Augmented Human Intelligence: A Partnership Measured by Evidence

In July 1945 Vannevar Bush, director of the American Office of Scientific Research and Development, published his essay "As We May Think" in The Atlantic. In it he described a device he called the "memex", in which a person stores his books, records and communications and consults them with speed and flexibility, and he called it an "enlarged intimate supplement" to memory. He did not aspire to a machine that thinks in a person's place, but to one that keeps in front of a person's mind what he might otherwise forget.

In October 1962 Douglas Engelbart, in his report for the Stanford Research Institute, defined augmenting human capability as "increasing the capability of a man to approach a complex problem situation". Since then the idea of partnership has endured, and it now takes a form that can be tested.

What do the experiments say?

In a field experiment with 758 consultants (Dell'Acqua et al., 2023), those who used artificial intelligence completed 12.2% more tasks, at quality more than 40% higher. Lower performers improved by about 43% and higher performers by about 17%. In a study of 5,179 support staff, productivity rose by about 14% on average and by about 34% among the least experienced, with a smaller effect among the experts (Brynjolfsson et al., 2023).

This supports the idea of augmentation: the tool brings the beginner closer to the expert, and the study's authors suggest it passes some of the most skilled workers' practices on to others. But note that the tasks in both studies lay within a range the tool handles well.

When the tool slows you down

The first experiment has another side: on a task designed to fall outside the model's capability, those using it were about 19 percentage points less likely to reach the correct solution. And in a randomised trial run by METR in 2025, 16 professional developers completed 246 tasks in projects they had known for years; they took 19% longer when allowed to use the AI tools available in early 2025, although before starting they had expected the tools to speed them up by about 24%, and afterwards they believed the tools had sped them up by about 20% (Becker et al., 2025). The sample is small and the tools change quickly, so the result should not be generalised, but it teaches us not to rely on a feeling of speed.

Augmentation is therefore a result we measure in each setting, not an assumption we announce. A tool may help, make no difference or do harm, and what distinguishes the cases is a test on a specific task against a clear standard.

The skills that rise in value

The World Economic Forum's The Future of Jobs Report 2025, based on a survey of employers, expects about 39% of workers' current skills to change or become outdated by 2030, and about 170 million jobs to be created while about 92 million are displaced. These are projections, not a fixed destiny. Analytical thinking tops the core skills employers seek, followed by resilience and adaptability, then leadership and social influence. That is to say, demand is heading towards what a person needs in order to direct the tool well and judge its output.

What should be done

• Keep a human in the loop: a decision that affects people and their values is not delegated to a machine (see the article "Responsible Artificial Intelligence: Four Principles for Governance and Trust").
• Treat the output as a draft: ask for the source, check it, and credit those to whom credit is due.
• Measure before you generalise: try the tool on your real work and compare time and quality with and without it.
• Train for the good question and disciplined doubt: not for memorisation alone (see "The End of the Age of Rote Learning?").

The finest thing about this idea is that it preserves human dignity: the person holds the intention and the responsibility, and the tool is a means. If we aim through it at benefiting people, take the necessary precautions and rely on God, it will help us in knowledge and in work, and will not stand in for the scholar and the worker.