Major Australian bank Westpac has said early experiments with using AI tools saw a 46 per cent increase in the productivity of its software engineers and caused no noticeable drop in code quality.

David Walker, CTO of Westpac Group, described the experiment in a LinkedIn post, saying the company randomly split 60 software developers into four different groups. Three of the teams used AI coding tools from Microsoft, Amazon, and OpenAI while a control group coded by hand.

“Each team was given the exact same challenges,” Walker said.

“The results were clear. All the AI tools provided significant benefits with an average productivity gain of 46 per cent across all AI tools.

“What’s more, this productivity uplift did not come at the cost of code quality, with no difference in code reliability, security vulnerabilities, and maintainability across the groups.”

Walker said it was important for developers to see these tools as aids designed to augment their existing skills “rather than rely on them completely”.

He also shared some testimonials from Westpac developers, including one who said they had minimal experience coding in Python yet was “quickly able to get functioning python code to do exactly what I wanted just by asking the AI the correct questions”.

Early this year, an expert in the future of work described to Information Age the type of experiment Westpac has now performed, saying it was important for business leaders to start gaining “realistic measures of what AI can do” or risk being left behind.

Meanwhile, academic researchers have started publishing preliminary, non-peer reviewed, results of their own experiments with generative AI tools in different contexts to measure productivity gains.

One experiment gave 444 professionals in different fields short 20-30 minute writing tasks such as writing press releases, reports, analysis plans, and emails. Half the participants were instructed to sign up for ChatGPT in between their first and second task.

The result was a reduction – by an average of 10 minutes – in the time taken to complete the task alongside an increase in the quality of their work.

Most (68 per cent) of participants said they submitted work straight from ChatGPT without any editing.

Another experiment saw customer service agents work alongside an AI chat assistant that monitored their chats with customers and offered real-time suggestions for how the agents should respond.

Having an AI assistant helped the customer service agents resolve 14 more conversations per hour.

Interestingly, the AI assistance disproportionately increased productivity for less skilled and less experienced workers, with higher skilled workers having “few positive effects” from AI interaction.

“We posit that high-skill workers may have less to gain from AI assistance precisely because AI recommendations capture the potentially tacit knowledge embodied in their own behaviours,” the researchers of that paper said.

Australia has a relatively low rate of AI adoption with local trainees “yet to embrace ChatGPT as a skill”, according to stats from learning platform Udemy.

A recent survey from Microsoft found almost half (46 per cent) of Australian respondents were worried AI would replace their jobs.

Yet Australians also said they would be more than happy to offload their administrative, analytic, and creative work to an AI with 64 per cent of respondents saying “they would delegate as much work as possible to AI in order to lessen their workloads”.

Local business leaders were around twice as likely to say that AI would add to productivity rather than lead to job losses, and 85 per cent of Australian employers telling Microsoft said they would need to hire employees with new skills to prepare for AI.