AI quickly identifies new species. Can we trust the results?

Lina Portman3 days ago

0 3 minutes read

Scientists are using artificial intelligence (AI) to identify new animal species. But can we trust the results?

For now, scientists only use AI to flag potentially new species; highly specialized biologists have yet to formally describe these species and decide where they fit in the evolutionary tree. AI’s effectiveness also depends on the data we train it on, and right now there are huge gaps in our understanding of Earth’s fauna.

But AI helps researchers understand complex ecosystems by making sense of large sets of data collected through smartphones, camera traps and automated monitoring systems.

“We're accelerating the pace of research to be able to answer larger questions, and that's exciting,” Christine Picardprofessor of biology at Indiana University, told Live Science.

Related: GPT-4 Passes Turing Test, Researchers Say

In a 2023 study published in the journal Methods in ecology and evolution, Picard and his colleagues trained an AI model to classify more than 1,000 insect species. Live Science spoke with Picard and the lead author Sarkhan Badirliwho completed the study as part of his doctorate in computer science at Purdue University in Indiana.

The model learned to recognize species from images and DNA data, Badirli explained. During training, the researchers hid the identity of some known species, so they were unknown to the model.

“It couldn't tell what species it was, but our model could tell what genus it most likely belonged to,” Badirli told Live Science.

The model correctly identified 96.66% of known species and assigned species whose identity was not disclosed to the correct genus with an accuracy of 81.39%. However, the success rate was significantly lower when the model did not have DNA data and relied only on images: 39.11% accuracy for described species and 35.88% for unknown species.

The researchers attributed this problem in part to the low resolution of the images, which came from a public database. They noted that the model's accuracy would improve with experience and higher-resolution images.

“Some of these photos were actually pretty bad, so I can’t believe the model did so well with this data,” Picard said.

A photo of an insect exhibit with many exotic butterflies and beetles

There are more species of insects than any other type of animal, and AI could help monitor and identify them. (Photo credit: Mint Images via Getty Images)

Giving meaning to biodiversity

Most of Earth's biodiversity – the variety of animal and plant life – lives in the tropicswhich are among the poorest and least studied regions. As a result, much of the world’s fauna remains unknown. Insects have more species than any other animal group, but most of them have not yet been identified. AI could help fill this huge knowledge gap.

“This allows us to delve into this unknown space of insect species diversity,” Picard said.

Some scientists are also using AI to monitor entire ecosystems. These tools combine AI with automated cameras to see not only which species live in a given ecosystem, but also what they are doing.

Live Science spoke with Jenna Lawsona biodiversity scientist at the UK's Center for Ecology and Hydrology, who helps manage a network of AMIs (automated insect monitoring) systems. Each AMI system has a light and white board to attract butterflies, as well as a motion-activated camera to photograph them, she explained. The systems also record audio to identify animal calls and ultrasonic acoustics to identify bats. Powered by solar panels, these systems constantly collect data and, with 32 systems deployed, they produce an enormous amount, too much for humans to interpret.

“We have this incredible hardware, and we can use it to collect all this data, but without AI, we have no chance of analyzing it,” Lawson said.

Katriona Goldmanna data scientist at the Alan Turing Institute, is working with Lawson to train models to identify animals recorded by AMI systems. Similar to Badirli’s 2023 study, Goldmann uses images from public databases. His models will then alert researchers to animals that don’t appear in those databases.

“They'll be able to point out images and say, 'This looks like something I've never seen before,'” Goldmann told Live Science.

AMI systems also allow researchers to monitor changes in biodiversity over time, including increases and decreases. Researchers have estimated that globally, due to human activity, species are disappearing between 100 and 1,000 times faster than they would typically do. Wildlife monitoring is therefore essential to conservation efforts.

Lawson's systems will measure how wildlife responds to environmental changes, including temperature fluctuations, and to specific human activities, such as agriculture.

“The advent of technology in biodiversity research has been fascinating because it has allowed us to record at a scale that was not possible before,” Lawson said.

The irony of these AI systems is that AI algorithms are incredibly energy-intensive and can therefore have an outsized impact on the environment, the report said. London School of Economics and Political Science.

So Goldmann trains his models on supercomputers, then compresses them to fit small computers that can be attached to the energy-saving units, which will also be powered by solar energy.