AI was using posture rather than beaks, which the team had initially thought would be easiest, to tell birds apart. Photo: Supplied
A study to identify individual kākā using AI has received a funding boost of $1 million, one of a number of research projects using artificial intelligence "for good".
The funding comes from the Ministry of Business, Innovation and Employment's (MBIE) Smart Ideas fund, which this year received 351 applications, and approved 53 of them.
The kākā project is a collaboration between behavioural ecologist Dr Rachael Shaw and senior lecturer in artificial intelligence Andrew Lensen, both from Victoria University of Wellington, who have been working on a new way of identifying individual birds to make monitoring easier.
Lensen said it was increasingly difficult to keep track of individuals as they spread out beyond Zealandia, where the species had flourished in recent years.
Not all birds were fitted with leg bands, and sometimes even those could become discoloured or hard to see, making keeping track of them tricky - and the city presented new threats, like disease, conflict with humans and unintentional poisoning.
The project began with photographs taken in nesting boxes at Zealandia, perfectly set up to allow snapshots of the birds heads in profile while they fed.
They were now achieving 90 to 95 percent accuracy with this method, Lensen said, but in a surprise twist, the AI was using posture rather than beaks, which the team had initially thought would be easiest, to tell birds apart.
The next step would be to scale up the complexity and test the AI's ability to identify birds against a background of trees, rather than the inside of a feeding box.
The kākā project is a collaboration between behavioural ecologist Dr Rachael Shaw and senior lecturer in artificial intelligence Andrew Lensen Photo: Supplied
Lensen said the funding would allow them to put in more cameras, test new methods like deep learning, and lean more into a mātauranga Māori approach.
"There's a big chunk of research around matauranga Māori, and trying to reconnect Māori with kākā," he said. "Before colonisation, they were seen as gods, or atua."
Three proposals from the Victoria University faculty of engineering had received MBIE funding this year, all using what Lensen called "AI for good": the kākā project, another using AI to recognise kiwi calls and another for ice sheet modelling.
"It's a reflection, I think, of where we are as academics," he said. "In contrast to in industry, where there is very much a capitalistic focus, we are quite privileged in that we have this freedom, or in some ways this mandate, to do good things with the technology."