- In short: The use of artificial intelligence (AI) is becoming more common among marine ecologists and conservationists. AI is being employed to listen to hundreds of thousands of hours of sound recordings to help map the whereabouts of critically endangered blue whales.
- Researchers from CSIRO are also using AI to map critically endangered seabird populations, a task that was previously done manually.
- What’s next? The technology is still prone to error, meaning humans will continue to play a role in monitoring and fact-checking results.
The Antarctic Ocean is remote and hostile. It's also teeming with life.
Brian Miller can verify this.
Over the course of his career, the ecologist and bio-acoustician has made hundreds of thousands of hours of audio recordings of rare and critically endangered marine mammals that call that ocean home.
The sounds he records provide crucial clues about the habits and whereabouts of whales, dolphins and seals that are often heard, but rarely seen.
But there's a problem.
"We generally have one opportunity per year. We get these huge volumes of data coming back from Antarctica," Dr Miller said.
The Southern Ocean is a notoriously difficult and expensive place to get to, with safe passage only possible during a short window over the summer months.
When Dr Miller does eventually make it to there, he sends half a dozen recording devices — ex-military hydrophones known as sonobuoys — into the dark, freezing ocean, where they record for a up to a year at a time.
"If I were to listen to that in real time, it would take me one year to listen to it," Dr Miller said.
"In order to manage species, in order to do conservation, you need to understand how many of these animals there are, where they are, when they're there, and how the populations are structured.
"And listening to them provides us a really good idea of when and where they are."
This is where AI comes in.
A 'renaissance' for conservation research
Machine learning — the ability for algorithms to find and analyse patterns in data — has become a valuable way to speed up what can be a tedious process for Dr Miller.
It's an unremarkable thing to watch — a computer whirrs through information, grey bars move across the computer monitor — but the implications of the technology for research is, for Dr Miller, the quietly revolutionary part.
"Algorithms are getting better and it's almost a renaissance for this field," Dr Miller said.
"It's fantastic to have these algorithms be able to listen to [the audio] for us and highlights the bits that are really novel, new, or just extremely high quality."
Each marine species he tracks has a unique soundwave pattern they make, which an AI algorithm can be trained to spot.
After the algorithm trains on a big enough data sample, it begins to recognise those sounds.
"That's turned out to be a really, really powerful technique for very quickly analysing large volumes of audio that we're recording," Dr Miller said.
"That allows us to get on with the task of understanding their ecology and focusing on the questions: Are they recovering after being nearly wiped out during industrial whaling, how has it changed, how are they faring in a changing climate?" he said.
And the more information the algorithm is fed, the "smarter" the technology becomes.
"Listening for them isn't a conservation action in and of itself, it's providing the knowledge that's needed for management," Dr Miller said.
"But before we can do that, we need the best available knowledge and listening for them is a great way to enhance the state of knowledge."
How a camera is learning to count threatened seabirds
The technology isn't just a good listener, it can also see.
Further north, off Australia's south-east coast, a team of researchers from the CSIRO have just wrapped up a six-week voyage to road test their new AI technology.
A CCTV-like camera fixed to the RV Investigator has been learning to "count" seabirds in the Tasman Sea, a task previously performed manually.
"It gives you an accurate idea of how many birds are out there, but also what kind of interactions these birds are having," CSIRO research technician Carlie Devine said.
It's crucial information, with bycatch in fisheries considered one of the greatest threats to seabirds while at sea.
Many seabirds, like the shy albatross, are already considered a threatened species.
Ms Devine said the technology is a step towards implementing real-time change for commercial fishing vessels, who can implement mitigation measures to stop birds getting caught in their nets.
"The beauty of [the technology] as well is that it performs really well when there's quite a lot of birds," she said.
"It's a difficult task for a human."
Humans still have a role to play
There are drawbacks to the new technology, however.
"Even though it's quite a fast output, you do need to spend the time to train the models," Ms Devine said.
"If you point a camera at a different location, then you do need to train more."
Ms Devine said humans were, for now, still necessary in the counting process.
"There were a few false negatives... the camera machine learning algorithm thinks that it has seen a bird and in fact, there's nothing there," she said.
"So, we have a little bit more training to do, but that's just the nature of machine learning."