Britain's largest ever woodland survey of bats would "simply not have been feasible" without artificial intelligence, the scientists behind it say.
Researchers left recording devices on poles among the trees of 310 woodlands across Britain and captured 98 million bat calls over three years, BBC News reported. Analysing that volume of audio without AI would have taken more than thirty years.
The survey was run by the Bat Conservation Trust alongside Forest Research, and involved University College London.
Some of Britain's rarest and most threatened bats were recorded in the south-west of England, including in Somerset, Devon, Cornwall and Dorset. Among them was the red-listed Barbastelle, a species that roams ancient forests and whose elusive nature has long made it difficult to track. Hotspots of Barbastelle activity detected by the survey were concentrated across the South West.
"AI has fundamentally changed the scale at which we can monitor wildlife," said Kate Jones of University College London, who called the project a "nice story of AI for good".
"Previously, analysing this volume of acoustic data would simply not have been feasible," Jones said. "These efficient open-source AI tools allow us to identify millions of bat calls across thousands of nights of recordings, identify patterns and flag species at a national scale."
Lia Gilmour of the Bat Conservation Trust called the project "unprecedented". "We're getting some really good bat activity in the south-west area of Somerset," she said. "Somerset is generally a really good stronghold for the bat species we have across the UK." She added: "In biodiversity monitoring, we view AI as a powerful tool. We were looking at huge volumes of data and 90,000 hours of recording, there was no way we would be able to do this manually."
The arithmetic inside that last sentence is the most interesting figure in the survey, and deserves to be set out. Three years of recording would have required more than thirty years of listening. The data arrived roughly ten times faster than a person could hear it.
Had the work been done by ear, the bats recorded at the start of the survey would have been identified somewhere around 2056. The report would have described the national distribution of British woodland bats as it stood a third of a century earlier, and it would have been entirely accurate.
Our analysts note that this would have been the first British wildlife census in which the population could have turned over while being counted, and in which an honest report would have had to carry two dates: one for when the bats were there, and one for when anybody found out.
The Department of Unnecessary Research has costed the alternative and finds the researchers' judgement sound. A listener working through 90,000 hours of woodland audio would face something close to a thousand bat calls an hour, in the dark, for thirty years, in order to establish the range of a species whose defining characteristic is that nobody hears it.
The bats, for their part, produced 98 million calls in three years and were understood within one. Nothing in the survey suggests that they noticed the difference.