Compiled by the editorial desk with reference to the study published in PNAS and public information about the Snapshot Serengeti project.

Ecologists have long struggled with a data bottleneck: motion-triggered cameras in the wild generate millions of images, but extracting meaningful information from them is a slow, manual process. A new artificial intelligence system, described in a study published in the journal PNAS, offers a way to automate this task, recognizing 48 animal species and their activities with near-human accuracy.

The algorithm was trained on a vast dataset from Snapshot Serengeti, a citizen-science project that uses camera traps in Tanzania's Serengeti National Park. Volunteers had previously labeled millions of photos, providing the labeled examples the AI needed to learn. After processing those images, the system could identify species such as wildebeest and gazelle, and classify behaviors like resting, eating, or social interaction.

In tests on new photos, the AI achieved 99.3 percent accuracy, only slightly below the 99.6 percent typically reached by human volunteers. The errors occurred mainly on ambiguous images, such as a close-up of an impala's leg that the AI mistook for a zebra. When such difficult cases were set aside for human review, the AI matched human performance on the rest.

The time savings are substantial. According to the researchers' calculations, the AI could process the entire Snapshot Serengeti dataset in 17,000 fewer hours than a team of human volunteers would require. That efficiency could free ecologists to focus on analysis rather than data sorting.

Why Automation Matters for Conservation

The backlog of unexamined wildlife images has real consequences. Many ecological discoveries come from revisiting archived data, and gaps in observation can leave basic questions unanswered. For instance, scientists only confirmed last year that wild aardvarks drink water, simply because no one had previously witnessed it.

Automated classification tools could help close such gaps by making it feasible to analyze large datasets quickly. With a clearer picture of how animals behave and interact, ecologists might identify patterns that inform conservation strategies, particularly as habitats face pressure from industrialization and development.

The study's authors note that the AI is not meant to replace human volunteers entirely. Instead, it can handle the bulk of the work, with humans stepping in for the most challenging images. This hybrid approach could make the most of both human judgment and machine efficiency.

As camera-trap networks expand globally, tools like this may become essential for turning raw images into actionable ecological knowledge. The technology is still in its early stages, but the potential for accelerating research is clear.