Have you ever seen advertisements on YouTube or TV created using AI? For example, advertisements from LG Uplus, LOTTE, and KRUSH. These advertisements are produced using Generative AI, and not only images but also music is created by AI. Earlier this year, Samsung Electronics also unveiled the Galaxy S24, the world's first AI-enabled smartphone. Through AI, the phone can translate calls in real time, summarize texts, and even help edit photos. With the development of artificial intelligence, many companies are using Generative AI or interactive artificial intelligence in diverse areas. In addition to these examples, various AI programs are also being utilized. For example, interactive AI such as Siri and Bixby, webtoon generators such as Tooning, and voice generators such as Murf.
As the pace of AI development accelerates, its use in everyday life continues to grow. We are already accustomed to AI, and it often feels natural. One of the most familiar and accessible AI tools for students is ChatGPT.
So, what is ChatGPT and what are its advantages and disadvantages?
ChatGPT is a compound word formed from “Chat” and “Generative Pre-trained Transformer (GPT)” and is a prototype interactive AI chatbot developed by OpenAI. When the user gives ChatGPT some topics, situations, or questions, it can answer like a human being through the interactive interface.
There are four major advantages of ChatGPT. First, it has outstanding analytical and summarizing abilities. Second, it doesn’t just provide results; it also explains the information, as if talking to a person. Third, ChatGPT is a versatile tool. It can be utilized in various contexts, such as writing, translation, image generation, and programming, with just the user’s situational setting. Lastly, users can train ChatGPT. It can think and learn like a person because it uses deep learning technology. Additionally, it offers both free and paid versions, so anyone can enjoy the above advantages for free if they accept limitations on the number of questions and quality.
The disadvantages can be summarized into four points. First, the information provided by ChatGPT is not completely verified and reliable. Second, because ChatGPT generates responses by synthesizing search results, it can be difficult to ascertain the sources of its answers, raising a risk of copyright infringement. Third, it tends to be easily biased based on its training data. This can lead to false statements or unfair treatment toward specific groups, including racial, gender, linguistic, or cultural minorities. Lastly, ChatGPT excels only within the confines of its training data and lacks the ability to generalize new data. Since ChatGPT generates answers by retrieving, organizing, and combining pre-learned data, it is prone to errors when confronted with entirely new information, as it may struggle to generalize it effectively.
However, a major disadvantage of ChatGPT that deserves our attention is its significant contribution to environmental pollution. The process of receiving responses from ChatGPT typically involves cross-questioning, aimed at minimizing errors in earlier interactions and obtaining answers as close to the truth as possible. This is because ChatGPT, as previously mentioned, does not always provide precise or verified information.
Did you know that an enormous amount of water and energy are consumed during the process of training ChatGPT?
Fundamentally, all technologies consume a vast amount of energy. Even a single Google search uses approximately 0.3 Wh of electricity. However, search engines powered by large language models (LLMs) like ChatGPT require nearly ten times that amount of energy per query. According to Stanford University’s AI Index Report, GPT-3, the foundation of ChatGPT, consumed 1,287 MWh of electricity during its training process. The estimated carbon emissions from this process amount to 502 tons, equivalent to the emissions generated by the average global citizen over a span of 100 years.
Nevertheless, it has been revealed that training artificial intelligence consumes not only electricity but also a significant amount of water. In the process of storing data for generative AI, data centers use GPUs (Graphics Processing Units) to handle large volumes of data. Since GPUs generate a lot of heat during operation, the temperature must be kept between 10 and 25 degrees Celsius to prevent malfunction. This process requires a substantial amount of cooling water.
Researchers from the University of California, Riverside, and the University of Texas, Arlington, have calculated that each conversation with ChatGPT consumes approximately 500 mL of water. According to their findings, GPT-3 is estimated to have used around 700,000 liters of water during its training process. Furthermore, the water used in data centers must be clean freshwater, such as drinking water or treated wastewater, to prevent corrosion or bacterial growth.
The researchers define "a single conversation" as an interaction involving 25 to 50 exchanges of questions and answers. While this benchmark is not insignificant, we must not overlook the environmental impact of ChatGPT, especially given its immense popularity worldwide and the countless users relying on it.
In the March 2024 United Nations’ World Water Development Report (WWDR), it was stated that “as of 2022, 2.2 billion people do not have access to safely managed drinking water,” and that “approximately half of the global population experiences severe water shortages for some part of the year.” Artificial intelligence was created to enhance human lives. Is it justifiable for some individuals to live on the brink of life and death due to water scarcity while others unknowingly consume water with just a few clicks? Before AI poses a threat to human existence, we must find a responsible path forward.
How can we address the environmental challenges emerging in the age of AI, which we once thought would only bring convenience? First, AI companies must recognize the global water shortage and take social responsibility regarding the energy waste generated during their development processes. They must act with a clear understanding that the purpose of developing AI is to enhance human life. At the same time, ChatGPT users should be aware of the value of their interaction with AI. It has become common among university students to submit assignments using ChatGPT, as it is efficient, often yields higher quality work, and is likely to receive better grades than those requiring time and effort. However, misuse of ChatGPT can compromise students’ ability to acquire knowledge and promote a focus on quick problem-solving, which raises ethical concerns. Even worse, it contributes to environmental issues. What we once did for convenience may not only hinder our own learning but also lead to unnecessary water waste. Therefore, understanding the value of a single interaction with ChatGPT and avoiding its overuse could be the first step toward resolving environmental concerns.
We must remember that the use of generative AI, such as ChatGPT, comes with responsibilities. Moving forward, we should seek new ways to ensure the coexistence of AI development and environmental protection. The key to solving environmental problems starts not with someone else, but with us.
