Invisible AI data trails a clear ESG risk

An almost complete lack of transparency around sources of data for AI will continue to present a challenge to investors looking for exposure to AI themes or to invest in companies using AI in their business, according to global research house MSCI.

“There's a change happening in the marketplace for data which is going to have a fundamental effect on what AI model developers can do in future, because to build a good model, you need plentiful, high quality data that doesn't come with built in biases and you need permission to use it,” says Meggin Thwing Eastman, MSCI Global Head of Sustainable Finance Research.

“So if you fail on either of those, you're in trouble.”

Data owners have now realised they don’t have to give up their data for free, Eastman told an ESG Trends event hosted by the Responsible Investment Association Australasia, which means good quality data, and the permission to use it, is becoming harder to find.

But the major hurdle for investors, and those advising them, lies in disclosure - or the lack of - of data sources. MSCI looked at the seven largest AI model developers in the tech sector, and found two of them reveal nothing about where their data comes from, while five said their data is sourced from publicly available information, although it is not clear whether they have permission to use it, whether it is subject to copyright or privacy laws.

MSCI also looked at the major companies known for heavy data usage as part of their operations.

“Keep a close eye on the companies that aren't being upfront about what they're doing, because it might be that they're doing something a little dodgy, or it might be that they just don't actually know. And neither of those scenarios is great.”

“They're not missing a chance to talk about it on their earnings calls, but only about half of them have said what their AI-related policies are on things like privacy and copyright and bias and discrimination,” says Eastman.

“And only one out of the 50 actually promised to respect copyright law and as public disclosures - that was Adobe.”

Meggin Thwing Eastman says it is not just those outside the company in the dark about the source of AI data, with boards and executives not necessarily having the right expertise to know what questions to ask their teams who are focused squarely on staying ahead of the pack.

Investors need to query data sources themselves and follow the trail as best they can.

“Companies which own the data, or have exclusive access to high quality data sets, they're going to have an advantage, especially as we start to see regulatory developments where more transparency is going to be required in jurisdictions like the EU or California.

“Keep a close eye on the companies that aren't being upfront about what they're doing, because it might be that they're doing something a little dodgy, or it might be that they just don't actually know. And neither of those scenarios is great.”