Human Capital: What Grows in Value When Everyone Has the Machine?

The first automated teller machines were installed in 1969, and by 1990 there were about 100,000 of them. Many expected the human teller to disappear. Yet the data of the economist James Bessen show that the number of tellers did not fall as ATMs spread; it stayed roughly stable. Two reasons are given. Opening a branch became cheaper because it needed fewer tellers, so branches multiplied, and the teller's work moved from counting cash to serving customers and offering a wider range of banking services.

This does not mean every profession will survive, or that history repeats itself in the same form. But it reminds us that the more precise question is not 'Will my job remain?' but 'Which of its tasks will pass to the tool, and what is left for me?'

What do the experiments say?

In a randomised experiment with 444 college-educated professionals, half were given access to ChatGPT. The time needed for a short writing task fell by about 37%, and quality scores rose by about 0.4 standard deviations. What matters most for our subject is that those who had performed worst on the first task benefited most: the correlation between scores on the first and second tasks fell from 0.49 in the control group to 0.25 among tool users. In other words, the tool brought people closer together in producing a draft.

The researchers themselves qualify this result: the tasks were short and self-contained, and may not apply to jobs that need organisation-specific context. But the idea stands: when a good draft is available to everyone with little effort, the draft is no longer the difference.

Where does the difference move?

The editorial reading is that it moves to the two ends of the draft. Before it: which problem deserves solving, and what context surrounds it. After it: is the output correct, and does it suit this customer, this patient or this organisation. What can be written down in a book or a rule, the tool copies at trivial cost; intuition honed over years of experience is harder to transfer. This is an inference from the evidence, not a measured finding in itself, and it deserves to be tested profession by profession.

When confidence tempts

A survey of 319 knowledge workers (Lee et al., 2025) collected 936 examples of generative AI use at work. It found that higher confidence in the tool was associated with less critical thinking, while higher confidence in one's own ability on the task was associated with more. The data are self-reported, and the relationship is a correlation, not a cause. Even so, it is enough to give pause: someone who always lets the tool think for them may find it hard to know when it is wrong.

A Saudi reading

In 2021 the Kingdom launched the Human Capability Development Program, one of the programmes for realising Saudi Vision 2030. Its goals include aligning educational outcomes with labour-market needs and developing skills at every stage of life, from early childhood to continuous learning. This direction agrees with what the evidence suggests, and with an idea the Nobel committee counted among the most noteworthy contributions of Gary Becker, the 1992 laureate in economics: that what is spent on knowledge and skill is treated as capital, not as an operating cost.

What is to be done?

• Make learning a renewed habit; a skill you trained in years ago is not a fixed balance.
• Try the tool in your own work, and note when it was right and when it was confidently wrong.
• Think through the problem yourself before you ask it, and keep a mental exercise that the tool does not do for you.

Taking these means does not conflict with trust in God; it is part of its fulfilment.