Key Points
Commitment to AGI research has been key to the rise of OpenAI and ChatGPT.
Although OpenAI claims its goal is to ensure AI benefits all of humanity, issues such as ChatGPT's carbon emissions and the use of low-wage labor through suppliers in Kenya have raised concerns.
ChatGPT: AI cannot replace human judgment and expertise in ESG disclosure.
ChatGPT: The most effective and sustainable way is for all stakeholders to share the cost of reducing Scope 3 emissions; Scope 3 emissions are not merely a responsibility shift.
If there's one topic that has recently become unavoidable in casual conversations, it's ChatGPT. On November 30th of last year, the U.S. artificial intelligence company OpenAI made headlines by launching ChatGPT, a chatbot based on a large language model. ChatGPT can understand users' questions and instructions, and assist in solving a wide variety of problems through dialogue. Upon its release, it caused a sensation: just five days after launch, it reached over one million users, and by January of this year, its monthly active users had surpassed 100 million. Some have even called ChatGPT's debut the "next iPhone moment" for the tech world.
What Sets ChatGPT Apart
It's not just ChatGPT as a product that leaves an impression but also the company behind it, OpenAI.
With ChatGPT's massive popularity, OpenAI has also entered the public eye.
Currently, OpenAI's relationship with tech giant Microsoft is under scrutiny. However, when it was first established, OpenAI was a non-profit organization, founded by figures like Tesla CEO Elon Musk, Sam Altman (then-president of Y Combinator), and notable Silicon Valley entrepreneurs and investors such as Peter Thiel.
What Was OpenAI's Founding Statement?
To promote the safe and beneficial use of AI, with social benefits as a priority. Today, debates over whether ChatGPT and AI technology will replace human jobs are endless, but OpenAI seemed to answer this question at its inception.
In its first blog post, OpenAI wrote: "We believe that AI should be an extension of human will, distributed as freely, safely, and widely as possible." On OpenAI's official company profile page, it prominently states, "Our mission is to ensure that general-purpose AI benefits all of humanity."
Unlike conventional AI, OpenAI has always emphasized its focus on Artificial General Intelligence (AGI). AGI refers to systems that have broad and general capabilities similar to humans, unlike the more narrowly focused AI we are familiar with, such as facial recognition technology. While these conventional AI systems require reconfiguration to maintain functionality when conditions change, AGI would have the broad capacity to adapt autonomously.
In ChatGPT, we can already glimpse OpenAI's grand vision. People use ChatGPT to organize literature, modify application code, or even to answer questions poetically. It has also been demonstrated to pass professional or academic exams with difficulty.
In an interview with Xinhua News Agency, Huang Tiejun, director of Beijing Zhiyuan Artificial Intelligence Research Institute, called ChatGPT's success a revolution for AI and the entire information industry.
Environmental and Social Risks of ChatGPT
OpenAI's founding story and mission seem inspiring, especially given the significant potential impact of AI on various industries, including finance and sustainable development. However, there are underlying concerns, rooted precisely in the reasons for ChatGPT's success in the current AI competition—extensive data-based model training. ChatGPT's neural network model contains 175 billion parameters, and the training used 300 billion words and 570GB of data.
For such a large data volume, the first concern is environmental impact, including the carbon emissions generated by ChatGPT's online operation.
A recent analysis estimated that ChatGPT generated over 814.61 metric tons of CO₂ equivalent during its first 60 days after launch on November 30, 2022. For comparison, China's per capita carbon emissions in 2019 were 7.3 metric tons, meaning that ChatGPT's emissions were equivalent to the yearly emissions of 111 Chinese individuals.
The analysis pointed out that the largest source of emissions came from the model training phase, which was based on the preceding GPT-3 model. GPT-3's model training produced 552 metric tons of CO₂ emissions, the highest among large language models. While Google's PaLM model training consumed more energy than GPT-3, the proportion of green electricity used for PaLM training was higher, resulting in only half of GPT-3's emissions.
Currently, ChatGPT is still in its early stages, and with further promotion and iterations, the energy required and resulting carbon emissions will rapidly grow. Therefore, carbon emissions management for data centers should become a key focus for OpenAI in the future.
Chart: Carbon Emissions from Training Large Language Models
Source: Ludvigsen, Mioying Research Institute
Additionally, a recent report revealed that in November 2021, OpenAI reached an agreement with an outsourcing company called Sama to complete data labeling for model training. Sama, an AI company headquartered in the U.S., employs workers in Kenya to complete data labeling for clients including Google, Meta, and Microsoft. According to Time magazine, Sama employees working on OpenAI projects had to view violent and graphic images, and they were paid as little as $1.32 to $2 per hour. Four Sama employees interviewed by Time expressed concerns for their livelihoods. OpenAI admitted that Sama employees were required to view harmful content to prevent such content from appearing in the data used to train OpenAI's models.
Since ChatGPT is essentially a chatbot, it requires reading user-submitted messages to function. This process carries the risk of data breaches. Reports have indicated that Amazon's corporate lawyers have warned employees against inputting any company-sensitive information into ChatGPT, as such information could be used in further iterations of ChatGPT's training data, and there have already been similar incidents.
Implications of ChatGPT for the ESG Sector
OpenAI's origin story and business model may not yet be fully sustainable, but tools like ChatGPT have the potential to inject new vitality into the global ESG/sustainability field. For instance, AI tools could help streamline ESG and financial information disclosures, improving efficiency.
When asked how ChatGPT and broader AI technology could help in this process, ChatGPT responded that it could provide information and guidance in accounting standards, financial analysis, and sustainable practices, as well as help analyze trends, patterns, and outliers in data. However, ChatGPT also "reminded" that AI cannot replace human judgment and expertise.
Chart: ChatGPT's Response to "How ChatGPT and Broader AI Technology Can Help in This Process"
Scope 3 GHG Emissions Debate
ESG and sustainable investing, after years of rapid growth, are now facing a unique period. As it is often said that "big data knows you better than you do," we asked ChatGPT a sharp question about Scope 3 GHG emissions.
Initially, we asked in Chinese, but ChatGPT didn't provide relevant answers, even inventing a concept called "Atmosphere Item Three." It was difficult to proceed with follow-up questions:
Chart: ChatGPT's Nonsensical Response to "What Are Scope 3 GHG Emissions?"
We speculated that this might be due to a lack of sufficient Chinese information on GHG emissions in ChatGPT's training data or possibly because ChatGPT's servers were overloaded at the time. So, we switched to English. This time, ChatGPT seemed to return to its comfort zone, giving the answers we were looking for.
Q: What are Scope 3 emissions?
A: (Omitted)
Q: Do you think it matters?
A: Yes, from a climate change and sustainability perspective, I believe Scope 3 emissions are extremely important.
Greenhouse gas emissions are the primary cause of climate change, as agreed upon by the scientific community. Therefore, it is crucial for all stakeholders, including all companies, to take action to reduce GHG emissions...
Q: Why is everyone struggling with Scope 3 emissions? What do you think?
A: I believe the main reasons include the complexity of managing value chain emissions, the lack of visibility and data for supply chain emissions, and the lack of sufficient incentives for companies to reduce Scope 3 emissions...
Q: Who do you think should bear the cost of reducing Scope 3 emissions—the ones who want it or the ones who do it?
A: The cost of reducing Scope 3 emissions should be shared by all stakeholders in the company's value chain, including the company itself, suppliers, and customers...
It is reasonable to think that the companies that benefit from emission reductions should bear the majority of the cost of this work. However, it is also important to recognize that the benefits of sustainable practices should be available to all stakeholders... Therefore, the most effective and sustainable way is for all stakeholders to share the cost of reducing Scope 3 emissions.
Q: Do you agree that Scope 3 emissions are a liability shift?
A: I would not classify Scope 3 emissions as a liability shift because liability implies legal responsibility for harm or damage. Although Scope 3 emissions have environmental and social impacts, they do not result in legal liability for the company generating the emissions.
I would classify Scope 3 emissions as a shared responsibility, requiring all stakeholders to cooperate and communicate to reduce the impact...
Chart: ChatGPT's Responses in English to the Above Questions
Despite its limitations, ChatGPT's emergence still gives us a sense that "the future is here."