute - servers running AGI (artificial general intelligence) - rather than human labour.

AGI, by definition, is capable of doing anything a human can in terms of knowledge work.

We do not have AGI yet, but many believe we may have it within the next few years.

Most would agree we’ll have it within a decade at the latest.

And so, the world we are rapidly approaching is one where knowledge work becomes a commodity: cheap and abundant.

This is a world where the only limit to intelligence is the substrate the intelligence runs on, which in our case is the data centres used for AI training and inference.

Those data centres require energy.

So, we can deduce that the real bottleneck to scaling intelligence to everyone is simply the cost of energy.

If labour becomes nothing more than ‘how much energy is needed to perform a task?’, and that is fungible because it runs on the silicon substrate (servers) instead of biological substrate (humans), can be parallelised and deployed across millions of instances at the same time, there is no functional distinction between capital and labour.

In other words, labour becomes fungible compute.

It’s fungible because it doesn’t require specialisation - one server can run your lawyer AGI, your doctor AGI, your software engineer AGI, etc (it’s all the same AGI instance, capable of doing any number of expert jobs).

AI is categorically different from previous technology revolutions, because it is a general purpose intelligence, not a general purpose utility.

Intelligence is capable of creating intelligence itself.

It’s recursive.

Why is this different from every other technology in human history, and why should you care?