‘Evolution & Impact of AI in Credit Unions’: The Recap and Report

Credit union leaders looking to understand the science, basics, and benefits of artificial intelligence (AI) applications to the financial services industry recently received much-needed guidance from Filene Research Institute’s “Evolution and Impact of AI in Credit Unions” webinar (scroll down on that webpage for entire recap video).

Audience members joined Lamont Black — fellow at Filene, finance professor at DePaul University, and cryptocurrency expert — and John Best, CEO of Best Innovation Group, for a deep-dive into key points of AI, the broader context, AI applications for credit unions, risks, ethics, and governance.

The rise in popularity of ChatGPT (Generative Pre-Trained Transformer) has intensified interest in the widespread adoption and use of text-generative AI (TextGen AI), which uses algorithms trained with vast datasets that have the capability to automate complex tasks in business and industries. In addition to ChatGPT (by OpenAI), other GPT-AI “prompt” interface online tools/technologies include Gemini (by Google), Claude (by Anthropic), and Copilot (by Microsoft).

Filene’s webinar — along with the recent Text Generative AI and the Future for Credit Unions report and summary slides — examines TextGen AI technologies to separate fact from fiction amid the hype in a way that allows credit unions to analyze real challenges, known limitations, and promising opportunities. TextGen AI tools (such as ChatGPT and others) are large language models (LLMs) that have been trained and optimized to talk — not to think, reason, or even know things.

Recently, better retrieval augmented generation (RAG) and retrieval augmented prompting (RAP) technologies are helping fine-tune AI’s response on topics so it won’t “hallucinate” (make false connections to generate false responses).

For credit union leaders, the most promising future of TextGen AI is in ChatGPT-like technologies, not as an end-service but as a component for existing information technology systems. Credit unions must uphold member trust, and careless use of new technologies will come at the expense of their members’ data security and privacy.

‘Artificial Intelligence-Ness,’ Regressive Bias, and Promising Futures
Filene’s report goes into more depth on the following:

• Artificial intelligence-ness is a matter of human perception — an appearance of intelligence.
• The important point for credit unions to recognize is that GPT is a language model that has been trained and optimized to talk and not think, reason, or even know things.
• ChatGPT and closely related GPT-based tools should not be seen as a replacement for human decision-making nor as a knowledge repository.
• LLMs like GPT show a regressive bias: because they are trained using, necessarily, data from the past, LLMs parrot opinions and facts that lag behind the real world.
• “Arming” TextGen AIs like GPT by giving them access to tools — a simple financial calculator, a list of financial products, real-life stock prices, etc. — represents the biggest opportunity in this field.
• The most promising future of TextGen AI for credit unions is in ChatGPT-like technologies, not as an end service but as a component for existing IT systems.
• Credit unions using text AI services should expect information they input into their models to be shared by the AI service providers with unspecified third parties for business purposes.

“TextGen AI promises to be a revolution for just about every industry, and credit unions are not an exception,” the report states. “From the promise of fully automated service chatbots to self-running social media accounts that engage with your members 24/7 to hyper-customized financial advice for your customers that is given instantly at almost no cost… well, we are not there yet.”

TextGen AI is great at talking but really bad at knowing what to say. All the might of Google, Microsoft, or Meta, and all the online data that modern computers can handle, have not been able to change that fact. LLMs powering TextGen AI are just that, models of language, not of the world, let alone the complex environment of finance. The underlying characteristics of these models come with serious limitations, including outdated data sources, unreliable accuracy, and noticeable biases.

“And yet, the implications of a machine that communicates meaning by forming words and sentences in human languages are tremendous,” the report adds. “The transformations in everyday efficiency for credit unions can be enormous because tools like ChatGPT may ‘only’ talk, but they do it well, and they do it with a degree of creativity that resembles human standards.”

Brainstorming, Creativity, Repetitive Tasks, IT Integration, and More
The following are two ways that credit unions can implement TextGenAI tools and technologies now:

• The first — and easiest — TextGen AI transformation for credit unions to do is already here: Credit unions can use TextGen AI tools for creative brainstorming tasks. For example, GPT tools can be used to brainstorm ideas like marketing slogans, HR first drafts of job postings, IT experimenting with code generation, etc. Managers and employees are overwhelmingly already exposed to these technologies (so are their members). It is important to understand its capabilities.
• The second TextGen AI transformation is already starting. Because ChatGPT or other GPT or LLMs do not have to be isolated systems relying only on their own strengths, credit unions’ IT departments can integrate TextGen AIs in their technology systems and have them communicate with other applications like social media apps, business insights tools, or decision algorithms. Having ChatGPT estimate loan risk is both unethical and managerially suspect.

But having ChatGPT reach out to your expert systems, algorithms, and employees and then articulating the findings in an automated report could be the next frontier.

“The excitement is justified, but it should be tempered with skepticism and renewed attention to data privacy and security,” the report states. “Credit unions have the trust of their members to uphold. Credit unions regularly handle highly sensitive, personally identifiable information. Like any third-party service provider, the creators of TextGen AI technologies reserve certain rights, and careless use of the technologies can come at the expense of data privacy.”

The legal and policy environment is likely to change radically in the following months, but the rights to public data of tech giants or the copyright implications of AI-generated content are still unsettled territory.