AI is making us faster. Why aren’t we more productive?

By Aaron Kosovich, Economist, Programmer and Founder

This seems odd given we are supposedly in the midst of an AI-fuelled technological revolution, one that is already consuming significant amounts of our energy, water and land. What strikes me is that, though AI is improving the productivity of the tasks we perform, it isn’t improving the productivity of the economy at scale. As an economist, I’d like to explore why this is and suggest how Australia can use AI to genuinely increase our productivity.

Technology usually increases productivity

Before computers, large organisations employed rooms of clerical workers to type documents, copy records and perform calculations by hand. Later on, computers allowed many of these tasks to be completed in a fraction of the time. They are now able to produce far more with fewer hours of labour.

It seems reasonable to expect something similar from AI as it is definitely making some people faster. After all, it can draft a report, summarise a document or write a piece of code in seconds. But these individual gains have not yet become economy-wide gains.

Why working faster hasn’t increased output

Most office workers are still employed for the same number of hours and, consistent with Parkinson’s Law, the time saved through AI is often absorbed into existing work. Some may be producing better work or completing individual tasks faster, but that doesn’t necessarily mean that their organisations are producing or selling more. These gains are yet to translate into a measurable productivity boost and as it turns out, task efficiency does not translate to economic productivity at scale.

In order to improve productivity in these industries without increasing unemployment, we need to help our firms use their additional capacity to serve more customers, improve what they offer, develop new services or compete in larger international markets. When our companies export their services internationally, they increase the overall size of our economic pie, their output, and our productivity.

That said, the largest increases to productivity from AI are likely to emerge elsewhere.

Where AI could make a bigger difference

We can expect AI to best scale output in industries like agriculture, mining, and manufacturing because their work depends more heavily on physical assets: land, machinery, energy and production systems. Small improvements in how these assets are used can translate into large increases in output.

AI could help farmers use water and fertiliser more precisely. It could help manufacturers detect defects earlier, reduce downtime and use less energy. It could help mining companies improve exploration, maintenance and safety. These applications can make production more efficient and more sustainable.

Yet adoption has been slow because the type of AI currently used by the services industry does not translate well to the rest of the economy. We can’t ask a chatbot to increase the yield of a farm or allow for more precise mining exploration, because it simply doesn’t have those features yet. These solutions require complex ecosystems of sensors, machinery, historical production data, location information and software arranged vertically for those use cases.

This explains why the first wave of AI adoption has been concentrated in offices. Giving an employee access to a chatbot is easy, but embedding AI into a physical production system with all of the bespoke hardware integrations and testing that would be required for each specific operation is really hard. Yet this is also what makes the opportunity so valuable. To realise these productivity gains we need to look beyond the first iteration of AI and invest in specific technologies.

Beyond chatbots

For most of us, our first interaction with AI was the release of ChatGPT, which gave millions of people direct access to a general-purpose AI model. But interacting with a chatbot is not the final form of AI any more than using a command line was the final form of computing. The larger productivity gains are coming from applications built around particular industries, workflows and production constraints. Making more of them will require backing smart people who can build tools tailored to real problems.

Industry needs startups, and startups need industry

Large companies are generally designed to execute proven business models. They are good at making established systems work reliably at scale, but usually less suited to pursuing uncertain ideas that may fail. Startups have the opposite characteristics. They can move quickly and test ideas cheaply, but they lack the assets, industry knowledge, customers and capital held by established companies. If we are to increase our productivity, we need both working together.

Established companies should become early customers and development partners for Australian technology companies. By doing so, they can provide the all-important problems, resources and operating environments that startups need. In return, startups can provide the speed and experimentation that large organisations find difficult to sustain internally. This interplay is much more ambitious than simply providing our employees with access to existing AI models and carrying on with business as usual.

From task efficiency to economic productivity

Using AI to do our jobs faster has improved the speed at which we can get things done, but is yet to translate into broader productivity growth. Meanwhile, the data centres and computing infrastructure behind it already consume vast amounts of our energy, water, and land.

Australia’s productivity will lift when the way we use AI moves beyond completing existing tasks faster and becomes embedded in the products, machinery and processes that expand what we can produce and sell. Done right, building and implementing these solutions could deliver the productivity multiplier we have been promised, help offset AI’s own environmental footprint, and create a powerful export industry in its own right.