Story | 20 Jul, 2023

Computer Conservation

AI and machine learning is helping to advance conservation efforts in many amazing ways. Sam Perrin and Tom Ireland explore the possibilities and limitations of this fast-moving technology

content hero image

As anyone who has interacted with the chatbot ChatGPT will know, artificial intelligence (AI) is developing at a remarkable speed. In just a few months the program has gone from a curiously talkative search engine to a powerful tool that can research, translate, code, problem-solve and even start an online business for you, with just a few simple prompts.

With humanity facing an urgent need for effective and affordable action to protect the natural world, many organisations are now turning to AI (and related technologies such as machine and deep learning) to refine and advance their conservation efforts.

The technology has initially found use in what is essentially very fast pattern-recognition, helping to identify glimpses of rare species hidden within millions of images or hours of video, or helping to identify early signs of environmental problems over vast areas. In China, satellite imaging has been combined with AI technology to detect forest fires earlier and automatically alert local management, cutting serious fires down to a third of what they were previously. In Wisconsin, USA, AI-driven camera systems have been installed on windfarms to instantaneously recognise threatened species of birds flying towards them and slow the turbines. And Kafue National Park, Zambia, has installed a 19km surveillance line, fitted with infra-red cameras and powered by intelligent technology that can identify poachers and alert local rangers.

CAPTAIN of conservation

More advanced applications of AI involve algorithms that can be trained to model large-scale conservation actions or suggest areas to prioritise. Several AI tools are now available, including Zonation, CAPTAIN and MARXAN, which can help conservationists identify regions most in need of biodiversity protection, or where action could have the most impact.

CAPTAIN (Conservation Area Prioritization Through Artificial Intelligence) feeds biodiversity data, conservation budgets, climate change models and human pressures into a neural network (a series of algorithms that aims to mimic the way the human brain operates). The program quantifies the trade-offs between the costs and benefits of area and biodiversity protection, exploring multiple biodiversity metrics.

CAPTAIN essentially plays a game in an artificial, simulated world, aiming to save as many species as possible from extinction in various scenarios. Each time, the software learns how to best place protected areas in its simulated world. Powerful platforms like Nature Metrics are helping organisations pull together information from things like environmental DNA to build a picture of the composition of an ecosystem and how it is changing. The Silicon Valley-backed start-up Basecamp Research is using a vast AI database of DNA sequences from around the world, to try to understand what the world’s unstudied proteins do. They hope this will help countries value the biodiversity within their biomes. 


Technology is helping to protect baboons and other species in Kafue National Park


Predicting threats

Other applications can take data from existing or past conservation work and use it to make predictions about habitats or species that have not yet been studied or assessed. For example, scientists at London’s Kew Gardens are testing if machine learning can take existing data on species extinction risk (such as IUCN’s Red List of Threatened Species) to predict which of their plants are most threatened2.

Dr Binbin Li, of Duke Kunshan University in North Carolina, USA, is currently using AI to track species throughout China, including charismatic and rare species such as takins, musk deer and giant pandas. “It’s so exciting,” she says. “We’re just scratching the surface at the moment, using AI to discern species from images. But in an ideal world we’ll be able to identify species from sound alone – even if they don’t show up on camera traps – and start to get an idea of population size.”

A key benefit of using these rapidly-advancing self-learning technologies for conservation is time. In a recent lecture on AI by the World Wide Fund for Nature, Professor Bistra Dilkina described sets of data that would previously have required well over 10 years of computing time to process. Yet with a machine-learning model that processed data and trained itself to recognise patterns on the go, her team was able to quickly produce predictions for bird migrations on a continental scale.

In the project from Kew Gardens mentioned above, Kew’s species risk software took just one day to model the threats faced by 47,659 plant species. Systems that can plough through data – and learn to do it more efficiently at the same time – should in theory free up more of conservationists’ time for fundraising, putting plans into action, and strategising.

AI is also helping people outside the scientific community to contribute to ecological monitoring and conservation. Data collected by non-specialists has been criticised in the past as being inaccurate, but the incorporation of AI into apps such as Seek/iNaturalist and Merlin/eBird enables people without years of academic learning to provide useful photo data that can identify organisms to a 

species level. It means places beyond heavily surveyed areas don’t go completely unsampled, as long as they’re available to local groups, hikers and even holidaymakers.

“It allows community scientists to be more productive in what they gather, and have less of a taxonomical bias,” explains Dr Wouter Koch, one of the researchers behind Artsobservasjoner, a Norwegian reporting tool for community scientists produced by the organisation Artsdatabanken (The Species Database). “They can report more than what they already know, which is often mainly birds.” As well as adding valuable species observation data to global databases, tools like Artsobservasjoner are helping bring more ordinary people into the global mission to understand and protect nature.

IUCN recently partnered with Chinese tech giant Huawei to release a report titled Tech4Nature: Solutions in Focus3, which featured numerous examples of AI aiding conservation efforts, from rebuilding coral reefs in Mauritius to boosting saiga populations on the Eurasian steppes, or the use of an AI-powered trap to protect Atlantic salmon from invasive rivals off the northern coast of Norway. The report was part of the wider IUCN Tech4Nature initiative, a global partnership to scale up success in nature conservation through digital technology innovation.

Better connected

IUCN can also play a role in connecting experts in AI and machine learning to the conservation community, says Dr Milind Tambe, Director of the Harvard Center for Research on Computation and Society. “Many AI researchers around the globe have the skills and desire to work on issues important to IUCN,” he says. “But they have no idea how they could get started, who they could talk to, or where they could get the data.”

Tambe’s team was one of the first to apply AI models to global anti-poaching efforts, as part of the PAWS project – the Protection Assistant for Wildlife Security. The system takes basic information about a protected area, and information about previous patrolling and poaching activities, and generates the most effective patrol routes for rangers. As they execute the patrol routes, more poaching data is collected and fed back to PAWS.

For those thinking of exploring AI in their conservation work, many AI platforms are easy to find online and are open source, so are free to use and modify. But experts warn of the perils of attempting to use ‘off-the-shelf’ solutions or algorithms ‘trained’ in other areas of the world. The story of the wildlife monitoring tool that began to identify giraffes in the snowy Canadian city of Edmonton shows how AI can be particularly prone to making errors until
it is given data that is fit for its purpose. If the data used to train the AI is poor, subsequent identifications of patterns will be poor too – or as computer scientists like to say, “garbage in, garbage out”.

The biases in existing studies may be amplified by an AI program, with potentially disastrous consequences. A poorly trained algorithm for recognition could result in false positives (potentially draining resources when looking for a rare species) or false negatives (which can be disastrous if you’re looking for an invasive species or a forest fire). The key to getting it all right is partnerships, says Tambe, and “for AI researchers to be partnering with conservation agencies, all the way from data to deployment”.

Dr Renee Sieber, an associate professor at McGill University and an expert in the use of IT by community groups, has identified six key points
for the good and ethical use of AI in conservation. They are: avoiding ‘off-the-shelf’ solutions; ensuring that quality of data is fit for the purpose and that the potential for harm (if it is not) is considered; that the technological process is fair, explainable and transparent; that issues around privacy and surveillance are addressed; that it balances the rights of individuals, communities and wildlife; and thatusers know when to draw red lines.

Many AI researchers around the globe have the skills and desire to work on issues important to IUCN

Satellite Images of wild fire