Dr Emma Spencer, WWF-Australia

AI algorithms aid image classification of Australian wildlife species in post-fire habitats as part of the Eyes on Recovery project.

Our natural world faces increasing pressure from global threats like climate change, deforestation, over exploitation, pollution, and the spread of invasive species. To track and manage these and other large-scale threats, we also must take a large-scale approach to data collection.

There already exists an impressive variety of technologies that can be used to quickly build these big datasets—from high-resolution optical remote sensing to camera traps, acoustic sensors and more. But manually processing the huge volumes of data that these technologies produce can be cost and time prohibitive.

This is where artificial intelligence can help. AI technologies increase efficiency by simulating human problem-solving and decision-making capabilities using computing systems, which can process most tasks considerably faster than the human brain. In the conservation space, AI can assist in processing massive datasets to inform management and conservation decisions far more quickly than is possible with human oversight alone.

There are many uses of AI currently being applied for conservation across the world. For example, in Africa, acoustic sensors and AI algorithms are being used to track real-time elephant movements and protect them from poaching in the Elephant listening project. Similarly, in Brazil a computer-based tool (PrevisIA) has been developed by applying AI algorithms to analyse satellite images and predict and combat deforestation in the Amazon.

In Australia, I have also been involved in a project that applied AI to help land managers assess wildlife recovery in the wake of the 2019-20 bushfires. This project, Eyes on Recovery, involved training a computer algorithm hosted on the Wildlife Insights platform to rapidly identify Australian species in images captured in the fire-impacted regions of south-eastern Australia. These technologies enabled our team to assess the impacts of a large-scale fire event on fauna, with the AI algorithm on Wildlife Insights allowing us to process over 8.5 million images and to identify around 140 wildlife species.

Checking a camera trap set up on Kangaroo Island, as part of the Eyes on Recovery initiative (image credit: Hannah Byrne-Willey).

Working in this space for the past three years has allowed me to see first-hand the incredible opportunity that AI represents in the field of conservation ecology. But it has also highlighted areas that require greater focus—especially concerning the ethics of using AI in conservation.

The ethics of AI is a widely discussed topic, with many examples of AI technology generating controversial outcomes. For example, certain recruiting methods that draw on AI algorithms were found to be gender-biased. Similarly, some facial recognition technology was shown to be less accurate for people with darker skin tones, with algorithms built to judge beauty also bias towards white people. There have also been several concerns raised around copyright protection and generative AI (like AI art generators), as many of these tools draw from extensive online databases without consent.

Discussions around ethics in conservation are less common but no less important. Limited, incorrect, or biased ecological training datasets can result in poor model performance, which could support ineffective and possibly detrimental management decisions. Biased training datasets could also perpetuate discrimination or unequal treatment in conservation efforts, for example, against less charismatic species or lower socioeconomic countries and regions.

The extensive cloud-based datasets generated real-time by AI technologies could also inadvertently provide cyber poachers (who hack online databases to guide illegal hunting activities) increased access to information on critically endangered species. Similarly, these technologies could be applied to more efficiently harm species considered worth conserving to some. For example, in Australia AI algorithms are being used to enable livestock producers to kill dingoes more efficiently.

There are also potential environmental impacts that arise from the production of associated computer hardware and the process of training and using various forms of AI technology. For example, training a single AI algorithm can emit more than 626,000 pounds of CO2 equivalent (or ~5x the lifetime carbon emissions of a car).

Privacy and data ownership is also a major consideration. Training datasets for AI algorithms may not be sourced ethically or respect the interests of the persons or groups who own the data. This is especially relevant when considering data generated by potentially marginalised groups like First Nations Peoples who protect around 80% of global biodiversity and are also responsible for collecting large volumes of data for conservation. Harvesting data without their consent risks human rights abuse, distortion of Indigenous culture and knowledge, and stalls collaborative design processes that could improve these technologies for all.

AI presents incredible opportunities in the field of conservation but as it become more common, we must all work hard to limit potential unethical applications of these technologies. Developing appropriate guidelines on AI that are complied with in research and conservation practices should be a priority. We also should ensure that these guidelines, as well as the development of the AI technologies themselves, are not exclusively informed by western perspectives but also take on the positions of many, including First Nations People.