Is Artificial Intelligence the Future of Financial Services?
August 11, 2019
Artificial intelligence, or technology that enables near-human levels of cognition, shows great promise for the financial services industry. Within capital markets, AI can be a means to produce better, faster, and more accurate predictions. But as a human-designed technology, it still ultimately suffers from some vulnerabilities: privacy concerns, biased or otherwise poor-quality input data sets, and unwarranted overreliance on the technology— to name a few.
The benefits, among them the potential to responsibly extend credit to millions of previously underserved individuals, however, are encouraging. Thus, the financial system is facing a turning point. By embracing AI, under the prudent and, hopefully, encouraging oversight of regulators, financial institutions will be able to unlock a whole new path towards stable and equitable growth. It is still too early to say for sure, but the benefits AI-enabled financial institutions could deliver to customers and other previously untapped sectors, will likely have multiplicative effects on the greater economy. Potential benefactors include traditionally underserved groups like women and minorities. AI credit-underwriting systems can incorporate dozens or even hundreds of variables, and are not inherently backwards-looking like traditional FICO ratings. Thus, they could help break the cycle of low credit scores for certain communities. By incorporating alternative data, it has been suggested that these systems could expand access to credit at a lower cost.
For example, the following image shows the extent to which geographic discrimination was codified in the 20th century. The Home Owners’ Loan Corporation rated neighborhoods, largely along racial lines, for suitability to receive mortgages. This practice eventually became known as redlining, and its effects are still evident today. Theoretically, AI loan and mortgage underwriting could help reverse this.
However, if data sets are not managed properly, or if certain types of data are used, AI algorithms are at significant risk of introducing bias.. For example, zip codes, a seemingly benign variable, are highly correlated to demographic information like race. If data like these are used to evaluate individuals, the credit-underwriting system is liable to perpetuate systemic inequalities. Furthermore, bias can also be introduced to these systems through unscrupulous programming, sometimes reflecting the personal prejudices of the team responsible for the system. But if the proper safeguards are put in place, the benefits can be brought to fruition.
As such, nearly half of US financial services executives believe that AI could greatly improve the customer experience. Automated “chatbots” like Bank of America’s Erica also have some of the same democratizing qualities as AI credit scoring. By providing low-cost and easy access to financial advice, this technology can expand the scope of consumer banking for more people. However, the question remains: are people willing to meaningfully interact with an automated system? As with most things, it is likely to require a gradual process of habituation and increasing personal experience to achieve significant impact.
Investment banks have been experimenting with the new technology as well. JPMorgan recently launched their Emerging Opportunities Engine to drive predictive investment recommendations, focused on equity capital markets. Early indications suggest that this system has been quite successful in correctly analyzing trades and making buy/sell predictions. More generally, machine learning, deep learning, and natural language processing, all subsets of artificial intelligence with more specialized capabilities, are being used to leverage valuable data and insights. For example, Bloomberg, a pioneer of the sentiment analysis technique, incorporated machine learning to flag news articles and social media trends to more accurately provide context to stock valuations.
The power of AI could also be harnessed by government agencies engaged in economic forecasting like the Federal Reserve. AI and machine learning are adept at dealing with variable correlation more nimbly than traditional programming. This means, essentially, that the system can eke out relationships and detect patterns between economic factors that otherwise would not be found through orthodox methods. Additionally, adaptive machine learning can respond far more quickly to changing economic conditions than traditional econometric models—easily recognizing economic turning points and discontinuities in macroeconomic data.
The chart below gives an indication of the sheer quantity of variables needed to assess the state of the economy. Within each subcategory, there could be dozens of measurable metrics, with an exponential amount of relationships between them. AI’s capacity for handling large data sets and for pattern detection could greatly improve the accuracy and flexibility of economic models that consider these elements.
While artificial intelligence is still in its infancy, along with machine learning and natural language processing, it is becoming increasingly clear that the financial industry will be witness to tremendous change as a result of this emergent technology. This new technology comes right at a time when individuals are demanding access to cheaper, more transparent, and more efficient financial services. In communities without access to adequate banking services, for those wanting to maximize their returns from their investments, and for the organizations most dependent on economic data analysis, AI represents a hopeful future.
Student Blog Disclaimer
The views expressed on the Student Blog are the author’s opinions and don’t necessarily represent the Wharton Public Policy Initiative’s strategies, recommendations, or opinions.