Generative AI threatens ESG initiatives & integrity of ESG reporting

WITH commercial confidentiality already clouding exact and independently audited answers on the power usage requirements of generative AI, the September launch of the latest OpenAI o1 “reasoning model for solving hard problems” makes an already difficult to parse ESG situation even less verifiable to my mind.

As generative AI and associated chatbots become increasingly ubiquitous so do the energy and climate costs. Indeed, since every time the latest models get asked a question they involve more extensive Chain of Thought (CoT) ‘reasoning’. One question the bots are definitely not answering is about generative AI aka what are the additional climate costs incurred here? Particularly when the annual (but discounted) server costs are estimated by the Information as $4 billion for OpenAI in 2024 to power ChatGPT and its underlying models.

Given the mystery that already surrounds the large energy-intensive deep neural networks that underlie the potential transformative effects of generative AI, how can we any longer trust – even good faith - ESG statements, measurements and metrics? Of course, not all ESG reporting is created equal. Let alone accurate, despite the Securities and Exchange Commission (SEC) reigning in the hyperbole with regulations to ensure greater veracity in ESG numbers and claims formally provided in company financial statements.

With scepticism already widely afoot in some quarters about the utility and efficacy of ESG, the last thing needed is the complicated intervention of increasingly powerful generative AI and its uneven hard to quantify environmental impacts. The black box that surrounds existing generative AI, its algorithms and CoT – along with the data farms and energy it requires – has as a bonus also badly muddied the already often cloudy waters of ESG (environmental social and governance) reporting. 

There is no dispute that large energy-intensive deep neural networks underlie even the most vanilla iterations of generative AI. In such a fast-moving context, failing to accurately quantify, represent or understand these environmental and climate impacts and inequalities represents significant trust and reporting problems for ESG. This is especially true if, as expected, these impacts escalate considerably due to increasing global AI energy demand. Until we agree the metrics for and measures of environmentally responsible generative AI, the ESG footprint of practically any company using generative AI are subject to question and likely scepticism.

Indeed, use of generative AI arguably represents a significant further obstacle to greater ESG financial reporting, never mind transparency and accuracy. The idea that we can salute and apply science in our workplaces when it comes to generative AI yet at the same time forget its currently unquantifiable ESG and climate crisis negative consequences is something of a contradictory position to maintain with a straight face.

Those already using generative AI do so – in my opinion – without being able to properly nor independently quantify its ESG and climate impacts against existing agreed standards. If nothing else, this stirs up ESG financial reporting exposure, doubt and risk. It is also a dereliction of governance towards the climate crisis that ESG is notionally supposed to help address.

Until reliable independently audited and verified information is available from generative AI companies about their environmental impacts and footprint, spurious commercial confidentiality will continue to further threaten the integrity and validity of existing ESG claims and financial reporting^ as well as call into serious question currently accepted standards, measurements and metrics.

Note

^ Industries and companies subject to US SEC financial disclosure obligations already face the need to understand, address and accurately report upon their ESG and climate crisis footprint with a stringency not yet applicable in the UK. The largest organisations here are currently required to disclose their Scope 1 and Scope 2 emissions in their annual reports. However, reporting on Scope 3 emissions are still mostly voluntary (and, before leaving office, the previous government was in the process of consulting on the needs of the future reporting framework). 

Photo credit: Getty Images/AI Business