Gen AI and Carbon Accounting - Risks & Opportunities

Generative AI (GenAI) is undeniably compelling. How should I structure this presentation? What does this error mean? Give me a list of themes for my birthday party. When these prompts are fed into tools such as ChatGPT a remarkably human-like response is given back. In addition, no computer science degree or any coding knowledge is needed to use these tools. The low barrier of entry has catapulted generative AI into the limelight as a solution to many problems both individuals and businesses face. Tech giants are developing or deploying their proprietary LLMs[1] and smaller corporations may opt to use pre-built GenAI solutions.

At Carbon Responsible, we have been thinking about how we might integrate GenAI technologies into our client services.

Generative AI in Carbon Accounting

Carbon Accounting is the process of measuring, recording, and reporting the amount of greenhouse gas emissions associated with an organisation’s activities.

Incorporating AI into our tools is an exciting prospect. Some aspects of carbon accounting may lend themselves to efficient automation through GenAI, offering the potential for increased efficiency. Currently, we view GenAI to be a powerful aide, but perhaps not the full solution for a couple of reasons.

Sense Check?

While GenAI is a powerful tool, it is not without its limitations. There exists a risk of generating incorrect or false information, often referred to as ‘hallucinations’.  The risk of generating hallucinations can be reduced through ‘prompt engineering’[2], where we learn how to phrase inputs effectively.  However, to be sure all outputs are accurate, outputs would require human review. This presents a challenge when aiming for a fully automated GenAI-powered carbon accounting solution.

Unknown Unknowns

The quality of an emissions inventory depends entirely on the data set we have. If we lack a comprehensive understanding of an organisation’s operations, a tool designed to automatically report emissions cannot identify potential omissions (e.g., missing data from an office) or anomalies (e.g., unrealistically high numbers) in the dataset.

At Carbon Responsible, carbon accounting specialists develop a relationship with clients enabling specialists to spot unusual data and clarify issues concerning reporting boundaries. Furthermore, they can guide and educate clients to improve their reporting practices, expanding the reporting boundary year on year.

Data Security

Carbon accounting often involves sharing and processing sensitive company and personal information.  One specific use case we’ve explored is using GenAI to extract data from utility bills. However, data security concerns arise regarding how this data might be stored and even used to train the LLMs. Many organisations have even banned the internal use of GenAI tools over data security concerns[3].

The emergence of enterprise-grade solutions such as ChatGPT Enterprise does offer a promising solution[4].  However, these solutions will still have to be thoroughly evaluated for their accuracy.

All of this is to say that GenAI is certainly something we are actively evaluating as an additional tool in our toolkit. Expectations and excitement surrounding GenAI are currently very high, and while we continue to explore its potential, our approach will be a risk conscious one.

—————————

[1] These companies are creating ChatGPT alternatives – Tech Monitor

[2] https://en.wikipedia.org/wiki/Prompt_engineering

[3] Samsung Bans ChatGPT, Google Bard, Other Generative AI Use by Staff After Leak – Bloomberg

[4] OpenAI releasing version of ChatGPT for large businesses | Reuters