New to Curatia?

See why finance counts on us for key market intelligence and industry trends.

Curatia Analysis for Tue, Feb 5

Why AI’s Weaknesses May Be Its Saving Grace

As fintech Davids and Wall Street Goliaths alike have dived deeper into AI projects, the heady early days of romanticizing AI’s ability to remake the financial landscape have given way to more sober media reflection on AI’s challenges and limitations.

That descent down the curve of AI’s innovation hype cycle could both help and hinder the progression of AI in the finance sphere, apprising firms of the technology’s common pitfalls but also discouraging investment in ambitious and potentially game-changing projects.

In the increasingly crowded world of ESG investing, fund managers are seeking to differentiate themselves by using AI to glean trading intelligence.

Scanning corporate filings to determine which companies have begun paying more attention to plastic pollution, for example, can yield insights on ESG-friendly firms as well as on the fortunes of plastics companies.

But the long-lamented lack of standardization around ESG factors also highlights a key weakness of AI: To facilitate comparisons, it requires data in a consistent format with few errors or omissions. Thin data sets can also cause hyper-analytical AI to view flukes as patterns.

In a finance sector where alt data is all the rage and historical financial data is often a hodge-podge of disparate data sets cobbled together as a result of mergers, that seemingly simple mandate can be a tall order.

Experts have also voiced concern over AI’s inability to provide transparency into its decision-making process and its potential to perpetuate biases lurking in historical data such as applications for mortgage loans and credit cards.

And while data must be extensive and robust, it must also be rare to be of value in the financial world. That paradox rules out the vast majority of data sets falling outside the Goldilocks zone.

The wealth of challenges around data quality in machine learning prompted MIT Sloan Management Review to contend that the machine learning race is really a data race.

Also challenging is the fact that the science of valuing companies has shifted from a focus on tangible assets (83% of markets in 1975) to harder-to-value intangible assets (84% of markets by 2015). That shifting dynamic has made obtaining data that bears on companies’ true value more difficult.

While such critiques of AI are hardly new, they have gained a greater slice of coverage in financial media of late. That trend has served in part to reassure financial professionals who worry AI poses a threat to their livelihood.

It also reflects as well as encourages a more skeptical approach to AI by executives. The percentage of executives who think AI is too opaque, for example, leaped from 29% in 2016 to 60% last year, per an IBM survey of 5,000 businesses.

Trending skepticism among execs owing to AI’s challenges and limitations may appear to be a stumbling block for the budding technology. But there are also reasons to believe it’s aiding AI adoption.

A PwC report released in December noted that AI initiatives led by CIOs “often fail to gain traction within the broader enterprise, with technologists focusing too much on IT capabilities and not enough on business needs or outcomes,” according to The Wall Street Journal.

Instead, AI-related projects fare better when ownership is shared between a firm’s business and technology arms, and when it’s delegated to groups with specific business needs rather than managed holistically from the top down.

Those findings illustrate businesses’ adoption of a more tempered, incrementalist approach to AI as they’ve gained a more nuanced understanding of not only the technology’s potential but also its weaknesses.

By fueling the technology’s application to real-world problems, a more tempered approach to AI could accelerate its widespread implementation. In that sense, AI’s weaknesses are saving it from the scathing rebuke critics have leveled at blockchain — namely, that it is a “solution in search of a problem.”

Ironically, AI incrementalism could also hinder bigger breakthroughs by curtailing investment in the types of visionary projects that, for now at least, remain central to AI’s identity.