Markets, not government stewardship, will drive AI adoption

Last week’s report by Jobs and Skills Australia on AI in the workplace underlines how rapidly platforms like ChatGPT are being adopted in Australian businesses.

But its recommendations about how to respond fall badly short.

Effective AI adoption in business will not be brought about by government coordination (what the JSA calls “stewardship”). It can only be driven by market pressure on managers - and the fear of failure that pressure creates.

The JSA surveys a range of studies that estimate between a quarter and half of Australian firms have adopted AI in the last two years.

Frankly, that seems low. These studies usually survey managers and business owners. But ChatGPT alone has 700 million weekly users – nearly a tenth of the population of the planet. And, as the JSA recognises, workers are using AI without any clear approval, even knowledge, of management.

In a paper published a few months ago in Innovation, we called this “shadow user innovation” – a type of workplace innovation where employees adopt technological innovations themselves rather than waiting for the firm to adopt it for them.1

Generative AI is the archetypal shadow user innovation. Access is extremely low cost (reports suggest that most people use the free version of ChatGPT) so it fits easily into household budgets. Because we prompt it with nothing but our natural language, it requires no technical skill. And it can increasingly use many of our work tools, like email and cloud storage platforms.

But most importantly it is a technology for individuals. It doesn’t necessarily make teams more productive, it makes individual workers more productive. Its benefits are highly personal. Some heavy users have AI write their emails, other heavy users cannot think of anything worse. The way it fits into my workflows and habits will be very different from yours.

The problem for us as workers, and therefore for economy more generally, is not how to use AI more. It is to discover where AI fits.

This is not the first time we’ve seen a bottom-up technology disruption.

Slack (the group chat-like communications platform used by many workplaces whose features were aped by Microsoft Teams) was adopted in much the same manner.

Group chats encourage informal, fast, constant and ephemeral communication. They are far superior to email for teams who need to coordinate complex work. When Slack was adopted, it was adopted by small groups within companies, who tested it and then lobbied their managers to introduce the platform company wide.

This comparison is revealing. A decade on from the release of Slack, it is hard to imagine a modern corporation operate without Slack or Teams or their equivalents. Group chats fundamentally reshaped how firms work at their most basic level – how they coordinate and manage tasks internally – and that reshaping was forced from below, not above.

In fact, we could say that the entire software-as-a-service industry has subtly changed the relationship many workers have with firms. By subscribing to SAAS products in their own right (not just generative AI but Dropbox, Calendly, Otter … ) white collar workers have started moving towards a “bring your tools to work” model of employment in the digital age.

In keeping with its mandate, the JSA makes a range of recommendations around the development of skills programs. These include building AI capabilities into education and training at all levels and all qualifications, building out short form training, and training the tertiary education workforce in AI so they can pass knowledge on to students.

More questionably, the JSA also recommends the Australian government “steward” the AI transition - acting in a coordination role to impose principles on AI adoption (equity, productivity, proportionality and technological neutrality), establishing a “National Compact” and creating some sort of national institution to coordinate AI adoption.

But these recommendations are starkly at odds with what we – and the JSA – know about how AI is being adopted in the real world. The AI transition is coming from workers themselves. It is not firms that are adopting AI but employees. Bottom-up adoption is a fundamental characteristic of generative AI. And a bottom-up problem implies that the solutions are going to have to be bottom-up too.

In our paper we argue that the real challenge is how firms can encourage employees to share among each other how they are using the technology, and how the organisation can capture the productivity gains of its use.

The unfortunate truth is the profound AI literacy deficit is more in management than workers.

When ChatGPT was first released there was a corporate panic over privacy – a fear that employees were dumping vast amounts of confidential information into their chatbot prompts. This fear was not entirely unjustified. But some firms responded with blanket ban policies that are hard to dislodge. Corporate policy revision cycles are measured in years. AI development is measured in weeks.

Many firms are genuinely confused about what is and is not AI. It is, admittedly, a bit of a term of art. Every software company now claims in its marketing to have “AI inside” or be “AI first”. This marketing is all very silly: I am happy to argue that computers have been doing “AI” since the very first devices were built during the Second World War. But the confusion has meant that natural organisational sluggishness is being compounded by managerial ignorance and fear.

In an important way the AI shadow innovation problem is a micro-level case study of the entire productivity debate.

The Commonwealth government wants to improve productivity across the economy. But productivity increases do not come from policymakers. They come from firms innovating and adopting new techniques and technologies.

The only way to accelerate such adoption is to further expose firms to competition, rewarding the innovators, and pushing productivity expectations higher.

Share