Capabilities to understand and generate human language give GPT technologies a potential to improve sustainability efforts and help businesses achieve their environmental, social, and governance (ESG) goals. GPTs, as advanced natural language processing (NLP) models are in center of attention to achieve this goal.
Yet, its usage also leads to carbon emissions and since it is a novel technology, ESG and business leaders are not clear about how GPT models can support their sustainability goals.
This article will explore GPT models’ potential to improve carbon accounting and other ESG practices with limited carbon emissions.
Improving carbon accounting
GPT models can process detailed information in business documents such as invoices and utility bills to create detailed carbon footprint calculations at reduced cost. For instance;
- Transportation data captured in invoices can give insights in fuel consumption and CO2 emissions
- Analysis of electric use on electricity bills can underline areas of energy inefficiency.
This level of detailed carbon footprint calculation can lead to more sustainable practices in fight against climate change.
Real life example from sponsor:
Hypatos’ GPT models can analyze transportation, logistics and utility invoices. With this data, companies can
- automate their carbon accounting
- get insights into
- fuel use and route efficiency in transportation. These insights can guide companies to reduce their carbon footprint by optimizing transportation methods or reconfiguring their supply chains.
- their electricity usage to improve energy efficiency.
It is important to experiment with models like Hypatos’ products that offer alternatives for a sustainable future.
Identifying scope 3 risks via public data
Compared to scope 1 and 2, scope 3 emissions present different challenges. GPT models can help solve them.
Scope 3 emissions refer to all indirect downstream and upstream emissions that occur in a company’s value chain. For some examples, see: Figure 1
In its latest environmental report, Microsoft revealed that more than 96 percent of their emissions are in Scope 3.1 Apart from the emissions caused by use of products like XBox, they include emissions from
- supply chain
- lifecycle of their hardware and devices
- travel, and other indirect sources.
Scope 3 is also related to non-financial and societal risks that can in the end effect companies’ reputation and sustainability goals. Environmental and labor related abuses are part of this. For instance, McDonalds is accused of deforrestation of Amazon.2
Apple’s reputational damage due to its suppliers’ unethical labor practices offers another example. 3
Models such as ChatGPT can help solving this by analyzing vast amounts of publicly available data, such as news articles, social media posts, industry reports, and more. By using a GPT model, companies can also analyze data in other categories such as supplier, transportation, distribution and product use risk. As Microsoft puts forward, scope 3 is the ultimate decarbonization challenge; it necessitates the coevolution of best practices for businesses. These features can help businesses to take actions to mitigate scope 3.
Faster supply chain data analysis
Supply chains can be complex. The ability of ChatGPT to analyze these supply chains‘ data can identify areas of waste or inefficiency, providing valuable insights for companies to automate their supply-chain operations, minimize waste, and reduce their environmental impact. Through its use, companies can have better grasp on information on WMS activities such as the capacity of inventories, shipments, worker efficiency and more.4
ChatGPT can interpret codes that contain information about the company’s supply chain. To exemplify, we asked ChatGPT to create a Python code with a supply chain dataset (Figure 2). 5 We wanted it to analyze this very code. In this most basic example, the model can detect several patterns (Figure 3), which can reduce waste significantly. Developed models can contribute further.
Informing decision making and shaping policies
Natural language processing (NLP) capabilities of ChatGPT can support informed decision making. The ability to digest and interpret large number of texts from research papers, policy documents, corporate reports, and more, ChatGPT can provide summaries and insights. This ability is especially important for policymakers, who can use these insights to make well-thought-out policies to fight the climate problem.
Businesses and organizations could also employ large language models to analyze trends, predict future scenarios, and design effective strategies for achieving global sustainability targets. This can contribute to addressing the sustainability challenge worldwide.6
This technology can help companies and governments address this challenge and make the world a better place for us from an environmental standpoint
Paula Assis, IBM’s General Manager for EMEA 7
It is important to acknowledge that the use of language models like ChatGPT and Bloom also contribute to the very problem they’re tasked to solve: increasing carbon emissions.8.Training large language models can be an energy-intensive process. Unless the energy comes from a renewable source, it can result in high rate of emissions.
A 2019 study from the University of Massachusetts, Amherst, highlighted that training a single AI model can emit as much carbon as five cars over their lifetimes.9
Mitigating the environmental impact of AI models
Addressing this issue is not straightforward, but there are steps that can be taken:
First, improving the energy efficiency (e.g. water) of AI training processes is key.10
Second, shifting to renewable energy sources for feeding these processes can reduce carbon footprint. Google, for instance, has committed to running its entire operation, including its data centers, on carbon-free energy by 2030.11
Third, it is vital to have transparency about the environmental impact of AI models. OpenAI, the organization behind ChatGPT, has taken steps in this direction by revealing the energy consumption and carbon footprint of their models, encouraging other organizations to follow suit.
Fourth, fine-tuning. Businesses can choose to fine-tune large language models by training model on focused dataset to accomplish a task rather than building one from zero. This can help reduce energy consumption significantly in case of LLMs.
If you have further questions regarding the topic, reach out to us:
Find the Right Vendors
Share on LinkedIn
This post originally appeared on TechToday.