A response to RI: Not all AI is created equal, but when it comes to ESG scores, it’s here to stay

Earlier this week, RI looked at why coal burner Adani Power scored so well on ESG. Thomas Kuh shares his thoughts on why algorithmic approaches are not to blame.

This article is Free, but to access more of our content, you can sign up for a no strings attached 28-day free trial here.

I read with interest Paul Verney’s Missing the Bigger Picture? How a coal giant scored better on ESG than a renewables firm (and made the FTSE4Good). As a long-time practitioner, I understand Ulf Erlandsson’s concerns about ESG research.

In his paper, Mr. Erlandsson cites two reasons to explain how a company like Adani could get a top-decile ESG score and qualify for an ESG index.

1.      “Big data or algorithmic approaches [that] could be subject to both stratification issues as well as reverse engineering/spoofing.”

2.      “Coal exclusion lists [that] have the requirement that revenues from coal production, as opposed to coal consumption, must not exceed 25% of company revenues.”

My response focuses on the first point. Regarding the second point – suffice it to say that screens based on “% of revenue from production” may miss the critical context of value chains and may be undermined by complex corporate structures.

But Adani Power is only the case in point here. ‘The bigger picture’ is that some of the self-designated arbiters of which companies meet ‘ESG standards’ got it way wrong on the company, revealing fundamental flaws in the prevailing paradigm for ESG research. ESG ratings are often inaccurate – sometimes with little consequence, sometimes grievously, as in this example. The question is why.

The application of AI to big data in the context of ESG analysis is both inevitable and imperative. The sheer growth and volume of ESG-relevant information necessitates the use of technology, and the opportunity to discern new insights is essential to progress in the sustainable investing. Of course, skepticism is warranted and deficient algorithms should always be a concern, especially as natural language processing and machine learning are increasingly prevalent in ESG analysis.

The real culprits in this case are traditional ESG ratings that rely on company reported data and ‘consensus’ models built on them. Like any analytical undertaking, ESG ratings risk being ‘garbage in, garbage out’ exercises. One would expect that consensus ratings based on diverse, and often incompatible, frameworks would be noisy. Without casting aspersion on motives, it is clear that unaudited data from companies may be biased and are not a good foundation for reasonably objective perspectives on companies. In the last decade, that’s most of what we had to go on. But not today.

Technology offers the chance to develop timely, transparent, decision useful data. Our firm’s data are drawn from more than 115,000 sources in 13 languages and reported daily. The information on Adani Power provides perspective on the company’s complex structure, including its coal power generation and renewables. So the problem is not algorithms, per se. In fact, well designed algorithms are a crucial part of the solution.

So, let’s not throw out the baby with the bathwater. The reality is that technology will profoundly shape the future of ESG research and investing. Properly deployed, AI will enable humans to make more intelligent decisions. For decades, those of us in ESG research forged ahead with the best tools and data available at the time, while acknowledging shortcomings. Our task now is to embrace technology and put it in the service of making ESG data and ratings more accurate, current, transparent and impactful.

Thomas Kuh is Head of Index at Truvalue Labs and former head of the ESG indices at MSCI. For more on this issue, see here.