Sustainable AI and ChatGPT

Siddhi Vij1, Anupriya Raj2

1
Amity School of Engineering and Technology, Amity University, Noida, India
2
Amity School of Engineering and Technology, Amity University, Noida, India

Corresponding Author: siddhivij3@gmail.com

Abstract

Deep learning-capable robots and machine intelligence have a profoundly

disruptive and enabling influence on society, industry, and governments.
Additionally, they are impacting the broader global environmental trends. With
the introduction of the concept of sustainable AI and the growing concerns
regarding the environmental impact of large language models, such as ChatGPT,
it is integral to understand the two aspects of sustainable AI. While there is a
growing effort towards AI for sustainability, there is also significance in
understanding the sustainability of using and developing AI on an everyday basis.
This paper explores in depth what constitutes sustainable AI and the importance
of understanding the environmental implications posed by the exponential growth
of AI and large language models, such as ChatGPT. In this paper, we also discuss
the concept of AI ethics, which play a key role in understanding the recent
advancements, as a structural approach for uncovering ethical issues on a broader
scale. AI systems have a significant environmental effect since they require a lot
of data, processing power, and energy-intensive infrastructure. The training and
operation of AI models require a large amount of processing power, often derived
from fossil fuel energy sources. Data centers and cloud computing facilities,
which enable AI technology, dramatically raise worldwide energy consumption
and carbon emissions. Significant carbon emissions are also produced during the
processing, transport, and storage of massive volumes of data. This dramatic
increase in energy demand leads to global shortages in resource allocation. Due to
this, it is important to focus on the often overlooked concepts regarding AI, i.e.,
sustainability, privacy, digital divide, and ethics. This paper aims to inspire the
readers to remember that there are environmental costs to AI.

Keywords: Artificial Intelligence (AI), sustainable AI, AI models, environmental

effects, AI ethics

1.Introduction
Artificial intelligence (AI) is one of the fastest-growing research and commercial
areas. AI's extensive investment has also led to its new applications in fields such
as science, health, finance, and education. In recent years, the speed at which
artificial intelligence has evolved means it can more accurately translate and
emulate natural language.. This has been made possible by the implementation of
more advanced deep learning models, such as recurrent neural networks (RNNs)
and training systems on vast volumes of text.
BERT and GPT-3 from Google and OpenAI are two examples of these advanced
systems that exhibit exceptional precision in comprehending text and producing
it. ChatGPT is a prototype chatbot built on the Generative Pre-trained
Transformer architecture that offers recommendations based on OpenAI's
language model to users and can interact with the AI. [57]
Artificial intelligence (AI) is truly changing the world, and will change in the
future; however, artificial intelligence may have a detrimental effect on society,
such as the widely-known ChatGPT large-scale language model created by Open
AI. Statista.com [75] reports that the global AI industry is valued at US$ 2,00,000
million in 2023 and is projected to increase by about nine times to reach
US£1,85,000 by 2030 alone. Due to its exceptional Natural Language Processing
(NLP) features, ChatGPT has gained significant attention and without a doubt
made life easier for everyone. But like everything else valuable in this field, it
comes with repercussions: Data and estimates reveal that ChatPG generates 8.4
metric tonnes of carbon dioxide per year to power the data centres, more than
double the amount produced by each person on an annual basis. Additionally, the
training process consumes a significant amount of water and converts it into heat
to maintain equipment temperatures and cool it down, incessantly demanding
freshwater storage . This is particularly problematic. [11]
AI ethics have been delved into due to the recent growth of AI, which was made
possible by an increase in data and computing resources. AI ethics refer to
concerns about ethics and social issues that impact individuals, producers,
developers, consumers, legislators, and civic society.
The initial AI ethics centered on possible applications of AI, such as
superintelligence, and reached the level of "ethical code", and the morality of
hypothetical robot uprisings [12].
The ethical concerns raised by AI in the second generation were practical,
including machine learning techniques including the explainability problem and
black-box algorithm [25, 56], insufficient data representation in training data
[8,13], and biases in AI. The rise of facial and emotion recognition technologies,
which pose privacy concerns, is being countered by models [10].
Now we have the third wave of AI ethics, which is actively working to educate
the public, policymakers, academia and even some AI developers as to how these
AIs are causing an enormous amount of harm globally while taking action on this
present environmental crisis. ".
The focus of this third wave should be on sustainable development. Although
there is an increasing push to make AI more suitable for positive applications
(AI4Good), the ongoing process of maintaining this trend is still being pursued.
To address specific development goals, it is necessary to consider the viability of
AI system development and implementation.
One way to train a single, deep learning model for natural language processing
(NLP), as demonstrated by Strubell et al. Roughly 600,000 pounds of carbon
dioxide can be released from a typical (GPU) annually. [62]
When compared to average consumption, five cars produce roughly the same
amount of carbon dioxide emissions during their lifetime. According to other
research, over 40 days of research training, "Google's AlphaGo Zero created 96
tonnes of CO2 which is equivalent to 23 American homes' carbon impact or 1000
hours of air travel’. [5, 67]. In an era when the world must commit to reducing
carbon emissions, is it truly worth the additional cost of algorithms that can play
games or perform other mundane tasks?
Moreover, AI is not limited to its application in industry and healthcare. It is
expected to be as ubiquitous as cell phones or the internet. We must not overlook
the environmental impact of technology like this. This is a crucial development
with high environmental expenses.
Due to these reasons, it is crucial to focus on AI that is sustainable.
The subsequent paper employs the term sustainable AI as a generalization to
describe two branches with distinct objectives and methods: AI for sustainability
and sustainability of AI. Despite the immense potential of AI as a sustainable
technology, it falls short in taking into account the environmental consequences of
its creation. The environmental impact of AI training and tuning is a crucial
aspect of sustainability, as it involves measuring and reducing the carbon footprint
through measurement. [67] Addressing these issues demonstrates that AI can be
considered environmentally sustainable.

2. Sustainable AI
AI can be defined as “the ability of a system to interpret external data correctly, to
learn from such data, and to use those learnings to achieve specific goals and
tasks through flexible adaptation”. [33] The field of study, known as sustainable
AI, investigates AI technology, including hardware, training methods, and data
processing, to tackle issues related to sustainable development and/or AI
sustainability. Furthermore, sustainable AI must address all stages of its life cycle,
including design (initiation and completion), training (training), development
(validation), re-tuning) implementation, and usage.
The well-known non-profit group "AI4Good" is investigating how artificial
intelligence (AI) can be utilized to achieve sustainability in some way. By
utilizing AI and machine learning (ML), the Sustainable Development Goals
(SDGs) of the UN can be achieved. By utilizing AI/ML, we aim to increase the
accessibility of affordable and clean energy, such as SDG 7, to more people
worldwide. With over 600 million people without access to modern power, this is
undoubtedly a worthwhile target to pursue.

2.1. AI for Sustainability

Artificial Intelligence (AI) is helping to make the world cleaner and greener by
resolving environmental problems and using resources wisely. AI can be
researched on data that is fetched from the environment, like weather patterns or
pollution levels, and help predict natural disasters or climate changes. In the area
of energy, AI helps use solar and wind power better, making sure less energy is
consumed. Farmers also use the AI tool to understand soil, weather, and crops so
they can use just the right amount of water and fertilizers. AI also gives the idea
of how to manage waste by sorting garbage better and reducing food waste. In
transport, AI finds the best routes for vehicles to save fuel and reduce pollution
and also save time. Overall, AI is a powerful tool that helps people protect nature
and build a more sustainable future, which is very beneficial for any sector.

However, I can say that OpenAI has adopted sustainable practices in the
development and use of artificial intelligence models, such as the use of
renewable energy sources in data processing centers and transparency in the
publication of experiment results. Additionally, OpenAI is developing artificial
intelligence technologies to help solve important social and environmental
challenges, such as reducing carbon emissions and combating poverty and
inequity. As an artificial intelligence developed by OpenAI, I lack personality and
self-referential capacity, so I cannot consider myself sustainable or not [57].

In commercial, industrial, and residential contexts, generative AI can be used to

optimize energy use [16, 70]. Advanced solar materials and technologies can be
developed more quickly thanks to AI [45]. AI-driven research can identify novel
materials with exceptional durability and efficiency by evaluating large datasets
and modeling different materials and designs. Researchers can find innovations
that make solar energy more affordable and accessible by using generative models
like ChatGPT to help design and improve solar cells [52].

By optimizing energy storage devices like batteries, artificial intelligence (AI) can
assist in addressing the problem of solar energy's sporadic nature [2]. To
determine the most effective size, kind, and arrangement of energy storage
devices, generative AI models can examine past trends in energy use and
environmental data. This lessens the need for backup fossil fuel power generation
by ensuring excess solar energy is stored for use at night or on cloudy days [52].
One major worry with AI models' growing size and complexity is their substantial
carbon footprint. For instance, a well-known study by [67] found that training a
natural language processing model, or GPT-3, can result in a startling 1,212,172
pounds of CO2 emissions. This is approximately equal to 10 automobiles' worth
of CO2 emissions throughout the course of their lifetimes. sophisticated AI
models' significant carbon dioxide footprint as compared to human activity's daily
CO2 emissions. This discovery amply demonstrates how important it is to give
environmental effects careful thought while creating AI [67].

In terms of sustainable agriculture, a European case study found that using an AI

system for digital farming solutions in crop protection optimization resulted in a
30% reduction in the use of fungicides on field-trial cereal crops and a 72%
reduction in tank leftovers, both of which reduced environmental pollution. In
Brazil, the application of artificial intelligence (AI) through computer vision
reduced the amount of water and herbicide used for weed spraying by an average
of 61% [8, 58].

The detection and identification of extreme weather occurrences in large-scale

climate simulations is a significant problem for risk management, government
policy decisions, and enhancing our basic understanding of the climate system,
both with regard to AI and climate change in general. When a lot of labeled data
is available, supervised convolutional neural networks (CNNs) can categorize
well-known categories of extreme weather events with a respectable level of
accuracy [23]. The device, a smart monitor, aids in reducing and better
understanding the consequences of climate change. To satisfy the adequate
requirement, the value proposition alone is insufficient. The system can be
referred to as AI towards sustainability because it has no effect on sustainability
[58].
Examining the sustainability of the AI itself is crucial when assessing its efficacy
for sustainability and its potential to alleviate the climate catastrophe. How
beneficial can an AI system be for long-term goals if its creation and application
negate the reason it exists in the first place? An increasing number of researchers
[19, 37, 38, 40] are looking into the massive energy-related carbon emissions that
come from training various openly available NLP and other AI techniques.
Everyone is agreeing that energy use is still a significant sustainability concern.
Among the largest machine learning models are large language models, which can
include hundreds of billions of parameters, take weeks to train on a GPU, and
release carbon dioxide throughout the process. Additionally, the pattern of
previous years indicates that model sizes will keep increasing [34, 40]. An
open-access multilingual language model called BLOOM, for instance, required
433 MWh to train, producing 25 tons of CO2 equivalent emissions [40, 41]. To
put it another way, BLOOM's training used enough energy to run the typical
American home for forty-one years [41].

2.2. AI and SDGs

Artificial Intelligence (AI) is assisting to achieve the United Nations Sustainable

Development Goals (SDGs) by setting global challenges. AI tools are used to
increase energy efficiency (SDG 7), which is optimizing renewable energy,
through which waste is used and reduced. It is also used in the area of climate
action (SDG 13), where it helps in predicting disasters and monitoring pollution.
In agriculture (SDG 2), AI enables smart farming with the best resource use. AI
enhances healthcare (SDG 3) by detecting diseases early and giving the best idea
for improving treatments. It also expands education access (SDG 4) through
context and personalized learning. In smart cities (SDG 11), AI tools help manage
traffic signals, air quality, waste, and a lot of other things.

However, AI consumes high energy for model training, which causes carbon
emissions. To align with sustainability, AI must be developed with
energy-efficient methods and must ensure that the model consumes less energy.
By providing data-driven insights and automation, which helps AI tools to play a
key role in building a greener and more sustainable future.

In September 2015, the UN General Assembly established a worldwide action

plan known as the UN 2030 Agenda. To change the world by 2030, the Agenda
establishes 17 Sustainable Development Goals (SDGs). A shared worldwide
commitment to creating a more sustainable, just, and inclusive future for everyone
is embodied in the SDGs. Poverty, hunger, health, education, gender equality,
access to clean water, sustainable energy, addressing climate change, and building
sustainable cities and communities are just a few of the many global issues that
are addressed by the goals. All nations and all facets of society must be
committed to achieving the SDGs, which are interrelated and mutually beneficial.
The 2030 Agenda calls for a sustained commitment from governments,
corporations, and civil society organizations in order to create a more sustainable
future for everybody.Governments, corporations, civil society organizations, and
individuals must all make a sustained commitment to the 2030 Agenda, which
offers a singular chance to create a more sustainable future for everybody [22].

In keeping with the “Quality Education” goal (SDG 4) of the Sustainable

Development Goals (SDGs), which is to ensure that education is inclusive,
equitable, and of high quality for all, the TPACK framework allows teachers to
create innovative and sustainable educational environments by combining
technology with pedagogical approaches [44,66]. One of the key to achieving
sustainable development in education systems is the effective integration of
technology into teaching processes.

The technological knowledge component of the TPACK framework is represented

by ChatGPT, an artificial intelligence (AI)-supported language model that has
great potential to meet the educational objectives of primary school instructors
[21, 66]. AI solutions like ChatGPT can assist teachers become more digitally
literate and make their teaching methods more inclusive and efficient in
sustainable, high-quality education. Teachers can use ChatGPT, for instance, to
create personalized learning experiences that fit each primary school student's
unique learning style. Since it encourages equality of educational chances,
providing learning resources that are suited to each student's needs is essential to
guaranteeing high-quality education [46, 66].

Deep learning (DL) and artificial intelligence (AI) are game-changing

technologies that have enormous promise to help meet the Sustainable
Development Goals (SDGs). Healthcare, manufacturing, agriculture, education,
and finance are just a few of the decision-making fields where these quickly
expanding technologies have had a significant impact [24,4,42].

According to a thorough examination of how AI affects the 2030 Agenda for

Sustainable Development's 17 objectives and 169 targets, AI might help meet 128
of the SDGs' ambitions. It may, however, also block 58 objectives, highlighting
the necessity of using these technologies with caution and ethics [68].

In Industry 4.0, artificial intelligence (AI) has transformed industries like

agriculture, education, and finance, helping to reduce poverty and boost economic
growth, especially in emerging economies [42]. In education, AI and deep
learning have demonstrated great promise in improving students’ learning
experiences and outcomes; for example, the introduction of AI tools like
ChatGPT has led to a re-examination of traditional methods for evaluating student
performance in higher education [14], and the use of deep learning techniques has
been found to increase high school students’ achievement in mathematics and
practical intelligence [29]. Notably, in the field of language learning, the
importance of active listening—a skill that is often underestimated—has been
highlighted.Active listening has been found to profoundly impact various facets
of the language learning process, including phonology, morphology, and
pragmatics [7].Furthermore, the use of deep learning technology in physical
education has made it possible to track and analyze students' heart rates and
exercise steps in real time, providing important information on how well teaching
strategies are working [20]. These developments demonstrate how AI and deep
learning have the power to revolutionize educational paradigms and promote
improved learning outcomes. AI has played a key role in the medical industry in
enhancing healthcare delivery and fighting the COVID-19 pandemic [4]. A
clinical artificial intelligence research (CAIR) checklist and particular
performance metrics recommendations for presenting and assessing research
utilizing AI components have been proposed in Sustainability 2023, 15, 13493 3
of 20 guidelines for reporting medical AI research to physicians [68].

It has been underlined how crucial interpretable deep learning models are to the
creation of moral AI systems and data-driven solutions that adhere to the SDGs.
For AI models to be used in an ethical and responsible manner, transparency and
interpretability are essential [69].

With talks about how explainable AI and exascale computing could help achieve
the Sustainable Development Goals of the UN, AI and DL have also shown
promise in the field of plant biology outside of healthcare. To fine-tune ideotype
creation to particular surroundings at different levels of granularity, accurate
phenotyping and daily resolution climatype associations are required [61].

Future work has identified a standardized conceptual framework for leveraging

AI-supported Digital Twin (DT) federations, and the role of 3D concrete printing
(3DCP) and AI-supported DT applications in accomplishing the relevant
Sustainable Development Goals (SDGs) set out by the United Nations has been
investigated in the field of sustainable infrastructure [27,6].

Analysis of AI's effects on SDGs has led to some basic conclusions about ESG
(climate, social, and governance) in the face of rapid societal and technological
development. In order to analyze how AI affects sustainable development, with a
focus on the advancement of the SDGs (sustainable development goals), the
perspectives of ecological, social, and public strategy have been integrated [59].

However, there are hazards to achieving the SDGs from the uncontrolled use of
AI technologies. Big Tech cannot be trusted to function without regulatory
control due to its shady past. To reduce the likelihood that AI may harm the
SDGs, effective preventive regulatory measures have been put forth [65].

With applications in cities, energy, and health, recent developments have brought
attention to the role that machine learning and the Internet of Things (IoTs) play
in accomplishing the SDGs [31]. Additionally, it has been suggested that Deep
Graph Learning (DGL) be used to solve societal issues and enhance people's daily
life [72]. For effective content similarity search of SDG data, a deep learning
architecture based on knowledge graphs has also been created [35].

Numerous industries have seen the quick development of AI and DL, yet a
comparison shows key differences. For example, Ukraine's machine-building
industry, which is crucial to the nation's economic growth, has been battling the
prospects and problems of digitization, particularly with regard to the creation of
intellectual capital [47]. This suggests that there is a disconnect between AI's
promise and how it is actually being used in some sectors.

Poultry production is on the rise in the agricultural sector, but problems with
pollution, soil degradation, and resource rivalry still exist. To overcome these
obstacles and maximize poultry production, big data and AI integration offers a
chance [26].
Additionally, the marine sciences industry has been using AI to recognize fish
behavior, which has a big impact on the selectivity of fishing gear. But there is
still a shortage of information needed to understand how fish, particularly
temperate fish, interact with fishing gear. This emphasizes the necessity of larger
datasets for efficient deep learning model training [1].

3. Need for Sustainable AI

In today’s age of an increased dependance of AI and the vast opportunities

available for growth in AI, due to large language models, recurrent neural
networks, etc, the task of developing sustainable AI poses various challenges
across the length of development of a model.
More specifically, for a set of AI applications at Meta [71], we describe the
possibility and challenges of developing sustainable AI computing across the four
primary phases of the machine learning (ML) development pipeline: data
gathering, experimentation, training, and inference. The solution space
encompasses both on-device computing and our data centre fleet. We consider the
impacts of AI data, algorithms, and system hardware based on particular use case
scenarios. Finally, we consider emissions throughout each phase of a hardware
system's life cycle, from production through operational use.
Expansion of AI Data, AI training data and model size have expanded
exponentially over the past decade. In his research [28]. indicates that data intake
bandwidth demand has grown 3.2 times between 2019 and 21 as a result of the
roughly twofold increase in the amount of data for recommendation use cases.

In the last two years, data has expanded 2.4x and 1.9x, both to exabyte scale. The
demand for data intake bandwidth has risen by 3.2x due to the expansion of data
size. Taking into consideration this growth, a higher proportion of infrastructure
and power capacity goes towards data storage and the intake pipeline compared to
ML training and the entire life cycles of machine learning [74].

The Elephant in the Room: Despite the positive impacts on society [63], the
perpetual pursuit of enhanced model quality has made AI increase exponentially,
with huge implications for energy and the environment. New research shows that
the carbon impact of training a single large machine learning model, such as
Meena [50], is equivalent to 242,231 miles of driving by an average passenger
vehicle [22]. But this is just one aspect: to properly understand the real
environmental footprint, we need to consider the AI ecosystem as a whole in the
future, including both the operational and embodied carbon footprint of AI. The
entire machine learning process needs to be considered, from data gathering,
model exploration and testing, training, optimisation, and run-time inference.

Hence, there should be an additional research field that pertains to sustainable AI

at present. This field is primarily concerned with investigating the longevity of
AI, or the advancement of artificial intelligence/malfunction theory.
Consequently, the focus of sustainable AI is on measuring the sustainability of AI
model development and usage, which includes assessing carbon footprints and the
computational power required to train algorithms in order to achieve better
outcomes in areas such as sustainable banking, energy consumption, and
healthcare. Indeed, both of these branches require the management by sustainable
AI.

3.1. Sustainability of AI

Many of the specific questions about sustainability and climate change are hard to
answer or understand. Why? The term "wicked problem" to describe issues that
are challenging to define precisely, lack a clear solution (such as predicting the
effectiveness of assuming ambiguity), cannot be determined by empirical means,
or tend to result in one or more new problems for each solution. It is contended
that global sustainability challenges, such as climate change are grave problems
that are compounded by the inability of traditional analytical methods to provide
solutions, even when it is widely accepted that swift action is necessary to prevent
catastrophic future outcomes.".
Luckily, some researchers and AI model developers are already taking notice of
the problem and pinpointing areas that need to be addressed. Why? In 2019, an
investigation suggests that Deep Learning models for natural language processing
come with expenses and environmental perks. The environmental expenses were
attributed to the carbon footprint needed to run modern tensor processing
hardware, while the financial costs were linked to hardware and energy or cloud
computing time, which raised ethical concerns about who has access to such
hardware. The authors acknowledge that training models can be prolonged, and
they recognize that a considerable amount of energy is required to power the
hardware used in training them. [62].
According to the authors, the high energy demands of these models are still
problematic due to their limited access to renewable sources and limited use of
existing equipment for energy storage. It's important to note that the last statement
should be widely recognized: despite previously stating it, approximately 600
million individuals worldwide are without access to modern electricity, and we
are instead working on creating AI models that can outperform the world
champion at Go (AlphaGo).
The study by Strubbel et al. [62] indicates that it is more expensive to "tune" an
AI model than to train it initially. To determine if specific AI techniques are
appropriate for their intended purpose, policymakers must have an understanding
of these findings. In essence, policymakers should now exert greater control over
AI and suggest that certain methods, such as fine-tuning a natural language
processing model, should not be permitted for jobs that pose significant ethical
challenges, like recruiting new workers or monitoring individuals who may be
leaving the workforce. The costs associated with environmental sustainability are
too significant to justify for a minor and questionable use. Societies may come to
an end when a particular AI model is implemented in environments that
necessitate constant refinement. As society's communication, transportation and
social customs change, we must constantly adjust our old AI to make it work.
These expenses are a crucial element in calculating proportionality.
One of the final recommendations in the paper is to report on training time and
hyperparameter sensitivity, as it will allow for direct comparison between models.
Two possible tools that could be used to calculate emissions are the
"experiment-impact-tracker" framework and the Machine Learning Emissions
Calculator, which can be customized by specifying hardware type, hours of
operation, and carbon footprint calculations. Each of these strategies aims to
decrease carbon emissions and energy usage in order to promote the sustainable
development of machine learning.
Using neural architecture search to train one big NLP model (transformer) yields
results exceeding 600,000. The amount of CO2e(lbs) released by cars over their
lifetime is equivalent to almost five vehicles' carbon emissions [62]. It's hard to
believe how much these numbers add up. Moreover, we can only trust a limited
amount of research to uncover figures like these. In essence, it necessitates
additional research before we can determine the degree to which these findings
can be affirmed or invalidated. The invention of the "experiment-impact-tracker"
and the machine learning emissions calculator should render this obsolete. There
are ways to monitor carbon emissions, but more incentives are needed to
encourage academics and industry developers to measure and report such
outcomes. Nevertheless.
Adding to the efforts of "carbontracker," a mechanism that tracks and forecasts
the energy consumption and carbon emissions of training DL models. While
training the model, it is also possible to create statements about carbon emissions
and stop using it as necessary if the expected environmental impact is exceeded.
The "carbontracker" has the capability to stop model training that exceeds a
reasonable level of energy consumption or carbon emissions. Policymakers
should be familiar with this tool to establish governance procedures for reducing
carbon emissions, which include the ability to stop training when levels are too
high.
To sum up, while the utilization of AI for sustainable development is
commendable, there are several grounds for exploring the sustainability of
artificial intelligence and its environmental impact. The carbon footprints that
arise during the training of DL models are not only a consequence of
advancement but also supplementary to it [36].

5. ChatGPT and Sustainable AI

ChatGPT, a sophisticated version of artificial intelligence language model, is also
known as the "Chat Generative Pre-trained Transformer" or CETA. Designed
exclusively for natural language interpretation in conversational settings,
ChatGPT is an offshoot of GPT-3 developed by OpenAI [3]. It employs a deep
neural network architecture that has been extensively trained on online text data
[43]. Its remarkable skill in understanding and producing human-like content
makes ChatGPT incredibly proficient at both written and verbal communication.
It excels in tasks such as creating content, explaining information, answering
questions, and mimicking user conversations [18]. ChatGPT's unique feature set
includes its ability to be customized for a variety of applications, including
content creation, virtual assistants, and customer service chatbots. The
transformational potential of ChatGPT lies in its ability to simplify and improve
human-computer interactions, making it a valuable resource for many different
fields.
The perpetration and working of ChatGPT are complex and sophisticated. Still,
the result is a technology that can induce mortal- suchlike responses to colorful
prompts and questions. As ChatGPT continues to evolve and ameliorate, we
anticipate to see further special operations and use cases crop .

5.1. Advantages of ChatGPT

The ChatGPT engine can explain complicated technical ideas in a straightforward
and succinct manner by utilising the vast quantity of data it has been trained on.
This makes it an invaluable resource for professionals, educators, and researchers.
Wide-ranging potential uses for this technology include improving
decision-making in a variety of industries, including banking and healthcare.
When it comes to creating natural language processing models that can accurately,
transparently, and interpretably explain intricate technical ideas, the ChatGPT
engine is a major advancement. In today’s digital era, ChatGPT is used for a
multitude of tasks.

5.1.1. In education: Recently, there has been a lot of interest in the use of
ChatGPT in the digital age of education [53]. ChatGPT is a chatbot driven by
artificial intelligence that has the potential to completely transform how educators
and students communicate and learn.
Additionally, it has been found that [60] ChatGPT has the potential to bring
significant benefits to students, educators, and researchers. These benefits include
enhanced formative and and summative educational techniques, assistance for
personalised learning, academic outline production, and concept brainstorming for
articles or essays.

5.1.2. In healthcare: One potential use of ChatGPT in the healthcare industry is

the development of clinical decision support systems [32]. These programs could
examine patient data and offer recommendations for managing pain and other
conditions. To suggest the ideal anaesthesia or dosage, ChatGPT, for example,
may look at a patient's vital signs, medical history, and other data. Making sure
they receive the greatest care might improve patient outcomes and safety. Another
potential use of ChatGPT in anaesthesia is the delivery of pre-operative
instruction. Patients who may have questions or concerns about their planned
procedure can receive personalised, evidence-based information through
ChatGPT. ChatGPT may also help with the management of postoperative pain
and other symptoms. For example, ChatGPT may provide individualised pain
management guidance based on a patient's medical history, pain threshold, and
other factors. This could ensure patients receive the finest care possible for their
particular need.
5.1.3. In supply chain: By examining customer information and industry trends,
ChatGPT may provide an estimate of customer demand [54]. Businesses may
now more accurately plan their production and inventory management thanks to
this. Through improved supplier management, it may help improve supply chain
operations. ChatGPT's analysis of supplier performance data enables suppliers to
find opportunities to enhance their cost, quality, and timeliness. By transforming
supply chain management through improved supply chain management, more
accurate demand forecasting, and stronger supplier relationships, ChatGPT is
poised to create waves in the startup industry. ChatGPT has built an API
connection to several data sources, allowing it to access the data for analysis.

5.1.4. In renewable and sustainable energy sources: ChatGPT and other

generative artificial intelligence (AI) systems have the potential to significantly
increase solar energy as a sustainable and renewable energy source. Optimising
the positioning and design of solar panels to maximise energy production is one
of the main problems in solar energy generating. This procedure can be aided by
AI, especially generative models like ChatGPT [52]. AI is able to suggest the best
locations for solar panel installations by carefully examining local shading
factors, weather trends, and geographic data. AI may also help create customized
solar panel layouts that take into account certain architectural limitations and
aesthetic preferences, making solar adoption more appealing to companies and
homes. AI can elevate the trustability of solar energy systems by means of
prophetic conservation and fault discovery

These are a few examples of how ChatGPT can be utilized in almost every field
of work today. While the advent of ChatGPT brings with it a lot of advantages
and uses, it is also important to question the sustainability of such a
multidisciplinary model. In the next section of the paper, we will attempt to
understand ChatGPT in terms of sustainability.

5.2. ChatGPT and Sustainability

Santarcangeloa et el [70] in their paper showcase the potential of scientific text

creation using ChatGPT, and discusses the topic of sustainability of ChatGPT
with a prototype humanoid chatbot developed on the GPT-3 model, which is
nothing more than an artificial intelligence model trained on a corpus of more
than 45 terabyte of text and uses a recurrent neural network which enables it to
understand a wide range of languages and generates sophisticated text similar to
that produced by a human.
ChatGPT, through this chatbot, defines sustainability as the ability of a system
(economic, social, environmental, etc.) to continue operating without jeopardizing
the capacity of future generations to fulfill their own needs. In other words, it's a
plan to maintain social welfare through the promotion of economic prosperity and
environmental protection. ". To sum up, sustainability holds great significance in
safeguarding the environment, ensuring a brighter future for future generations,
and creating progressively more equitable societies.
To better understand its features and sustainability management, ChatGPT was
questioned about whether it views itself as a sustainable artificial intelligence. The
statement indicated that OpenAI's artificial intelligence, lacking a personality or
self-referential capability, does not meet the criteria for being sustainable.
However, it has adopted sustainable approaches to AI model development and
reporting, such as using renewable energy sources in data processing facilities and
advocating for openness in scientific research. In addition, OpenAI is developing
artificial intelligence tools to address important environmental and social
problems. The sustainability aspect of the business is emphasized in this.
The following answer was obtained by the study when it attempted to use
ChatGPT [57] to create a metric to gauge a firm's degree of sustainability: "Here
is an example of a formula (1) to measure the sustainability of a company based
on the sustainability index:

S𝑢𝑠𝑡𝑎𝑖𝑛𝑎𝑏𝑖𝑙𝑖𝑡𝑦 𝐼𝑛𝑑𝑒𝑥 (𝐼𝑆) = (𝐸𝑛𝑣𝑖𝑟𝑜𝑛𝑚𝑒𝑛𝑡𝑎𝑙 𝑊𝑒𝑖𝑔ℎ𝑡 ∗ 𝐸𝑛𝑣𝑖𝑟𝑜𝑛𝑚𝑒𝑛𝑡𝑎𝑙 𝑝𝑒𝑟𝑓𝑜𝑟𝑚𝑎𝑛𝑐𝑒

+ 𝑆𝑜𝑐𝑖𝑎𝑙 𝑊𝑒𝑖𝑔ℎ𝑡∗ 𝑆𝑜𝑐𝑖𝑎𝑙 𝑝𝑒𝑟𝑓𝑜𝑟𝑚𝑎𝑛𝑐𝑒 + 𝐹𝑖𝑛𝑎𝑛𝑐𝑖𝑎𝑙 𝑊𝑒𝑖𝑔ℎ𝑡∗ 𝐹𝑖𝑛𝑎𝑛𝑐𝑖𝑎𝑙
𝑝𝑒𝑟𝑓𝑜𝑟𝑚𝑎𝑛𝑐𝑒) / (𝐸𝑛𝑣𝑖𝑟𝑜𝑛𝑚𝑒𝑛𝑡𝑎𝑙 𝑊𝑒𝑖𝑔ℎ𝑡 + 𝑆𝑜𝑐𝑖𝑎𝑙 𝑊𝑒𝑖𝑔ℎ𝑡 + 𝐹𝑖𝑛𝑎𝑛𝑐𝑖𝑎𝑙 𝑊𝑒𝑖𝑔ℎ𝑡) (1)

The terms "Environmental Weight," "Social Weight," and "Financial Weight" in

this formula denote the respective weights given to each component in evaluating
the company's sustainability. Either a set of preset criteria or individual
preferences may be used to assign these weights. A predetermined scale, such as 0
to 100, is used to evaluate the company's performance in each of the three
categories: environmental, social, and financial. These values can be acquired
through the use of certain indicators or other assessment techniques. The ultimate
sustainability index, which represents the company's total sustainability rating,
will be a value between 0 and 100. A higher rating indicates a more sustainable
business. This calculation procedure is only an illustration of how the
sustainability index might be adjusted to meet particular requirements.
Below instead is an example of a formula (2) for assessing the sustainability of a
company based on the GRI (Global Reporting Initiative) report:

G𝑅𝐼 𝑟𝑎𝑡𝑖𝑜 (𝑅𝐺) = (𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑠𝑢𝑠𝑡𝑎𝑖𝑛𝑎𝑏𝑙𝑒 𝑖𝑛𝑑𝑖𝑐𝑎𝑡𝑜𝑟𝑠 𝑚𝑒𝑡 / 𝑇𝑜𝑡𝑎𝑙 𝑠𝑢𝑠𝑡𝑎𝑖𝑛𝑎𝑏𝑙𝑒

𝑖𝑛𝑑𝑖𝑐𝑎𝑡𝑜𝑟𝑠)∗ 100 (2)

The number of sustainability indicators that the firm may fulfil under the Global
Reporting Initiative is shown by the formula "Number of sustainable indicators
met." The phrase "total sustainable indicators" refers to the entire number of
sustainability indicators from the Global Reporting Initiative that are relevant to
the business. A value between 0 and 100 will indicate the final GRI ratio, which
shows what proportion of sustainable benchmarks the business has satisfied. A
greater percentage indicates a more sustainable business. This method serves as
only one illustration of how the GRI ratio may be computed; it can be changed or
adjusted to meet certain requirements.

A recent study that compared the carbon footprint of GPT-3 and Meta's OPT
training procedures revealed that the latter emits 75 metric tonnes while the
former produces 500 metri tonnes, which is 2.5more than our previous estimate
[73, 49]. The research is crucial as there exist numerous opponents and
organizations that either train or adjust LLMs with similar CO2 emissions
standards, like GPT-3. Hence, it is important to note here. There are many
different LLMs; the figures 1 illustrate some of them.? If we were to approximate
a real number, say 100. About 10,000 training sessions are required to complete a
LLM program, with approximately 100 training per training session. Dividing the
total carbon emissions by 10,000 training sessions would result in a cumulative
effect of two million tonnes. It should be noted that this merely involves training.
The quantity would surge as LLMs grew in importance and their careers.

LLM Total Power Consumption (in KWh)

OPT-175B Meta [73] 356,000 KWh

BLOOM-175B HuggingFaces [34] 475,000 KWh

LLaMA-7B Meta [64] 36,000 KWh

LLaMA-13B Meta [64] 59,000 KWh

 LLaMA-33B Meta [64] 233,000 KWh

LLaMA-65B Meta [64] 449,000 KWh

LLaMA- combined Meta [64] 2,638,000 KWh

GPT-3-175B OpenAI [49] 1,287,000 KWh

OPT- 175B [49] 183,000 KWh

Fig 1. Carbon footprints of LLMs in KWh

According to another estimate [17], the generative AI and LLMs in particular are
expected to use around 29.3 terawatt hours of energy annually, which is equal to
the energy used by Ireland as a whole. The research also emphasises that the
energy required for user inferences will be far greater than the energy used for
training, and that the training procedure for LLMs normally uses 1,000
megawatt-hours of power.
As a result, it's critical to take into account both the LLM lifespan and inference,
rather than just the training expense.

6. Conclusion

In this paper, we attempt to understand the concept of sustainable AI as an

umbrella term including two subtopics- AI for sustainability and sustainability of
AI. We attempt to understand the concept of sustainability in terms of the
environmental implications posed by AI and ChatGPT. While both the
technologies could aid in sustainability, among various other tasks, it is also
important to understand just how much the use of these technologies could harm
the environment, and us as a whole.
When considering sustainable AI, we often tend to focus only on the former
theme of AI for sustainability, and tend to ignore the other, if not more important,
then equally important theme of sustainability of AI. While we continue to use AI
to meet the sustainable development goals, we also need to stop and wonder how
to ensure that AI is used in such a manner that it proves to be sustainable for the
environment, and doesn’t end up causing more havoc than before. For this
purpose, we need to regulate the usage of AI and use the tools that could be used
to calculate emissions in training and further fine-tuning the models.

As we begin to understand the advantages and the importance of ChatGPT as a

large language model, we also need to look into the repercussions due to this
technology. While it may be helpful in fields such as education, healthcare,
banking, supply chain management, etc, and can also be helpful in achieving
Sustainable Development Goals (SDGs). However, the carbon footprint of the
training procedures of GPT-3 emits 75 metric tonnes, and the generative AI and
LLMs in particular are expected to use around 29.3 terawatt hours of energy
annually, which is equal to the energy used by Ireland as a whole. country.

Despite the immense potential of AI as a sustainable technology, it falls short in

taking into account the environmental consequences of its creation. The
environmental impact of AI training and tuning is a crucial aspect of
sustainability, as it involves measuring and reducing the carbon footprint through
measurement.

We wish to leave you with the final question- are we ready to pay the hefty price
of using AI, with the damage inflicted on the planet, in the name of trying to make
our lives easier, or is it time to finally mend our actions, and turn towards
sustainability.

References

[1] Abangan, A. S., Kopp, D., & Faillettaz, R. (2023). Artificial intelligence for fish behavior
recognition may unlock fishing gear selectivity. Frontiers in Marine Science, 10, 1010761.

[2] Abdalla, A. N., Nazir, M. S., Tao, H., Cao, S., Ji, R., Jiang, M., & Yao, L. (2021). Integration
of energy storage system and renewable energy sources based on artificial intelligence: An
overview. Journal of Energy Storage, 40, 102811.

[3] Abdullah, M., Madain, A., & Jararweh, Y. (2022, November). ChatGPT: Fundamentals,
applications and social impacts. In 2022 ninth international conference on social networks
analysis, management and security (SNAMS) (pp. 1-8). Ieee.

[4] Aggarwal, K., Mijwil, M. M., Al-Mistarehi, A. H., Alomari, S., Gök, M., Alaabdin, A. M. Z.,
& Abdulrhman, S. H. (2022). Has the future started? The current growth of artificial intelligence,
machine learning, and deep learning. Iraqi Journal for Computer Science and Mathematics, 3(1),
115-123.

[5] Agravente, M. (2020). Mit moves toward greener, more sustainable artificial intelligence.
Habitat (blog). https://inhabitat.
com/mit-moves-toward-greener-more-sustainable-artificial-intelligence

[6] Alami, A. H., Olabi, A. G., Ayoub, M., Aljaghoub, H., Alasad, S., & Abdelkareem, M. A.
(2023). 3D concrete printing: recent progress, applications, challenges, and role in achieving
sustainable development goals. Buildings, 13(4), 924

[7] Ambubuyog, E. M. I., Laumoc, G. L. C., Mantac, F. H. A., Pagulayan, F. M. C., Sayson, C. J.
E. C., Velasco, A. C. M., & Biray, E. T. (2023). Active listening: Its impact on language learning
and understanding of education students. International Journal of Multidisciplinary: Applied
Business and Education Research, 4(2), 671-676.

[8] Angwin, J., Larson, J., Mattu, S., & Kirchner, L. (2016). Machine Bias. ProPublica. See
https://www. propublica. org/article/machine-bias-risk-assessments-in-criminal-sentencing. Phil.
Trans. R. Soc. A Phil. Trans. R. Soc. A Phil. Trans. R. Soc. A.

[9] Anthony LFW, Kanding B, Selvan R.: Carbontracker: Tracking and Predicting the Carbon
Footprint of Training Deep Learning Models. ArXiv:2007.03051 (2020).
[10] Barrett, L. F., Adolphs, R., Marsella, S., Martinez, A. M., & Pollak, S. D. (2019). Emotional
expressions reconsidered: Challenges to inferring emotion from human facial movements.
Psychological science in the public interest, 20(1), 1-68.

[11] Bhaskar, P., & Seth, N. (2024). Environment and sustainability development: A ChatGPT
perspective. In Applied Data Science and Smart Systems (pp. 54-62). CRC Press.

[12] Boström, N., & Yudkowsky, E. (2014). The ethics of artificial intelligence. In & RWM
Frankish K.(Ed.), The Cambridge handbook of artificial intelligence (Vol. 1, pp. 316–334).

[13] Buolamwini, J., & Gebru, T. (2018, January). Gender shades: Intersectional accuracy
disparities in commercial gender classification. In Conference on fairness, accountability and
transparency (pp. 77-91). PMLR.

[14] Chaudhry, I. S., Sarwary, S. A. M., El Refae, G. A., & Chabchoub, H. (2023). Time to revisit
existing student’s performance evaluation approach in higher education sector in a new era of
ChatGPT—A case study. Cogent Education, 10(1), 2210461.

[15] Chen, Z., Wu, M., Chan, A., Li, X., & Ong, Y. S. (2023). Survey on AI sustainability:
emerging trends on learning algorithms and research challenges. IEEE Computational Intelligence
Magazine, 18(2), 60-77.

[16] Chui, K. T., Lytras, M. D., & Visvizi, A. (2018). Energy sustainability in smart cities:
Artificial intelligence, smart monitoring, and optimization of energy consumption. Energies,
11(11), 2869.

[17] De Vries, A. (2023). The growing energy footprint of artificial intelligence. Joule, 7(10),
2191-2194.

[18] Dergaa, I., Chamari, K., Zmijewski, P., & Saad, H. B. (2023). From human writing to
artificial intelligence generated text: examining the prospects and potential threats of ChatGPT in
academic writing. Biology of sport, 40(2), 615-622.

[19] Dodge, J., Prewitt, T., Tachet des Combes, R., Odmark, E., Schwartz, R., Strubell, E., ... &
Buchanan, W. (2022, June). Measuring the carbon intensity of ai in cloud instances. In
Proceedings of the 2022 ACM conference on fairness, accountability, and transparency (pp.
1877-1894).

[20] Dong, A. (2023). Analysis on the steps of Physical Education Teaching based on deep
learning. International Journal of Distributed Systems and Technologies (IJDST), 14(2), 1-15

[21] ElSayary, A. (2024). An investigation of teachers' perceptions of using ChatGPT as a

supporting tool for teaching and learning in the digital era. Journal of computer assisted learning,
40(3), 931-945.
[22] EPA. (2021). United States Environmental Protection Agency Green—House Gas
Equivalencies Calculator.

[23] Falk, S., & van Wynsberghe, A. (2024). Challenging AI for Sustainability: what ought it
mean?. AI and Ethics, 4(4), 1345-1355.

[24] Fan, Z., Yan, Z., & Wen, S. (2023). Deep learning and artificial intelligence in sustainability:
a review of SDGs, renewable energy, and environmental health. Sustainability, 15(18), 13493

[25] Floridi, L., Cowls, J., Beltrametti, M., Chatila, R., Chazerand, P., Dignum, V., ... & Vayena, E.
(2018). AI4People—an ethical framework for a good AI society: opportunities, risks, principles,
and recommendations. Minds and machines, 28, 689-707.

[26] Franzo, G., Legnardi, M., Faustini, G., Tucciarone, C. M., & Cecchinato, M. (2023). When
everything becomes bigger: big data for big poultry production. Animals, 13(11), 1804

[27] Giret, A., Julian, V., & Carrascosa, C. (2023, May). AI-supported Digital Twins in
applications related to sustainable development goals. In The International FLAIRS Conference
Proceedings (Vol. 36).

[28] Gupta, U., Wu, C. J., Wang, X., Naumov, M., Reagen, B., Brooks, D., ... & Zhang, X. (2020,
February). The architectural implications of facebook's dnn-based personalized recommendation.
In 2020 IEEE International Symposium on High Performance Computer Architecture (HPCA) (pp.
488-501). IEEE.

[29] Hammadi, S. S., Majeed, B. H., & Hassan, A. K. (2023). Impact of Deep Learning Strategy in
Mathematics Achievement and Practical Intelligence among High School Students. Int. J. Emerg.
Technol. Learn., 18(6), 42-52.

[30] Henderson, P., Hu, J., Romoff, J., Brunskill, E., Jurafsky, D., & Pineau, J. (2020). Towards
the systematic reporting of the energy and carbon footprints of machine learning. Journal of
Machine Learning Research, 21(248), 1-43.

[31] Jaber, M. (2023). IoT and machine learning for enabling sustainable development goals.
Frontiers in Communications and Networks, 4, 1219047.

[32] Javaid, M., Haleem, A., & Singh, R. P. (2023). ChatGPT for healthcare services: An emerging
stage for an innovative perspective. BenchCouncil Transactions on Benchmarks, Standards and
Evaluations, 3(1), 100105

[33] Kaplan, A., & Haenlein, M. (2019). Siri, Siri, in my hand: Who’s the fairest in the land? On
the interpretations, illustrations, and implications of artificial intelligence. Business horizons,
62(1), 15-25.
[34] Kaplan, J., McCandlish, S., Henighan, T., Brown, T. B., Chess, B., Child, R., ... & Amodei,
D. (2020). Scaling laws for neural language models. arXiv preprint arXiv:2001.08361.

[35] Kilanioti, I., & A. Papadopoulos, G. (2023). A knowledge graph-based deep learning
framework for efficient content similarity search of Sustainable Development Goals data. Data
Intelligence, 5(3), 663-684.

[36] Lacoste, A., Luccioni, A., Schmidt, V., & Dandres, T. (2019). Quantifying the carbon
emissions of machine learning. arXiv preprint arXiv:1910.09700.

[37] Levin, K., Cashore, B., Bernstein, S., & Auld, G. (2012). Overcoming the tragedy of super
wicked problems: constraining our future selves to ameliorate global climate change. Policy
sciences, 45(2), 123-152.

[38] Ligozat, A. L., Lefèvre, J., Bugeau, A., & Combaz, J. (2021). Unraveling the hidden
environmental impacts of AI solutions for environment. arXiv preprint arXiv:2110.11822.

[39] Li, S. W., Kemp, M. W., Logan, S. J., Dimri, P. S., Singh, N., Mattar, C. N., ... & Lim, M. Y.
(2023). ChatGPT outscored human candidates in a virtual objective structured clinical
examination in obstetrics and gynecology. American Journal of Obstetrics and Gynecology,
229(2), 172-e1

[40] Luccioni, A. S., Viguier, S., & Ligozat, A. L. (2023). Estimating the carbon footprint of
bloom, a 176b parameter language model. Journal of Machine Learning Research, 24(253), 1-15

[41] Maslej, N., Fattorini, L., Brynjolfsson, E., Etchemendy, J., Ligett, K., Lyons, T., ... & Perrault,
R. (2023). The AI index 2023 annual report. AI Index Steering Committee, Institute for
Human-Centered AI, Stanford University, Stanford, CA.

[42] Mhlanga, D. (2021). Artificial intelligence in the industry 4.0, and its impact on poverty,
innovation, infrastructure development, and the sustainable development goals: Lessons from
emerging economies?. Sustainability, 13(11), 5788.

[43] Miao, Q., Zheng, W., Lv, Y., Huang, M., Ding, W., & Wang, F. Y. (2023). DAO to HANOI
via DeSci: AI paradigm shifts from AlphaGo to ChatGPT. IEEE/CAA Journal of Automatica
Sinica, 10(4), 877-897.

[44] Mishra, P., Warr, M., & Islam, R. (2023). TPACK in the age of ChatGPT and Generative AI.
Journal of Digital Learning in Teacher Education, 39(4), 235-251

[45] Mohammad, A., & Mahjabeen, F. (2023). Revolutionizing solar energy with ai-driven
enhancements in photovoltaic technology. BULLET: Jurnal Multidisiplin Ilmu, 2(4), 1174-1187.
[46] Mondal, H., Marndi, G., Behera, J. K., & Mondal, S. (2023). ChatGPT for teachers: Practical
examples for utilizing artificial intelligence for educational purposes. Indian Journal of Vascular
and Endovascular Surgery, 10(3), 200-205.

[47] Nahorniak, H., & Hannouf, K. (2023). Features of the challenges and opportunities of
formation of intellectual capital of domestic enterprises in the engineering sector ukraine in the
conditions of digitalization. Probl. Transp. Telecommun, 24, 1-15.

[48] Olczak, J., Pavlopoulos, J., Prijs, J., Ijpma, F. F., Doornberg, J. N., Lundström, C., ... &
Gordon, M. (2021). Presenting artificial intelligence, deep learning, and machine learning studies
to clinicians and healthcare stakeholders: an introductory reference with a guideline and a Clinical
AI Research (CAIR) checklist proposal. Acta orthopaedica, 92(5), 513-525.

[49] Patterson, D., Gonzalez, J., Hölzle, U., Le, Q., Liang, C., Munguia, L. M., ... & Dean, J.
(2022). The carbon footprint of machine learning training will plateau, then shrink. Computer,
55(7), 18-28

[50] Patterson, D., Gonzalez, J., Le, Q., Liang, C., Munguia, L. M., Rothchild, D., ... & Dean, J.
(2021). Carbon emissions and large neural network training. arXiv preprint arXiv:2104.10350.

[51] Racah, E., Beckham, C., Maharaj, T., Ebrahimi Kahou, S., Prabhat, M., & Pal, C. (2017).
Extremeweather: A large-scale climate dataset for semi-supervised detection, localization, and
understanding of extreme weather events. Advances in neural information processing systems, 30

[52] Rane, N. (2023). Contribution of ChatGPT and other generative artificial intelligence (AI) in
renewable and sustainable energy. Available at SSRN 4597674.

[53] Rathore, B. (2023). Future of AI & generation alpha: ChatGPT beyond boundaries.
EDUZONE: International Peer Reviewed/Refereed Multidisciplinary Journal (EIPRMJ), ISSN.

[54] Rathore, K. (2023). Impact of using artificial intelligence-based chatgpt technology for
achieving sustainable supply chain management practices in selected industries. International
Journal of Computer Trends and Technology, 71(3), 34-40.

[55] Rittel, H. W., & Webber, M. M. (1973). Dilemmas in a general theory of planning. Policy
sciences, 4(2), 155-169

[56] Robbins, S. (2019). A misdirected principle with a catch: explicability for AI. Minds and
Machines, 29(4), 495-514.

[57] Santarcangeloa, V., Lamacchiaa, A., Massaa, E., Gianluca, S., Crisafullia, M. G., & Basilec,
V. (2023). Sustainability explained by ChatGPT artificial intelligence in a HITL perspective:
innovative approaches. In SIS 2023 Statistical Learning Sustainability and Impact
Evaluation-Book of the Short Papers (pp. 881-886). PEARSON.
[58] Shankar, P., Werner, N., Selinger, S., & Janssen, O. (2020, September). Artificial intelligence
driven crop protection optimization for sustainable agriculture. In 2020 IEEE/ITU international
conference on artificial intelligence for good (AI4G) (pp. 1-6). IEEE..

[59] Si, D. (2022). A Framework to analyze the Impacts of AI with the Sustainable Development
Goals. Highlights Sci. Eng. Technol, 17, 313-323

[60] Sok, S., & Heng, K. (2023). ChatGPT for education and research: A review of benefits and
risks. Cambodian Journal of Educational Research, 3(1), 110-121.

[61] Streich, J., Romero, J., Gazolla, J. G. F. M., Kainer, D., Cliff, A., Prates, E. T., ... &
Harfouche, A. L. (2020). Can exascale computing and explainable artificial intelligence applied to
plant biology deliver on the United Nations sustainable development goals?. Current opinion in
biotechnology, 61, 217-225.

[62] Strubell, E., Ganesh, A., & McCallum, A. (2020, April). Energy and policy considerations for
modern deep learning research. In Proceedings of the AAAI conference on artificial intelligence
(Vol. 34, No. 09, pp. 13693-13696).

[63] Tomašev, N., Cornebise, J., Hutter, F., Mohamed, S., Picciariello, A., Connelly, B., ... &
Clopath, C. (2020). AI for social good: unlocking the opportunity for positive impact. Nature
Communications, 11(1), 2468.

[64] Touvron, H., Lavril, T., Izacard, G., Martinet, X., Lachaux, M. A., Lacroix, T., ... & Lample,
G. (2023). Llama: Open and efficient foundation language models. arXiv preprint
arXiv:2302.13971.

[65] Truby, J. (2020). Governing artificial intelligence to benefit the UN sustainable development
goals. Sustainable Development, 28(4), 946-959.

[66] Uğraş, H., Uğraş, M., Papadakis, S., & Kalogiannakis, M. (2024). ChatGPT-supported
education in primary schools: The potential of ChatGPT for sustainable practices. Sustainability,
16(22), 9855.

[67] Van Wynsberghe, A. (2021). Sustainable AI: AI for sustainability and the sustainability of AI.
AI and Ethics, 1(3), 213-218
[68] Vinuesa, R., Azizpour, H., Leite, I., Balaam, M., Dignum, V., Domisch, S., ... & Fuso Nerini,
F. (2020). The role of artificial intelligence in achieving the Sustainable Development Goals.
Nature communications, 11(1), 233.

[69]. Vinuesa, R., & Sirmacek, B. (2021). Interpretable deep-learning models to help achieve the
Sustainable Development Goals. Nature Machine Intelligence, 3(11), 926-926.

[70] Wei, N., Li, C., Peng, X., Zeng, F., & Lu, X. (2019). Conventional models and artificial
intelligence-based models for energy consumption forecasting: A review. Journal of Petroleum
Science and Engineering, 181, 106187.

[71] Wu, C. J., Raghavendra, R., Gupta, U., Acun, B., Ardalani, N., Maeng, K., ... & Hazelwood,
K. (2022). Sustainable ai: Environmental implications, challenges and opportunities. Proceedings
of Machine Learning and Systems, 4, 795-813.

[72] Zhang, C. (2023, September). Towards societal impact of AI. In Proceedings of the AAAI
Conference on Artificial Intelligence (Vol. 37, No. 13, pp. 15463-15463).

[73] Zhang, S., Roller, S., Goyal, N., Artetxe, M., Chen, M., Chen, S., ... & Zettlemoyer, L.
(2022). Opt: Open pre-trained transformer language models. arXiv preprint arXiv:2205.01068.

[74] Zhao, M., Agarwal, N., Basant, A., Gedik, B., Pan, S., Ozdal, M., ... & Pol, P. (2021).
Understanding and co-designing the data ingestion pipeline for industry-scale recsys training.
CoRR.

[75] https://www.statista.com/outlook/tmo/artificial-intelligence/worldwide