AI’s Deep Dive into Green Data
More sophisticated use of the technology requires good data and risk management, according to Marsal Gavalda, Clarity AI’s Chief Technology Officer.
Responsible investors are becoming braver in their use of artificial intelligence (AI), increasingly using the technology to pose complex questions across multiple entities, looking for industry-wide trends and assessing how these fit into wider decarbonisation goals.
A recent survey of 150 asset managers across various classes, conducted by consultancy Mercer, showed that 91% currently use or are planning to use AI within their investment strategy or asset-class research. Nearly half of those already using AI (43%) are deploying capabilities to incorporate alternative datasets, filling data gaps across areas including sustainability factors and potentially improving insight into fundamental value.
Marsal Gavalda, Chief Technology Officer at technology provider Clarity AI, says that the way in which responsible investors deploy AI becoming more sophisticated.
“A few years ago, a human analyst would have had to carefully go through all the data, pull each data point up and come up with their own interpretation,” said Gavalda. “As AI gets better and investors become more used to what it can do, investors are increasingly using it for this more complex assessment.”
Gavalda, who is from Spain, moved to the US to study machine learning and speech recognition in the 1990s. He ended up staying, working at a number of technology companies that included Microsoft and a startup that was eventually acquired by Cisco.
Clarity AI provides a sustainability-focused platform for investors to interrogate in order to derive actionable insights that then feed into their decision-making processes. In February, the New York-based company launched a new tool to help investors monitor funds across an increasingly complex network of markets with different labelling and other regulatory requirements.
Examples of more complex requests that Clarity AI has been fielding in recent months include benchmarking companies against their peers, determining whether a company is on track to meet its decarbonisation targets and assessing which sectors are most at risk of non-compliance due to upcoming requirements under the Corporate Sustainability Reporting Directive (CSRD).
“More complex questions involve analysis, comparison, or interpretation across multiple entities or dimensions. These questions often require reasoning, synthesis, or applying specific methodologies,” said Gavalda.
They also require access to robust data and good risk management practices.
Data disclosures
An AI system is often only as good as the data that underpins it. With two of the world’s major markets – the US and the European Union – seeming intent on rolling back disclosure standards, good data for feeding AI algorithms may not flow as freely from mandatory disclosures as had been anticipated.
Gavalda said that, since its launch nearly eight years ago, Clarity AI has built up a wealth of data that it can draw on when calibrating AI models. If there is not sufficient information in the database to address a particular query, then Clarity AI’s platform will respond that it can’t provide an answer.
Since data is so integral to AI modelling, the firm is constantly trying to improve its datasets, in line with what customers are asking for. Gavalda remains optimistic that the ability to source high quality data is only going to increase, despite proposals which are expected to weaken disclosure regimes such as the EU’s CSRD.
“It is true that the legislation that the European Commission has proposed somehow diminishes scope, but I think the overarching trend is nonetheless towards greater disclosure and transparency, not only in Europe but also in Asia,” said Gavalda. “More companies are starting to think about how to better position themselves against their peers, because then investors are going to start ranking them. The CSRD changes are unfortunate, but overall the trend is positive.”
There are also many other data sources, beyond official disclosures, that Clarity AI has been looking to tap into. This includes satellite imagery data, which, when combined with other pieces of information, can help build up a more accurate view of a company’s footprint and impact.
Clarity AI also draws upon third-party AI systems, such as Open AI’s ChatGPT and Google’s technology. It has even taken China’s DeepSeek for a spin, to explore potential advantages in terms of speed and ease of use.
Making sure the database is as up-to-date as possible is another issue that AI platforms have to contend with. This can be a problem in cases where sustainability metrics, such as emissions data, only get published once or twice a year.
“We work very hard to make sure that the data in our database is as fresh as possible, and the time between a report being published and it being in our database is as short as possible,” said Gavalda.
Safety concerns
One of the biggest risks that users of AI models face is the risk that the technology gets things wrong. So-called ‘hallucinations’ can be hugely problematic for sustainability-focused investors, who may find themselves exposed to inappropriate industries or companies, or – worse – find themselves on the wrong side of regulators.
“We need to be very careful about hallucinations. The way that we avoid these is by making sure that the model provides the source data for any explanation,” said Gavalda.
For example, if a particular car manufacturer is trending well towards their decarbonisation objectives, then Clarity AI’s platform would typically include a chart with the actual emissions in any claim that it makes about this, along with detailed information about the company’s targets and clear indications of where this data comes from.
“Everyone is worried about AI models providing something that sounds plausible but is in actual fact incorrect. But this risk materially decreases with newer tools that force the model to back up each statement and number provided,” said Gavalda.
“We have a very high threshold for confidence. If we are not confident enough to respond, then we will say so. We need to have the traceability and the references to back up that response. This is far better than just giving a response that may not be totally accurate.”
How green is AI?
There is one other problem for investors who want to incorporate sophisticated AI technology into their investment strategies: does it align with their green credentials and objectives?
As Clarity AI Chief Sustainability Officer Lorenzo Saa pointed out in a recent commentary piece for ESG Investor, sophisticated AI systems can demand large amounts of electricity and water, potentially threatening progress towards climate goals. In developed economies, running data centres, a key component of AI infrastructure, accounts for 2-4% of national electricity consumption.
Gavalda sees this as a valid concern, but he says that much of it can be mitigated with careful planning and preparation.
“At Clarity AI, to minimise this consumption we use the smallest possible model for each of the tasks that we need to carry out,” said Gavalda. “We have a dedicated model for tabular ingestion, another for processing infographics, and so forth.”
By thinking about AI adoption in this way, sustainability-focused asset managers can continue to benefit from the technology while at the same time adhering to their responsible investment mandate.
“My personal take is that, overall, AI is a good thing, with the significant benefits outweighing the drawbacks of having to consume more energy,” said Gavalda.
The productivity advantages, he suggests, are only just beginning to be understood.
“These models are generic, so you train them once and they can be applied in many, many different use cases. The benefits will extend well beyond responsible investing. They will be felt in medicine, in education, right the way across society. There’s really going to be some huge societal benefits because of the use of AI in the years ahead.”
The post AI’s Deep Dive into Green Data appeared first on ESG Investor.