In the wet areas of Senegal, researcher Alexander Dale Plink Pilot Pilot for Waterbirds Counting: Pelican, Flemingo and Trinns. He said he flies the drone, but analyzes photos to count on people in the AI flock, and accelerates thousands of hours of analysis in each survey. And time is the essence.
Since 1970, the population of forest life has fallen More than seventy percent. The world is under the auspices of the Geovioida crisis and, according to some researchers, it is undergoing The sixth large scale. The planet has previously endured five massive events at the end of the Cretesis period: the time of the infamous Kishodarrah’s effects that released nuclear winter and killed dinosaur. It was sixty -six million years ago.
First, first to protect the species from the brink of extinction, first you have to know what you have, and how many – which are often easy, especially in fields that have a lot to count. Scientists make less estimates than 20 % of worms species Identified on the ground. When AI reviewed only a week -long camera trap footage in Panama, researchers say they say Found more than 300 species The first was unknown to science.
Scientific Research is the basis of AI Not without critics. High -tech supporters refer to AI’s ability to analyze large datases in seconds, which will take months, with AI to understand the samples in the conversation and distribution of unidentified species for humans and expose a cycle of genomes. Critics indicate its environmental impact, the possibilities of prejudice and moral standards.
Most of the AI work in protection is focused on analyzing the thousands of hours of footage taken from a remote camera or aerial survey, but there is unlikely to end it. For now, researchers are focused on taking footage with object detection models, a type of AI that can identify and detect items in an image or video. These models are often made with virtual nervous networks (CNN) and are trained to identify the species or detect their presence or absence.
Plans for AI employing AI to “save species” often create media frenzy. Researchers in South Africa had a stir of headlines as to what AI can save “The lonely plant in the world.” Scientists have deployed drones at the inaccessible jungle of a female partner in London’s Qi Botanical Gardens in search of a female partner. AI scan footage for a species of species in the wild, whose researchers hope that is not really extinct – just under the umbrella. But some say the headlines are suppressed without considering the results.
“AI, which looks at environmentally and socially spending, is a wave of enthusiastic research on environmental and less critical research,” said Hamesh Van Dear Wen, head of the business, stability, and technology lab at the University of British Columbia.
The training process for the AI model, such as a large language model (LLM), can use more than a thousand megawatts of electricity. Scholi Ren says, whose research is focused on minimizing the effects of AI’s health, says the data centers have water consumption.
The data centers have the infrastructure required to provide processing power for AI, and through the fresh water obtained from local water supply, all technology must cool down. Due to its cooling requirements, AI is likely to withdraw between 4.2 Billion and 6.6 Billion By 2027, every year cubic meters of water, most of which lose vapor. And environmental effects are not felt equally, as tech companies export to their data centers abroad. Google’s plan to build new data centers in Latin America was born Mass protests in Chile and UruguayBaodores region are already in severe drought.
“The data centers also create a health crisis due to air pollution emissions, including fine particle substances (pm2.5) and nitros oxide (NOX),” Ren said. In the United States, the burden of public health has been created by data centers located mainly in low -income areas. Twenty billion by 2030.
“The models we are running are not too much – they are big for us, but it’s not like a big social network data.”
Nevertheless, most of the biology AI works, for the moment, is not equal. For this part, Delplanic has a local computer processing, and its hernite model – which helps in the population of densely filled animals like elephants and deer on Swanna, which takes twelve hours over weekends during the training process.
“We are concerned as a scientist all the time: are we really hurting the environment we are trying to help? At least for the issues we are talking about, I don’t think, because the model we are running is not too big for us – but this is not a great social network, like a social network. Extraptive types have to be deployed.
But Competition Acologist Tanya Berger Woolf says current low -power applications are not using the full potential of this technology, citing image identification as “Old School AI”. Berger Woolf and Polk wrote a paper by writing a paper “Unrealized ability AI to enhance the knowledge of biological diversity ”
“We want to go beyond scaling and rapidly what people already do with something new, such as creating trial assumptions or removing non -visible patterns and combinations,” says Berger Woolf.
“What we have been doing so far with AI is clear, which is all about this fast image detection and sound surveillance, but we should work more than that: using AI to ask the right environmental questions,” says Polk.
A potential application that points to both the eligible and the condemnation, the concept of the use of AI to de -code the animal communication. Earth species project Generative AI and LLM are using the translator to communicate with inhumane life. There is also Project CitiWhich is focused on using a similar approach to understanding sperm whales, which communicate through the Morris Code -like clicks, which can be considered theoretically. Already, scientists have been able to employ machine learning Suggest elephants address individuals With unique names in their family. But the major basis for regulating animal communication raises ethical questions and concerns about success. In other words: Will it work? Is this a waste of resources to try? Should we talk to animals exactly?
“We have to choose where this model will make a difference, not just use them because you have a shiny new toy,” he warned. Applications like LLM promote a major mark to environmental maps, so “spending resources if the result of research does not change is irresponsible. And data is a source.”
The models are just as good as they are being trained on their data, which can potentially lead to misunderstandings in prejudice and protection measures. The most common issues include local prejudice, where species are better offended in some regions in data sets, and texinomic bias, where charismatic species such as pandas receive maximum financing and thus more and more data are readily available, they say an unclear. Van Derne Wayne argued, but AI can prejudice our comments and even form the questions we are asking. A paper writer How LLMS reduce environmental challenges.
“There are many more options for AI to offer prejudice, extract resources, and to work more than protection applications,” he said. “If we weigh the benefits of protection, how effective Amazon is being used to buy more and more things to consumers, this is a very uneven scale.”
In 2024, for its part, Google announces the deployment of AI model to listen to coral rocks: surforch. Biocostics play a key role in assessing stability. At the time of the deployment of this device, Google also announced that it lacks climate promise goals. AI’s environmental demands.
“It is not hypocritical to use AI in protection-it needs to be used responsibly,” said Barjar Wolf. But when it comes to the rules, neither biological diversity nor in accordance with geographical political boundaries, he said with astonishment.
