By W.B. King
Among keynote speakers at Origence Lending Tech Live, which took place May 31 to June 1, 2023 at The Cosmopolitan Hotel in Las Vegas, was Hilary Mason, co-founder and CEO of Hidden Door. The event, only open to credit union attendees, was billed to deliver a unique conference experience that covers all aspects of lending — from innovation and technology to marketing and digital transformation.
Noting that artificial intelligence (AI) is often in the news today, Mason said it’s perplexing because one headline has AI as the “future of all business” and another headline has leading researchers saying that AI poses an “existential threat.”
Adding to this news cycle is the Writer’s Guild of America ongoing strike demanding that contract language includes AI regulations designed to protect writers and the works they create. SAG-AFTRA has also joined the strike, with actors demanding that AI-generated digital likeness is not used without their permission.
“We’re in this moment where technology has come to the point where it is useful and potentially able to do harm,” Mason said during her keynote address, “Innovating Through Data: The Future of AI and Machine Learning.”
Deep Learning
Referring to herself as an “AI graybeard,” Mason began teaching computer science at the college level 20 years ago with a focus on tech that could make decisions in real-time with limited amounts of data. Over the subsequent years, she worked with startup companies and financial institutions that were focused on social media and human behavioral data—analyzing billions of points of data.
She later co-founded the Brooklyn, N.Y.-based Hidden Door, a game technology studio that is building the “first narrative” AI, the company notes. The platform, as noted on its website, is designed to transform any work of fiction into an infinite social roleplaying experience, bringing together players, authors and other creators.
Machine learning, she explained, is data science modeled into feedback loops that “gets better over time,” noting that this analytics movement began picking up steam around 2010. This, in turn, has allowed many businesses, including credit unions, to more accurately predict the future wants and needs of membership. In 2014, AI was becoming the buzz word along with “deep learning,” both of which have evolved over time creating even more insights and actionable intelligence.
“So today in 2023, what is AI? It’s in every department, in every industry,” Mason said, adding that data is informed and supported through generative AI technology that benefits greatly from human interactions.
Invent User Experiences
When the topic of ChatGPT was offered, Mason explained that the open source data model aggregates words with associated data points existing on the internet, all prompted to quickly deliver outcomes in the manner and format the user seeks.
“What we have is an engine that aspires to be the most average idiot on the internet…based on the data it’s been trained on,” she said, adding that this information sourcing is getting smarter every day because reinforcement learning is layered on top of it. This allows [real] people to say an answer to a question was either “really good or really bad,” thus improving the final outcome. But the process is still evolving and is not quite accurate, yet.
In an effort to illustrate her point, she used Midjourney to produce an image of a credit union executive giving a presentation about data and AI and asked for it to be photorealistic. The San Francisco-based company offers a generative AI program that generates images from natural language descriptions, called “prompts,” similar to OpenAI’s DALL-E and Stable Diffusion.
Mason’s prompt resulted in four candidate pictures.
“I present to you what a credit union executive looks like in the world of data: He is very white, very dude and a little old,” she said. “This is magnifying the bias the underlying data is associated with.”
Not immune to the inaccuracies of AI, Mason has had issues with ChatGPT impersonating her. She was recently quoted in article that was posted on LinkedIn. The problem: she wasn’t interviewed.
“I never said the quote and it wasn’t a bad quote but it was just a weird choice of words,” she reflected.
For credit unions interested in adopting or enhancing AI and machine learning strategies, Mason said it’s a matter of getting a good sense of what exists in the aggregated data set and then “poke it” to pull out meaningful data that can build upon.
“On the application layer, I think we cannot be shy about saying it will largely disrupt and provide a set of automations in the production of creative systems, but also entitle new products and that’s something I’m really excited about,” she continued. “We are in the moment we are able to invent these [user] experiences.”
