On the 30th of November 2022, ChatGPT was publicly released. In the two years since, the internet has been awash with experts (and those who claim to be) extolling the revolutionary powers of Large Language Models (LLMs) for each sector. Sustainable investing has not been spared.
In an industry accustomed to quantitative analysis, where machine learning models were already being deployed in mature processes to capture and analyse data, the arrival of the ‘AI age’ was less a revolution and more an evolution. What will the next phase of that natural progression look like? Where, if we don’t depend on sensationalistic headlines, will the impact of AI really lie for sustainable investors?
AI applications that actually create impact
Quantifying ‘sustainability’ is a complex proposition. Reliant on data that has historically been imperfect or even inaccessible, and metrics that are typically less standardised and often qualitative, matching the precision found in traditional investments has been an ever-present challenge for sustainable investors, hence the protracted popularity of ESG scores despite being demonstrably uncorrelated between providers, and unhelpful in advanced investment decision making.
Quantitative, factual data, and the analyses that can be conducted on it, are the antidote to this challenge. However, accessing this data at scale presents an additional challenge. This application is the area in which LLMs have had an immediate and practical impact, proving themselves to be truly disruptive.
By understanding and reasoning about unstructured data, AI models can dramatically accelerate the collection of relevant metrics. They can gather sentiment data at scale across large investment universes. They can understand and classify fund prospectuses. Not least, they can pull disclosed metrics from unstructured company reports. And the applications don’t stop at quantitative data: amidst the cacophony of regulatory changes, LLMs can help researchers zero in on relevant and material updates. In short, AI’s greatest impact so far is in accelerating any task which previously required labour-intensive reading, understanding, summarising and extracting quantitative data from unstructured written documents, paving the way for the rapid curation of usable and varied datasets.
AI wins on scale, humans on strategy
Removing the historic barriers to data collection has led to an explosion in the claims made about data availability. Whereas before, measuring sustainability in any form was the challenge, now a new challenge emerges for investors: how to meaningfully use the vast amounts of data suddenly available to us. Further: how to distinguish between data for data’s sake and actionable insights.
For seasoned sustainable investors, traditional investment frameworks centred on risk and return are now expanding to include factorised sustainability data gathered through AI, unlocking a new, three-dimensional approach to investing. While AI excels at quantifying these factors, human expertise remains essential to interpret their relevance, validate their impact, and align them with investor objectives.
Challenges unique to AI in sustainable investing
Whereas previously, the primary challenge faced by sustainable investors lay in data accessibility, the volume of data made available through AI now presents a number of new challenges. A whole article could be written on this subject – for brevity, I have chosen three to highlight here.
Data quality remains a major issue. LLMs are vulnerable to ‘hallucinations’, meaning they can generate incorrect but, critically, plausible data. Traditional statistical QA methods, based on detecting outliers, are not sufficient when dealing with AI-extracted information. More inputs, more fine-tuning, more advanced risk-based modelling of data, and more targeted human intervention are needed to avoid new datasets being compromised by quality issues.
There is the challenge of model transparency and explainability. Many AI models, particularly deep learning-based approaches, are often seen as ‘black boxes’ due to their complexity. When taken beyond data curation and used in portfolio construction and investment decision making, this lack of transparency can jeopardise investor confidence. This is especially true in sustainable investing, where (perhaps previously burned) stakeholders require a clear understanding of how investment decisions are made and how they align with their goals. Building explainable AI models that can provide insights into their decision-making process and applying techniques like Shapley values to attribute importance to different factors, is crucial in overcoming this barrier.
There is also an elephant in the room. AI models, especially large ones, are extremely energy intensive. Gartner projects that 3.5% of the worldwide electricity demand will be used for AI by 2030, twice the power demand of France. And that’s before we consider the water usage required for cooling data centres. It is hardly surprising when we now see big tech companies furiously backpedalling on their climate commitments in the race to capture the AI market. Outside of the environmental scope, AI technology can also exacerbate societal disparities, potentially widening the gap between richer nations with access to cutting-edge tools and poorer nations without. For example, a 2023 report from Oxford Insights found large disparities in AI government strategy, data and infrastructure, and technology between high income and lower- and middle-income countries.
Any investor making use of AI and LLMs in sustainable investment decision making needs to be prepared to answer the question: does the benefit of using AI to enable the shift of capital towards a sustainable future outweigh the risks, and environmental and societal cost of doing so? I believe it can, but this is not a given.
Conclusion: a balanced perspective
AI has undoubtedly made significant strides in the sustainable investment space, particularly in scaling data collection and making previously inaccessible information available. Its value lies not in replacing traditional methods, but in complementing them. When AI-driven data extraction is combined with human expertise, the result is a powerful partnership capable of generating deep, actionable insights for more informed investment decisions.
AI also introduces new and unique challenges, not least in data quality and energy intensity, which are ignored at the peril of investors who truly want to align with a sustainable future.
AI is a great tool, but even the best tools are useless in the hands of bad craftsmen. Is AI the future of sustainable investing? Yes, but only for those who navigate the intersection of AI and human judgment with care.
David Forster
Head of Technology, Impact Cubed
