Content uploaded by Deborah Lupton
Author content
All content in this area was uploaded by Deborah Lupton on Jun 10, 2024
Content may be subject to copyright.
2
RMIT Classification: Trusted
Acknowledgement of Country
In the spirit of reconciliation, we acknowledge
the Traditional Custodians of country
throughout Australia and their connections to
land, sea and community. We pay our respect
to their elders past and present and extend
that respect to all Aboriginal and Torres Strait
Islander peoples today.
Suggested citation
Butler, E, Lupton, D. (2024) Generative AI
Technologies Applied to Ecosystems and the
Natural Environment: A Scoping Review,
ADM+S Working Paper Series 2024 (10), ARC
Centre of Excellence for Automated Decision-
Making and Society, DOI: 10.60836/k7fv-7991.
Copyright © 2024 Ella Butler, Deborah Lupton.
This is an open-access article distributed
under the terms of the Creative Commons
Attribution 4.0 International License (CC BY
4.0). The use, distribution or reproduction in
other forums is permitted, provided the
original author(s) and the copyright owner(s)
are credited.
ARC Acknowledgement
This research was conducted by the ARC
Centre of Excellence for Automated Decision-
Making and Society (CE200100005), and
funded by the Australian Government through
the Australian Research Council
3
RMIT Classification: Trusted
ABSTRACT
This report provides a scoping review of the literature on the ways that novel generative AI tools are being
applied to living things and other elements of ecosystems and the natural environment. The report outlines
several areas where generative AI is being deployed in new research projects and industry applications.
These include animal communication and agriculture and plant cultivation as well as environmental
sustainability, biodiversity, climate change and nature conservation initiatives. The report also details
some of the negative environmental costs and ethical issues associated with the manufacture, training
and infrastructure that support generative AI and large language models more generally.
KEYWORDS: GENERATIVE AI; LARGE LANGUAGE MODELS; ECOSYSTEMS; NATURE CONSERVATION; ANIMAL
COMMUNICATION; BIODIVERSITY; PRECISION AGRICULTURE; PLANT CULTIVATION; E-WASTE; ENVIRONMENTAL
SUSTAINABILITY; AI ETHICS; SCIENCE COMMUNICATION
4
RMIT Classification: Trusted
TABLE OF CONTENTS
1. Introduction ..................................................................................................... 5
2. Communicating with animals ........................................................................... 5
3. Livestock and poultry monitoring and management ........................................ 7
4. Plant cultivation ...............................................................................................8
5. Biodiversity preservation and nature conservation ....................................... 10
6. Climate change science and communication ................................................. 13
7. Environmental costs of generative AI ............................................................ 14
8. Conclusion ...................................................................................................... 16
9. References ..................................................................................................... 17
5
RMIT Classification: Trusted
1. INTRODUCTION
Digital technologies are currently used in many ways to monitor and measure living things and other
elements of ecosystems, such as seas and waterways, air, geological features and weather and climate
conditions (Fox et al., 2020; Lupton, 2023; McGovern et al., 2024; Steeneken et al., 2023). This report
outlines a number of areas where new data-driven software involving generative AI and large language
model (LLM) tools are used in applications directed at the living and non-living elements of the natural
environment and ecosystems. Generative AI is a type of machine learning that uses massive datasets to
produce text, audio or image-based outputs in response to prompts. LLMs specialise in natural language
processing to generate text that is similar to human language. These novel technologies have garnered a
high level of industry, research and mass media attention in recent years, led by OpenAI’s ChatGPT
software, which was quickly followed by competitors’ models (Sætra, 2023).
Generative AI is being used for projects involving a variety of non-human animal and plant species. Its
possible uses cover scales from the microscopic to the vast – from identifying the proteins that contribute
to basic biological functioning to monitoring the world’s oceans; from researching the habitats of tiny
zebra finches to tracking the movements of massive sperm whales. The technology is being used in
initiatives that directly implicate human and planetary health, such as assisting farmers in the Global South
who are most vulnerable to the effects of climate change. This report details the deployment of generative
AI in sustainability projects and biodiversity conservation as well as how this use creates environmental
impacts such as increased carbon emissions, energy use and water consumption. The report outlines
various uses of generative AI aimed towards animals, plants, biodiversity, conservation and climate
change. It concludes with some considerations of the impacts on the natural environment of these tools
and the need for ethical consideration of how they are deployed.
2. COMMUNICATING WITH ANIMALS
Recent developments in recording technologies such as hydrophones, biologgers and drones, as well as
the affordability of sensors, has meant that the field of biology has collected an ever-increasing trove of
digitised data about animals and their habitats. The vast amount of data presents a challenge, as there is
far too much for researchers to sort through manually. LLMs can handle these large datasets (Parshley,
2023). Animal communication is also multimodal, and vocalisations frequently occur simultaneously with
other kinds of behavioural gestures. This feature adds another layer of complexity to analysis. New AI
systems can encompass multimodal elements as part of sorting data (Parshley, 2023).
Several organisations are using AI both to decode animal communication and ‘talk back’ to animals. Recent
work in this area is related to generative AI through its focus on developing foundation models that can
6
RMIT Classification: Trusted
sort and categorise patterns within a dataset (Bromley, 2023; Parshley, 2023). Key initiatives include the
Earth Species Project, Project CETI (Cetacean Translation Initiative) and Interspecies Internet. These
initiatives are building on recent developments in AI in which languages are cross-translated into one
another, not by semantic understanding but rather by ‘aligning their shapes’ – mapping the frequency with
which particular words are close to one another in order to relationally predict them (Parshley, 2023).
The Earth Species Project is experimenting with generating animal vocalisations, ‘like an audio version of
GPT’ (Bromley, 2023). One study aims to do this with zebra finches held in captivity. These ‘interactive
playback experiments’1 involve vocalisations generated by an AI model. Playback experiments are being
conducted with captive animals rather than wild animals because AI can ‘speak’ but does not necessarily
‘know’ what it is saying, and the effects on animal groups are not yet understood.
The Earth Species Project has a second study that is focused on the Hawaiian crow. The Hawaiian crow is
extinct in the wild, but there are a limited number of individuals in captivity. The research aims to inventory
and catalogue the calls of the crows in captivity and compare these sounds to recordings of the last
known wild populations. The goal of the research is to identify calls that might be lost, such as those that
alert others to predators or that relate to courtship. The researchers hypothesise that these lost calls
might be part of the reason that efforts to reintroduce the crows back to the wild have been unsuccessful.
Generative AI translation projects applied to animals raise several significant ethical issues. One potential
issue is embedded in the development of the technology itself. Because generative AI could ‘speak’ to
animals without understanding what specific communications mean, it is possible that whatever it ‘says’
could cause harm to animal groups (Rutz et al., 2023). It is well known that LLMs such as ChatGPT can
produce convincing but factually incorrect answers to queries (Kocoń et al., 2023; Sallam, 2023). So too
might an AI chatbot generate ‘convincing’ calls that are potentially harmful to animals, with meanings that
are not yet understood by human researchers. In other words – as in many areas where generative AI
technologies are deployed – there is the potential to spread ‘misinformation’ to animals through
communicating with them using LLMs. Additional ethical issues include the targeted use of AI for uses
such as hunting, for example by listening for calls of specific species or imitating those calls to bring
groups of animals nearer (Parshley, 2023).
1 https://www.earthspecies.org/what-we-do/projects
7
RMIT Classification: Trusted
3. LIVESTOCK AND POULTRY MONITORING AND
MANAGEMENT
‘Precision farming’ (otherwise known as ‘precision’ or ‘smart’ agriculture), which involves generating data
about land and livestock from a variety of digital sensors and other data points, has been employed in
farming since the 1990s. It began with the use of global positioning systems (GPS), yield monitors and
geographic information systems (GIS). Since then, precision farming technologies been extended with the
development of ‘smart’ machinery such as internet-connected tractors, robotic feeding and milking
systems, sensors embedded in soil or carried on livestock, and drone surveillance systems – many of
which are part of Internet of Things networks that can share data with each other (Fox et al., 2020; Lockie
et al., 2020; Lupton, 2023).
With the advent of generative AI, attention has been directed towards how to use LLMs to contribute to
precision farming, including the management of livestock and poultry. In relation to the breeding and care
of animals, generative AI has been proposed as a way to set breeding goals by analysing consumer
sentiment and preferences. The use of AI systems drawing on large datasets of drone images to monitor
livestock and analyse milk samples from cattle has also been explored (Hayes et al., 2023). Predictive
modelling and simulation applied to animals and farming environments have further developed with the
use of LLM tools.
The possibilities of enhancing digital twin simulations with the latest generative AI software have been
investigated in this context. Digital twins are digital replicas of physical objects or systems that are
updated with data generated by sensors placed on or in the object or environment in real time. In precision
farming, these simulations are seen as a way to improve efficiency, leading to better profits, as well as to
contribute to the health and wellbeing of livestock (Neethirajan, 2023b, 2024). Sensor-generated and
remote monitoring data about the animals, including such aspects as their heart rate, body temperature,
drinking and feeding habits, body fluids and movements, are used in the digital twin simulations to create
scenarios to help farmers decide how best to manage their animals. Internet of Things hardware and
software architecture can be implemented for 24/7 monitoring of livestock and poultry. These data points
are then fed into the simulations to help make data-driven decisions about the health, growth and welfare
of the animals. It has even been suggested that digital twins could be used for real-time monitoring of
farmed animals’ emotions by recreating these emotions in an artificial environment, allowing for
projections of future animal behaviour (Mishra & Sharma, 2023).
Generative AI has also been proposed as a way to better manage livestock exporting. Generating and
analysing large sets of data points related to factors such as animal weights and counting of the animals
during the export process is viewed as a way to reduce tedious, error-prone and time-consuming tasks
8
RMIT Classification: Trusted
and promote supply chain efficiency. The issue of animal welfare is often raised in such accounts. It is
suggested that the health and wellbeing of farmed and exported animals can be better managed and
promoted through the use of generative AI as the latest form of precision agriculture: mostly by alerting
farmers early to any health problems or distress indicated by the data (Neethirajan, 2023b). Further, the
possibility has been raised that generative AI used in precision farming could contribute to environmental
sustainability by improving the use of resources such as water, feed, energy, pharmaceuticals and
chemicals in farm animal management, leading to outcomes like reducing methane emissions from cattle
and handling animal waste better (Neethirajan, 2024).
Potential harms from the application of generative AI–powered tools in livestock and poultry include
ethical issues. Using remote monitoring systems on animals can distance farmers from their livestock and
poultry in undesirable ways, meaning they have fewer opportunities to learn about the animals through
their physical encounters with and intimate knowledge of them. While better animal welfare is one of the
selling points of precision agriculture approaches such as those that use sensors and AI data-driven
decision-making, it has been argued that these approaches can objectify animals more than when farmers
rely on their embodied knowledge of their animals to make decisions about their wellbeing (Coghlan &
Parker, 2023; Lupton, 2023). The introduction of further generative AI–driven technologies using remote
monitoring can lead to greater ethical problems unless they are viewed as supplements to farmers’
expertise and knowledge of their animals rather than as substitutes (Neethirajan, 2023a).
4. PLANT CULTIVATION
Across both commercial agriculture and domestic gardening, a number of start-up companies have
developed applications for generative AI to assist with the care and cultivation of plants. Several studies
have explored the potential to use LLMs and computer vision technologies to diagnose plant diseases
(Demilie, 2024; Sykes et al., 2023) as part of precision agriculture initiatives (González-Rodríguez et al.,
2024). One article reports on the adoption of ‘AgriTech’ in India and the possibilities for generative AI.
Uses include crop optimisation and yield prediction, precision agriculture, disease and pest detection, and
crop breeding and genetic optimisation (Limaye, 2023). Indoor agriculture is one field of horticulture that
is deploying AI tools as part of standard growing methods. Growing crops indoors can be done in
greenhouses or hydroponic vertical farms. Companies tend to self-describe their products as organic,
pesticide-free and sustainable (Vedantam, 2023). At the same time, they are heavily tech-dependent, as
AI is used to monitor growing aspects such as light, humidity and water.
Hortiya is a German start-up that is using generative AI for indoor agriculture. Hortiya’s goal, as stated on
their website, is ‘a plant foundation model, to accurately predict plant behavior and performance. We call it
9
RMIT Classification: Trusted
Greenbrain AI’.2 The start-up is still in the development stage, and in two research pilots as well as one
greenhouse it is currently working on an AI model that is able to help growers manage their crops by
understanding how different conditions affect plant health (Smith, 2023). The company is currently
amassing data to build its foundation model, which is ‘based on the same principles’ as existing LLMs
(Smith, 2023). Data for the model will come from greenhouse sensors that measure how plants react to
changes in light, temperature and moisture. One of the main stated goals of the company is to help indoor
agriculture operators to conserve money through judicious and optimal use of resources such as light and
fertiliser.
Indoor agriculture implicitly markets itself as an application for generative AI that links human and
planetary health by improving both food security and environmental sustainability. For instance, the CEO
of Hortiya, Marc Weimer-Hablitzel, has directly linked the value-added offering of his start-up to the
mounting crisis of climate change. Indoor agriculture often results in saving water, but it is incredibly
energy-intensive because plants need artificial light to grow. Weimer-Hablitzel claims that the technology
will save growers about 5 to 10% in terms of their energy use. This is significant given the resources
frequently consumed by the industry. He notes: ‘you can’t imagine how much our tomatoes cost in terms
of energy so that we can get them in winter’ (Smith, 2023). The market opportunity for Hortiya is that
climate stress caused by more frequent droughts puts food security in peril. Indoor agriculture is seen as
an alternative because it does not require as much water as traditional agriculture. The company’s website
lists some of the impetus for its technology as the energy crisis, climate change, uncertainty in the food
chain and water scarcity.3
Resource scarcity due to climate change is also the driving factor of another initiative of generative AI for
agriculture, this time directed at farmers in the Global South. Farmer.CHAT is an initiative of the
development organisation Digital Green in partnership with Gooey.AI, a generative AI start-up. It is
specifically meant for agricultural producers ‘on the frontlines of climate change and water security’.4 The
initiative was designed in collaboration with the Indian and Ethiopian governments and developed to
provide information to farmers on how to ‘optimize crop management, reduce waste, and increase yields’.5
Farmer.CHAT is a GPT-4 based AI platform trained on best practice videos, agricultural call centre
transcripts and other ‘vetted data sources’. It is available in Hindi, Telugu, Bhojpuri, English, Amharic and
Swahili. Farmers can interact with the AI via WhatsApp or type questions directly into the Farmer. CHAT
2 https://hortiya.com/
3 https://hortiya.com/
4 https://www.help.gooey.ai/farmerchat
5 https://www.help.gooey.ai/farmerchat
10
RMIT Classification: Trusted
website. Ama KrushAI, a chatbot that provides agronomic advice to farmers, is a similar initiative run by
the state of Odisha’s Department for Agriculture and Farmer’s Empowerment in India.6
Another area where generative AI is being applied to plant cultivation is in the development of apps to
assist domestic gardeners. These have a similar purpose to generative AI initiatives aimed at farmers in
that they provide horticultural advice. One such app is the ChatBotanist, developed by the Royal
Horticultural Society in partnership with technology company Publicis Sapient.7 ChatBotanist uses
generative AI to answer plant-related queries from gardeners. A similar app called PLNT is designed to
help users care for plants with information on ideal light and watering conditions, as well as help to identify
plants and diagnose and treat plant health problems (Willey, 2023).
Some scholarship has investigated the usefulness of generative AI for basic biological research. One
research group used ChatGPT to generate a list of important questions for plant science (Agathokleous et
al., 2023). Studies testing the accuracy of LLM tools in plant biology applications have produced varying
results. Another study investigated the quality of responses to ChatGPT queries about general biological
process in living things without specifying which kingdom (i.e. animal, plant or fungi). It found that ‘plant
awareness’ in the AI answers was highly variable, demonstrating the importance of having expert
scientists to contribute to validating the data and methods used to train LLMs. A bias was evident in
responses that focused on animal biology without acknowledging that the same processes occurred in
plants (Geitmann & Bidhendi, 2023).
5. BIODIVERSITY PRESERVATION AND NATURE
CONSERVATION
Some new engagements by technology companies using generative AI are targeted at nature
conservation efforts and biodiversity preservation. Many applications in conservation take advantage of
the ability of generative AI to analyse large datasets and combine diverse sources of data or, alternatively,
to create synthetic data and simulation models.
IBM has described one such initiative in a blog post. It is partnering with the Reef Company, which works
on coral reef restoration by building engineered reefs. These reefs include sensors and cameras to collect
data such as salinity, temperature, pH and carbon dioxide. This information is then uploaded to the cloud.
IBM is offering support through its AI and data analytics. Using its watsonx generative AI and foundation
6 https://www.amakrushi.in/en/
7 https://www.publicissapient.com/news/rhs-partners-with-ps-to-launch-generative-ai-powered-chatbotanist
11
RMIT Classification: Trusted
models, IBM can combine various data sources (such as those from sensors with satellite imagery and
historical records) and vast amounts of data. The company claims that this use of generative AI will aid
conservation efforts by, for example, predicting potential threats to coral reefs in the analysis of patterns
in the data (Martinho & van de Waal, 2023).
The possibility for generative AI to create synthetic data is being explored as part of the creation of digital
twins for biodiversity research (Podder et al., 2023). For example, the BioDT project is developing
prototype digital twins to model aspects such as species responses to environmental change and threats
to species of concern.8 This modelling is meant to be used for biodiversity restoration. BioDT features
eight use cases related to land ecosystems, clustered in four groups: i) species response to environmental
change; ii) genetically detected biodiversity; iii) dynamics and threats from and for species of policy
concern; and iv) species interaction with each other and with humans.
Generative AI is further applied by researchers to the protection and management of forests, including the
use of simulations such as digital twins. As highly diverse ecosystems, forests are complex to monitor and
manage, and many are under major pressure from climate change and human encroachment due to land
clearing. Digital twins of forests have been developed that include trees as core forest elements together
with data points from the environments in which trees grow, as a form of data-driven prediction and
decision-making. Data from a variety of sources are combined in these models using LLMs to test the
efficacy of various alternative management systems through simulations (Döllner et al., 2023).
Researchers studying biodiversity have also used generative AI to learn about species coexistence (Hirn
et al., 2022). Coexistence patterns are complex because they do not simply entail direct interactions
between two species but also indirect interactions between multiple species. Generative AI can be used
for this research because it can handle complex data.
The Hawaiian crow project is a further example of how AI research can be deployed for animal
conservation efforts. Project CETI (Cetacean Translation Initiative) also claims to potentially contribute to
conservation through the greater awareness of the intelligence of sperm whales that their translation
research might produce. The 1970s album ‘Songs of the Humpback Whale’, which is thought to have
contributed to the ‘save the whales’ movement, is cited as an example. Members of the CETI research
group suggest that decoding animal communication using AI and machine learning could also lead to
greater societal impetus for environmental protections (Anthes, 2022; Kolbert, 2023).
8 https://biodt.eu/
12
RMIT Classification: Trusted
In addition to its potential for conservation, researchers suggest that AI-powered translation projects
could also help to improve the welfare of animals in captivity. By identifying ‘animal signals that are
associated with stress, discomfort, pain, and evasion, or with positive states, such as arousal or
playfulness’, the lives of farm animals and other animals in captivity could potentially be improved.
Furthermore, these tools could be used for conservation by ‘the assaying of wild populations to measure
the impact of anthropogenic stressors’ (Rutz et al., 2023, p. 154).
Another initiative is being developed by Meta AI, which has created a generative AI model that can predict
protein structures. Proteins are found throughout living creatures and underpin diverse biological
functions such as the production of antibodies, photosynthesis and vision. Predicting and discovering
proteins has the potential to benefit human and planetary health. Meta AI claims that it could ‘help cure
diseases, clean up the environment, and produce cleaner energy’.9 Researchers trained an LLM to make
predictions of protein structures 60 times faster than current technology. Discovering new proteins is one
side of the research, but Meta AI is also working on designing proteins which it claims will assist in ‘solving
challenges to health, disease, and the environment’.10
Wildlife researchers are using generative AI to analyse images of animals. AI programs can identify
differences and similarities between animal morphology and physical traits that human observers cannot
detect. Due to the rigours and expense of collecting data in the field, there is currently a deficiency of data
about 20,054 species on the International Union for Conservation of Nature’s Red List of Threatened
Species. Because of this lack of data, conservation researchers are unable to accurately know which
species are threatened or at what rate they are moving towards extinction (Wong, 2022). Generative AI
programs are being used for initiatives such as scanning thousands of images of whale sharks to assist
researchers to identify individuals, track their movements and estimate population size, which enables
conservation scientists to make more accurate assessments of the endangered status of the species as a
whole (Wong, 2022).
While generative AI technologies offer benefits such as these to the field of nature conservation, there is
some concern that they may have negative impacts. One area where generative AI could potentially be
misused is in wildlife photography, where AI-generated images are almost indistinguishable from those
taken by photographers in the field. Because conservation activism uses photography as a powerful
argumentative mode for the importance of species protection, the possibility that images could be AI-
9 https://ai.meta.com/blog/protein-folding-esmfold-metagenomics/
10 https://ai.meta.com/blog/protein-folding-esmfold-metagenomics/
13
RMIT Classification: Trusted
generated has raised concerns about their potential to be used as misinformation or that they might make
it more difficult to trust images taken from nature (Murabayashi, 2023).
6. CLIMATE CHANGE SCIENCE AND COMMUNICATION
Generative AI and LLMs play a role in both communicating about and researching climate change. As in
many domains, there is strong potential for the role these tools can play in supporting research into
climate science. For instance, generative AI could help researchers to review the available literature and
bring different sources of information into the frame, such as documents and images, or texts that fall
outside traditional scientific journals, including policy documents and working papers (Larosa et al., 2023).
In one commentary, the authors suggest that AI might offer an expansive view of the possibilities for
climate change solutions and ‘a more holistic understanding of the landscape in which decisions are taken’
(Larosa et al., 2023, p. 498).
Along these lines, a research team from New Zealand provided three case studies of situations where
generative AI could potentially help with what they described as ‘nature-based solutions’ for the
integration and support of ecosystems (Richards et al., 2024). The first case study addressed the
possibilities of using generative AI to report scientific information on ecosystem services, future land use
options and nature-based solutions for farms in accessible ways for a public audience. The second case
study suggested that the software could be deployed to help answer questions from homeowners about
garden design to support biodiversity; for example, with the use of an interactive chatbot system. The
third study involved applying generative AI to image generation to create visualisations showing potential
future scenarios of landscape change using different solutions. These case studies all focus on how
scientific information about ecosystems and the environment can be processed by generative AI tools to
help decision-making by the public and other community stakeholders.
Researchers have further argued that AI systems can help with climate predictions, as they can manage
the complexity of climate information and the vast amount of constantly changing data (Debnath et al.,
2023). Generative AI is being introduced as a technology for weather forecasting, in particular for extreme
weather. A generative AI model called NowcastNet can predict extreme rain earlier than the software that
is currently used. However, there are also concerns about the potentially negative effects of AI as a
source of climate information. One is potential bias, a problem with generative AI and AI more generally. In
the case of climate information, an AI model might incorporate bias by being trained on data only from
certain regions or time periods, thereby compromising its capacity for climate prediction (Debnath et al.,
2023). Furthermore, the utility of LLMs in weather forecasting is itself potentially affected by climate
change, as it is unclear how these models will respond to drastic alterations in the environment, given that
they are trained on data from previous weather and climate conditions (Heikkilä, 2023).
14
RMIT Classification: Trusted
Another bias problem that has been identified is the sources on which LLMs are trained and draw from
when answering queries. An analysis of 30,000 responses from ChatGPT on ecological restoration
expertise, stakeholder engagements and techniques found that two-thirds of the responses relied on the
expertise of male academics working at US universities, while largely ignoring evidence from researchers
and other experts in less wealthy countries, including First Nations knowledge holders. The authors argue
that ChatGPT is therefore reinforcing Western science perspectives, overlooking diverse sources of
expertise and contributing to global conservation injustices (Urzedo et al., 2024). Other studies have
suggested the possibility that AI models could inaccurately represent the emissions from certain
industries, thereby supporting vested interests rather than disinterested climate science (Debnath et al.,
2023).
Because LLM chatbots such as ChatGPT are readily accessible by the general public, these models are of
interest to researchers and commentators in terms of how effectively they can be used as reliable
information sources, as well as to what extent they are at risk of spreading misinformation. As in other
areas of education, this general accessibility means that ChatGPT might be useful for environmental
education (Chang & Kidman, 2023). One study evaluated climate information and generative AI through
the lens of public communication of science. The researchers concluded that LLM chatbots were fluent
and articulate in terms of communication style, but they were problematic regarding the content of climate
information and prone to making generic statements that were not very useful (Bulian et al., 2023).
Generative AI could also be used to spread climate misinformation. A report by Stockholm University
suggests several ways this could be accomplished: by bots generating tweets in the style of climate
deniers, by the creation of deep-fake images of high-profile figures such as climate activist Greta
Thunberg, and by producing long-form blog posts or pieces of writing espousing climate misinformation.
The report also speculates that generative AI tools could aid in misinformation campaigns by virtue of their
labour-saving possibilities through the replacement of human writers or by creating varied content that
allows small groups to appear online as though they have many different members (Galaz et al., 2023).
The role AI tools can play in climate information is therefore contradictory. They are potentially useful in
terms of accessibility and potentially damaging in terms of the veracity of information that is disseminated.
7. ENVIRONMENTAL COSTS OF GENERATIVE AI
As outlined in this report, generative AI can be used productively for various sustainability initiatives.
However, these technologies generate their own significant environmental costs in the form of carbon
emissions, electricity consumption, use of rare minerals and excessive water use related to the hardware
15
RMIT Classification: Trusted
and infrastructure requirements of data training, processing and archiving, and the e-waste generated by
the hardware demands of data-driven AI.
Numerous studies have estimated the energy consumed by the manufacture and training of a generative
AI model. For example, researchers calculated that the development of GPT-3 ‘consumed 1,287 megawatt
hours of electricity and generated 552 tons of carbon dioxide equivalent, the equivalent of 123 gasoline-
powered passenger vehicles driven for one year’ (Saenko, 2023). Training the models is the most energy-
intensive aspect of generative AI. Researchers estimate that training a model such as GPT-4 generates
approximately 300 tonnes of carbon (Kumar & Davenport, 2023), comparable to the amount of carbon
emitted by 125 round-trip flights between New York and Beijing (Deeb & Garel-Frantzen, 2023). As the
technology advances, the carbon cost is likely to increase, as the greater complexity of the models and
amount of data they are trained on will require even more energy (An et al., 2023).
Added to the carbon emissions are other environmental costs associated with parts manufacture and
physically running models, such as mining of rare minerals for graphics-processing units and use of water
to cool large data centres (Luccioni, 2023). Data centres consume massive amounts of energy as well as
water to run air conditioners. Training the LLM LaMDA used an estimated million litres of water (Dolby,
2023). Some jurisdictions, including the US state of Virginia, are developing new fossil fuel plants or
retaining old coal plants due to be retired simply to service these resources needs (Fanger, 2024).
Generative AI also uses more hardware than other digital technologies which has to be replaced more
frequently. This means not just greater manufacturing demands but also the production of more e-waste
(Pratt, 2023).
On the user side, it is estimated that a generative AI query is considerably higher in terms of carbon
emissions (four to five times) than a Google Search or other search engine query (Saenko, 2023). While a
single query uses less energy than the process of training the model, queries are typically performed many
times over and may therefore constitute up to 90% of the energy consumption of generative AI (Kumar &
Davenport, 2023). Using generative AI also consumes water: for example, ‘Chat-GPT needs to “drink” a
500-milliliter bottle of water for a basic conversation of 20 to 50 inquiries, depending on where the
electricity is generated’ (Dolby, 2023). Further, there are place-specific circumstances that affect the
energy and water use of LLMs. Some areas may use renewable energy or freshwater alternatives such as
wastewater, and these factors shape the environmental cost of LLMs. Microsoft has said that its Asian
data centres are far less water-efficient than its centres based in the Americas (Dolby, 2023). Additionally,
there are seasonal shifts in water use because hot summers require much more water to cool data centres
due to higher evaporation rates (Dolby, 2023).
Activists and commentators have argued that one of the reasons that the environmental toll of generative
AI is difficult for users to perceive is because AI itself appears ‘ephemeral’, rather than something with a
16
RMIT Classification: Trusted
physical, material existence (Stokel-Walker, 2023). Estimating the costs of specific models is also
compounded by a lack of transparency from companies making the technology (Pratt, 2023). For industry
actors considering adopting generative AI, questions of sustainability have been overshadowed by other
concerns regarding the technology such as its tendency to generate wrong answers and the in-built
structures of bias that feature in models as a result of their training (Pratt, 2023). Researcher Alex de
Vries, who has published on the energy use of AI, suggested in an interview that people should be
‘mindful’ of their AI use, as running large LLMs is a ‘potential big waste of power’ (Calma, 2023).
8. CONCLUSION
The paradox of generative AI is that the technology is hyped for its promise in contributing to
sustainability and environmental health, yet at the same time its manufacture, training and use all entail
significant costs to the environment. The vested interests that may underlie major commercial initiatives to
apply generative AI to ecosystems should be closely examined. The extent to which promissory narratives
put forward in technology developers’ promotional materials may be yet further examples of industry
‘greenwashing’ (de Freitas Netto et al., 2020) in an attempt to make claims about sustainable practices, or
are over-hyped to attract investors, require further analysis and critical attention.
As the language of ‘nature-based solutions’ implies, many developers and promoters of current or
predicted generative AI technologies tend to objectify ‘nature’ as a resource to be extracted and
manipulated for the use of humans to find ways of redressing the harms they have caused to the natural
environment. As such, these discourses, imaginaries and practices reproduce a worldview that positions
elements of the non-human world as separate from, and manipulable by, humans for their own interests
(Lockhart et al, 2023; Lupton, 2023; Turnbull et al., 2022). This is an approach that completely ignores the
ways that people are always inextricably embedded within ecosystems and disavows the multispecies
relationships that are essential to human and non-human flourishing.
The ethical implications of applying this novel software to digitise and datafy living creatures and
ecosystems and the potential intended or unintended consequences and impacts of such activities have
yet to be fully confronted. There is a small but growing literature that has raised these issues in relation to
previous applications of data-driven systems for the care, monitoring and control of animals, noting how
such technologies frequently objectify creatures (for example, Bossert & Hagendorf, 2021; Coghlan &
Parker, 2023; Lupton, 2023; von Essen et al., 2023). So too, a move towards recognising more-than-
human rights is gathering pace in legal circles, drawing on First Nations’ understandings of the vitalities of
other-than-human entities, including not only other living creatures but also elements of the landscape
such as geological features, waterways and the atmosphere ((Rodríguez-Garavito, 2024). As these
generative AI and LLM tools continue to develop, detailed examination of the assumptions underpinning
17
RMIT Classification: Trusted
their design and applications when used in relation to ecosystems is crucial to limit these human
tendencies to exploitative, extractivist, and potentially deeply harmful approaches to other species and
the natural world.
Finally, there is a range of questions still to be explored that focus on how these technologies are
accepted and used by members of the public as well as by experts in the fields of environmental
sustainability, conservation and science communication. The utility of generative AI and LLMs will vary
according to the sociodemographics, cultures, geographical and ecological contexts, biographical
experiences and literacies of the different social groups that are targeted by developers and promoters of
these tools. Unless these situated experiences and differences are acknowledged in future research, the
possibilities, risks, accessibility and impacts of generative AI when applied to ecosystems and the natural
environment will not be fully recognised or addressed.
9. REFERENCES
Agathokleous, E., Rillig, M. C., Peñuelas, J., & Yu, Z. (2023). One hundred important questions facing plant
science derived using a large language model.
Trends in Plant Science
.
https://doi.org/10.1016/j.tplants.2023.06.008
An, J., Ding, W., & Lin, C. (2023). ChatGPT: Tackle the growing carbon footprint of generative AI.
Nature
,
615
, 586. https://doi.org/10.1038/d41586-023-00843-2
Anthes, E. (2022, August 30). The animal translators.
The New York Times
.
https://www.nytimes.com/2022/08/30/science/translators-animals-naked-mole-rats.html
Bossert, L., & Hagendorff, T. (2021). Animals and AI. The role of animals in AI research and application – An
overview and ethical evaluation.
Technology in Society
,
67
.
https://www.sciencedirect.com/science/article/pii/S0160791X21001536
Bromley, C. (2023, August 29). How to use AI to talk to whales – and save life on Earth.
Wired
.
https://www.wired.com/story/use-ai-talk-to-whales-save-life-on-earth/
Bulian, J., Schäfer, M. S., Amini, A., Lam, H., Ciaramita, M., Gaiarin, B., Huebscher, M. C., Buck, C., Mede,
N., Leippold, M., & Strauss, N. (2023). Assessing large language models on climate information.
arXiv
,
2310.02932v1
. https://doi.org/10.48550/arXiv.2310.02932
Calma, J. (2023, October 11). The environmental impact of the AI revolution is starting to come into focus.
The Verge
. https://www.theverge.com/2023/10/10/23911059/ai-climate-impact-google-openai-chatgpt-
energy
18
RMIT Classification: Trusted
Chang, C.-H., & Kidman, G. (2023). The rise of generative artificial intelligence (AI) language models –
challenges and opportunities for geographical and environmental education.
International Research in
Geographical and Environmental Education
,
32
(2), 85–89.
https://doi.org/10.1080/10382046.2023.2194036
Coghlan, S., & Parker, C. (2023). Harm to nonhuman animals from AI: A systematic account and
framework.
Philosophy & Technology
,
36
(2). https://doi.org/10.1007/s13347-023-00627-6
Debnath, R., Creutzig, F., Sovacool, B. K., & Shuckburgh, E. (2023). Harnessing human and machine
intelligence for planetary-level climate action.
NPJ Climate Action
,
2
. https://doi.org/10.1038/s44168-023-
00056-3
Deeb, D. J., & Garel-Frantzen, A. (2023). The generative AI revolution: Key impacts to the environmental
and energy sectors.
The National Law Review
. https://www.natlawreview.com/article/generative-ai-
revolution-key-impacts-to-environmental-energy-sectors
de Freitas Netto, S. V., Sobral, M. F. F., Ribeiro, A. R. B., & Soares, G. R. d. L. (2020). Concepts and forms of
greenwashing: A systematic review.
Environmental Sciences Europe
,
32
(1), 1–12.
https://doi.org/10.1186/s12302-020-0300-3
Demilie, W. B. (2024). Plant disease detection and classification techniques: A comparative study of the
performances.
Journal of Big Data
,
11
(1). https://link.springer.com/article/10.1186/s40537-023-00863-9
Dolby, N. (2023, September 11). Artificial intelligence can make companies greener, but it also guzzles
energy.
Wall Street Journal
. https://www.wsj.com/articles/artificial-intelligence-can-make-companies-
greener-but-it-also-guzzles-energy-7c7b678
Döllner, J., de Amicis, R., Burmeister, J.-M., & Richter, R. (2023).
Forests in the digital age: Concepts and
technologies for designing and deploying forest digital twins
. Proceedings of the 28th International ACM
Conference on 3D Web Technology, San Sebastian, Spain. https://doi.org/10.1145/3611314.3616067
Fanger, E. (2024, February 22). The dirty energy fueling Amazon’s data gold rush.
The Nation
.
https://www.thenation.com/article/environment/data-centers-virginia-amazon-environment/
Fox, B., Goggin, G., Lupton, D., Regenbrecht, H., Scuffham, P., & Vucetic, B. (2020).
The Internet of Things
.
Australian Council of Learned Academies. https://acola.org/hs5-internet-of-things-australia/
Galaz, V., Metzler, H., Daume, S., Olsson, A., Lindström, B., & Marklund, A. (2023).
AI could create a perfect
storm of climate misinformation
.
19
RMIT Classification: Trusted
https://www.stockholmresilience.org/download/18.889aab4188bda3f44912a32/1687863825612/SRC_Cli
mate%20misinformation%20brief_A4_.pdf
Geitmann, A., & Bidhendi, A. J. (2023). Plant blindness and diversity in AI language models.
Trends in Plant
Science
,
28
(10), 1095–1097. https://doi.org/10.1016/j.tplants.2023.06.016
González-Rodríguez, V. E., Izquierdo-Bueno, I., Cantoral, J. M., Carbú, M., & Garrido, C. (2024). Artificial
intelligence: A promising tool for application in phytopathology.
Horticulturae
,
10
(3).
https://www.mdpi.com/2311-7524/10/3/197
Hayes, B. J., Chen, C., Powell, O., Dinglasan, E., Villiers, K., Kemper, K. E., & Hickey, L. T. (2023). Advancing
artificial intelligence to help feed the world.
Nature Biotechnology
,
41
, 1188–1189.
https://doi.org/10.1038/s41587-023-01898-2
Heikkilä, M. (2023, July 5). New AI systems could speed up our ability to create weather forecasts.
MIT
Technology Review
. https://www.technologyreview.com/2023/07/05/1075897/new-ai-systems-could-
speed-up-our-ability-to-create-weather-forecasts/
Hirn, J., García, J. E., Montesinos-Navarro, A., Sánchez-Martin, R., Sanz, V., & Verdú, M. (2022). A deep
generative artificial intelligence system to predict species coexistence patterns.
Methods in Ecology and
Evolution
,
13
, 1052–1061. https://doi.org/10.1111/2041-210X.13827
Kocoń, J., Cichecki, I., Kaszyca, O., Kochanek, M., Szydło, D., Baran, J., Bielaniewicz, J., Gruza, M., Janz,
A., & Kanclerz, K. (2023). ChatGPT: Jack of all trades, master of none.
Information Fusion
.
https://www.sciencedirect.com/science/article/pii/S156625352300177X
Kolbert, E. (2023, September 4). Talk to me.
The New Yorker
.
https://www.newyorker.com/magazine/2023/09/11/can-we-talk-to-whales
Kumar, A., & Davenport, T. (2023, July 20). How to make generative AI greener.
Harvard Business Review
.
https://hbr.org/2023/07/how-to-make-generative-ai-greener
Larosa, F., Hoyas, S., García-Martinínez, J., Conejero, J. A., Nerini, F. F., & Vinuesa, R. (2023). Halting
generative AI advancements may slow down progress in climate research.
Nature Climate Change
,
13
,
497–499. https://doi.org/10.1038/s41558-023-01686-5
Limaye, P. (2023, July 12). Breaking new ground: the role of generative AI in agritech.
Express Computer
.
https://www.expresscomputer.in/artificial-intelligence-ai/breaking-new-ground-the-role-of-generative-ai-
in-agritech/100930/
20
RMIT Classification: Trusted
Lockhart, A., Marvin, S., & While, A. (2023). Towards new ecologies of automation: Robotics and the re-
engineering of nature.
Geoforum
,
145
.
https://www.sciencedirect.com/science/article/pii/S0016718523001513
Lockie, S., Fairley-Grenot, K., Ankeny, R., Botterill, L., Howlett, B., McBratney, A., Probyn, E., Sorrell, T.,
Sukkarieh, S., & Woodhead, I. (2020).
The future of agricultural technologies
. Australian Council of Learned
Academies. https://acola.org/hs6-future-agricultural-technologies/
Luccioni, S. (2023, April 12). The mounting human and environmental costs of generative AI.
Ars Technica
.
https://arstechnica.com/gadgets/2023/04/generative-ai-is-cool-but-lets-not-forget-its-human-and-
environmental-costs/
Lupton, D. (2023).
The Internet of Animals: Human–animal relationships in the digital age
. Polity Press.
Martinho, R., & van de Waal, J. (2023, June 8). Celebrating World Oceans Day: Revitilizing the marine
ecosystem with technology-driven engineered reefs to accelerate CO2 capture.
https://www.ibm.com/blog/celebrating-world-oceans-day-revitalizing-the-marine-ecosystem-with-
technology-driven-engineered-reefs/
McGovern, A., Tissot, P., & Bostrom, A. (2024). Developing trustworthy AI for weather and climate.
Physics
Today
,
77
(1), 26–31. https://doi.org/10.1063/PT.3.5379
Mishra, S., & Sharma, S. K. (2023). Advanced contribution of IoT in agricultural production for the
development of smart livestock environments.
Internet of Things
,
22
.
https://doi.org/10.1016/j.iot.2023.100724
Murabayashi, A. (2023). What does generative AI mean for bird and nature photography?
Audubon
Magazine
. https://www.audubon.org/news/what-does-generative-ai-mean-bird-and-nature-photography
Neethirajan, S. (2023a). Artificial intelligence and sensor innovations: Enhancing livestock welfare with a
human-centric approach.
Human-Centric Intelligent Systems
,
online first
. https://doi.org/10.1007/s44230-
023-00050-2
Neethirajan, S. (2023b). Artificial intelligence and sensor technologies in dairy livestock export: Charting a
digital transformation.
Sensors
,
23
(16). https://www.mdpi.com/1424-8220/23/16/7045
Neethirajan, S. (2024). Net zero dairy farming – advancing climate goals with big data and artificial
intelligence.
Climate
,
12
(2). https://www.mdpi.com/2225-1154/12/2/15
21
RMIT Classification: Trusted
Parshley, L. (2023, October). Artificial intelligence could finally let us talk with animals.
Scientific American
.
https://www.scientificamerican.com/article/artificial-intelligence-could-finally-let-us-talk-with-animals/
Podder, S., Singh, S. K., Robinson, M., & Kuntagod, N. (2023). Biodiversity needs AI: Infusing intelligence
into biodiversity preservation and restoration.
California Management Review
.
https://cmr.berkeley.edu/2023/09/biodiversity-needs-ai-infusing-intelligence-into-biodiversity-
preservation-and-restoration
Pratt, M. K. (2023, October 20). Generative AI’s sustainability problems explained.
TechTarget
.
https://www.techtarget.com/sustainability/feature/Generative-AIs-sustainability-problems-
explained#:~:text=Generative%20AI%20has%20been%20linked,in%20other%20types%20of%20computing
Richards, D., Worden, D., Song, X. P., & Lavorel, S. (2024). Harnessing generative artificial intelligence to
support nature-based solutions.
People and Nature
,
6
(2), 882-893. https://doi.org/10.1002/pan3.10622
Rodríguez-Garavito, C. (Ed.). (2024).
More than Human Rights: An Ecology of Law, Thought and Narrative
for Earthly Flourishing
. New York University.
Rutz, C., Bronstein, M., Raskin, A., Vernes, S. C., Zacarian, K., & Blasi, D. (2023). Using machine learning to
decode animal communication.
Science
,
381
(6654), 152–155. https://doi.org/10.1126/science.adg7314
Saenko, K. (2023, May 23). Is generative AI bad for the environment? A computer scientist explains the
carbon footprint of ChatGPT and its cousins.
The Conversation
. https://theconversation.com/is-
generative-ai-bad-for-the-environment-a-computer-scientist-explains-the-carbon-footprint-of-chatgpt-
and-its-cousins-204096
Sætra, H. S. (2023). Generative AI: here to stay, but for good?
Technology in Society
,
75
.
https://www.sciencedirect.com/science/article/pii/S0160791X2300177X
Sallam, M. (2023). ChatGPT utility in healthcare education, research, and practice: Systematic review on
the promising perspectives and valid concerns.
Healthcare
,
11
(6).
https://doi.org/10.3390/healthcare11060887
Smith, T. (2023, April 19). PlantGPT – the company building a generative model to speak the language of
plants.
Sifted
. https://sifted.eu/articles/plantgpt-chat-gpt
Steeneken, P. G., Kaiser, E., Verbiest, G. J., & ten Veldhuis, M.-C. (2023). Sensors in agriculture: towards
an Internet of Plants.
Nature Reviews Methods Primers
,
3
(1), 60. https://doi.org/10.1038/s43586-023-
00250-x
22
RMIT Classification: Trusted
Stokel-Walker, C. (2023, August 1). TechScape: Turns out there’s another problem with AI – its
environmental toll.
The Guardian
. https://www.theguardian.com/technology/2023/aug/01/techscape-
environment-cost-ai-artificial-intelligence?CMP=share_btn_url
Sykes, J. R., Denby, K. J., & Franks, D. W. (2023). Computer vision for plant pathology: A review with
examples from cocoa agriculture.
Applications in Plant Sciences
. https://doi.org/10.1002/aps3.11559
Turnbull, J., Searle, A., Hartman Davies, O., Dodsworth, J., Chasseray-Peraldi, P., von Essen, E., &
Anderson-Elliott, H. (2022). Digital ecologies: materialities, encounters, governance.
Progress in
Environmental Geography
,
2
(1-2), 3-32. https://doi.org/10.1177/27539687221145698
Urzedo, D., Sworna, Z. T., Hoskins, A. J., & Robinson, C. J. (2024). AI chatbots contribute to global
conservation injustices.
Humanities and Social Sciences Communications
,
11
(1).
https://doi.org/10.1057/s41599-024-02720-3
Vedantam, K. (2023, March 17). Here are 3 big areas where AI is cropping up in agtech.
Crunchbase News
.
https://news.crunchbase.com/ai-robotics/agtech-indoor-farming-openai-plenty/
von Essen, E., Turnbull, J., Searle, A., Jørgensen, F. A., Hofmeester, T. R., & van der Wal, R. (2023). Wildlife
in the digital anthropocene: examining human-animal relations through surveillance technologies.
Environment and Planning E: Nature and Space
,
6
(1), 679-699.
https://doi.org/10.1177/25148486211061704
Willey, D. (2023, March 14). Been trying ChatGPT out? PLNT app utilizing generative AI is now helping
people take care of their plants.
Yahoo! Finance
. https://finance.yahoo.com/news/trying-chatgpt-plnt-
app-utilizing-131500376.html?soc_src=social-sh&soc_trk=tw&tsrc=twtr
Wong, K. (2022, April 20). Wildlife photos are a new treasure trove for AI-driven conservation research.
The Verge
. https://www.theverge.com/23032327/wildlife-photos-ai-conservation-research
