Understanding (and using) ChatGPT in banking

By Ryan Jackson

Anyone paying attention to tech headlines over the past few months likely has been unable to avoid one topic: ChatGPT. Going within two months from zero to 100 million users should tell you all you need to know in terms of its buzz. Not surprisingly, headlines ranged from questioning whether ChatGPT should be banned or used at schools to how it allows anyone to write code to predicting the demise of Google. Though there is some truth to all these headlines, like most new technologies, the reality lies somewhere in the middle.

So what is ChatGPT? Released in November 2022 by artificial intelligence research laboratory OpenAI, ChatGPT is a chatbot that leverages machine learning on large language models to mimic conversational answers to complex questions. To simplify even more, a user asks ChatGPT a question and receives an answer based on what the technology has “learned” as if a human were answering it.

There are two unique features to highlight regarding ChatGPT. First, ChatGPT is a large language model, or LLM, meaning it is trained on massive amounts of data — in this case primarily text — which it uses to predict what word should come next in a sentence. According to Stanford University, GPT-3, which ChatGPT is built on, “has 175 billion parameters and was trained on 570 gigabytes of text. For comparison, its predecessor, GPT-2, was over 100 times smaller, at 1.5 billion parameters.” Second, it was trained using supervised learning and reinforcement learning. In both approaches, humans improve the model’s performance by providing feedback on the outputs. For the reinforcement approach, human trainers rank ChatGPT’s responses to questions and create reward models for better responses. Both of the factors enable ChatGPT to better understand meaning behind questions and to recognize mistakes.

AI-based chatbots have been leveraged for years in the customer service context, but the speed and clarity with which ChatGPT provides content in a relatively simplified user interface has been eye opening. For example, most users can use Google to research a given topic by searching via keywords, clicking on any number of links, reading articles, and then formulating an outline — ChatGPT does all this based on a natural language question and generates a summary within seconds.

For all the initial hype surrounding this technology, it is not all positive news. First, some of ChatGPT’s responses to questions are factually incorrect. This problem seems to be exacerbated by the fact the answers to queries are returned as if a real-life human were responding, which can give the user a false sense of security or accuracy. Second, early testers of Microsoft’s search engine Bing, which has integrated OpenAI’s product, have reported some bizarre interactions with the technology. Third, and probably most troubling, yet not surprising, is the opportunity for criminals to leverage ChatGPT. A highly referenced article by Check Point Research walks through how this technology could generate a phishing email and code in order to instigate a malicious attack.

ChatGPT has implications for the banking industry, and banks should start to familiarize themselves with it and similar chatbot technologies. Banks should begin with reviewing how these technologies fit within their current controls and review of third-party software. Due to this reason, several large banks — such as JPMorgan, Bank of America, Citigroup and Goldman Sachs — have restricted employees’ use of ChatGPT. In addition, banks should continue to monitor regulatory guidance as the technology has attracted the attention of a number of different government agencies. Banks should also be vigilant on the cybersecurity front and monitor for any increase in phishing attempts that could be exacerbated by the use of this technology.

Once banks are comfortable with leveraging these technologies, they can easily experiment with the products and evaluate potential future use. Early use cases banks could explore include research and summation (for example, understanding emerging technologies), idea formulation (developing an outline of topics), and building customer personas for marketing campaigns.

Of course, even with all the promise of ChatGPT, and similar conversational AI technologies, humans should be part of the loop and review the outputs that are produced. Not only is this technology becoming more robust by the day, but with news that Microsoft plans to build OpenAI and ChatGPT features into all its products, more people are likely to access and use these technologies. In the meantime though, let’s see directly from ChatGPT what it could do for banks:

Output of a ChatGPT conversation on banking.

Access resources from ABA’s Office of Innovation.


About Author