With ChatGPT reaching 100 million users within two months of its release, generative AI has become one of the hottest topics, as individuals and industries ponder its benefits and ramifications. This has been further spurred by the fact that ChatGPT has inspired a slew of new generative AI projects across industries, including in the financial services ecosystem. Recently, it was reported that JPMorgan Chase is developing a ChatGPT-like software service to be used by its customers.
On the flipside, as new stories about generative AI tools and applications spread, so do conversations about the potential risks of AI. On May 30, the Center for AI Safety released a statement — signed by over 400 AI scientists and notable leaders, including Bill Gates, OpenAI Chief Executive Sam Altman and “the godfather of AI,” Geoffrey Hinton— voicing concerns about serious potential risks.
Finastra has been closely following developments in AI for many years, and our team is optimistic about what the future holds — particularly for the application of this technology in financial services. Indeed, at Finastra, AI-related efforts are widespread, touching areas from financial product recommendations to mortgage process document summaries and more.
However, while there is good to come from AI, bank leaders — responsible for keeping customers’ money safe, a job they do not take lightly— must also have a clear picture of what sets tools like ChatGPT apart from past chatbot offerings, initial use cases for generative AI for financial institutions and the risks that can come with artificial intelligence, particularly as the technology continues to advance rapidly.
Not your grandma’s chatbots
AI is no stranger to financial services, with artificial intelligence already deployed in functions such as customer interaction, fraud detection and analysis well before the release of ChatGPT.
However, in contrast to today’s large language models (LLM), previous financial services chatbots were archaic — far simpler and more rules-based than the likes of ChatGPT. In response to an inquiry, these previous iterations would essentially look to find a similar question and, if such a question was not registered, they would return an irrelevant answer, an experience many of us have no doubt had.
It takes a much larger language model to understand the semantics of what a person is asking and then provide a useful response. ChatGPT and its peers excel in domain experience with a human-like ability to discuss topics. Massive bots like these are heavily trained to provide a far more seamless experience to users than previous offerings.
Potential use cases
With a better understanding of how new generative AI tools differ from what has come before, bank leaders next need to understand potential use cases for these innovations in their own work. Applications will no doubt expand exponentially as the technology develops further, but initial use cases include:
Case workloads: These documents can be hundreds of pages long and often take at least three days for a person to review manually. With AI technology, this is reduced to seconds. Furthermore, as this technology evolves, AI models may develop such that they not only review but actually create documents after having been trained to generate them with all their necessary needs and concepts baked in.
Administrative work: Tools like ChatGPT can save bank employees meaningful time by taking over tasks like curating and answering emails and supporting tickets that come in.
Domain expertise: To provide an example here, many questions tend to arise for consumers in the home mortgage market process who may not understand all of the complex terms in applications and forms. Advanced chatbots can be integrated into the customer’s digital experience to answer questions in real time.
While this technology has many exciting potential use cases, so much is still unknown. Many of Finastra’s customers, whose job it is to be risk-conscious, have questions about the risks AI presents. And indeed, many in the financial services industry are already moving to restrict use of ChatGPT among employees. Based on our experience as a provider to banks, Finastra is focused on a number of key risks bank leaders should know about.
Data integrity is table stakes in financial services. Customers trust their banks to keep their personal data safe. However, at this stage, it’s not clear what ChatGPT does with the data it receives. This begs the even more concerning question: Could ChatGPT generate a response that shares sensitive customer data? With the old-style chatbots, questions and answers are predefined, governing what’s being returned. But what is asked and returned with new LLMs may prove difficult to control. This is a top consideration bank leaders must weigh and keep a close pulse on.
Ensuring fairness and lack of bias is another critical consideration. Bias in AI is a well-known problem in financial services. If bias exists in historical data, it will taint AI solutions. Data scientists in the financial industry and beyond must continue to explore and understand the data at hand and seek out any bias. Finastra and its customers have been working and developing products to counteract bias for years. Knowing how important this is to the industry, Finastra actually named Bloinx, a decentralized application designed to build an unbiased fintech future, as the winner of our 2021 hackathon.
The path forward
Balancing innovation and regulation is not a new dance for financial services. The AI revolution is here and, as with past innovations, the industry will continue to evaluate this technology as it evolves to consider applications to benefit customers — with an eye always on client safety.
Adam Lieberman, head of artificial intelligence & machine learning, Finastra