Artificial intelligence is making the leap from the imagination of sci-fi authors and computer scientists into consumer and corporate life. But even as the theories become reality, the general idea about AI — the lens through which the technology is viewed — often carries with it misconceptions and faulty assumptions that, over time, could impede AI’s progress.
And that could end up costing financial institutions that don’t get it right.
In a new PYMNTS podcast, Karen Webster spoke with Brighterion CEO Akli Adjaoute about those misconceptions. In the backdrop of their discussion was research from PYMNTS and Brighterion about the significant gaps between the use of true AI — unlike machine learning, AI offers insights from unsupervised computer learning — and the perception of it.
That gap, indeed, is a factor that explains the relatively low use of (true) AI by banks and credit unions. According to that PYMNTS research report, The AI Gap: Perception Versus Reality in Payments and Banking Services — its findings come from interviews with executives at 200 American financial institutions — only 5.5 percent of FIs employ genuine AI systems.
Misconceptions vs. Progress
Misconceptions can hinder further progress.
For instance, one common AI misconception concerns the assumption that the data fed into the AI algorithm — data that enables the system to arrive at its own, unsupervised insights — must be “clean.” That term itself can mean a few different things, not the least of which is that only properly labeled data serves as proper fuel for AI systems, including those deployed by banks and credit unions to manage everything from fraud to optimizing resources allocated for collections.
In fact, financial institutions being financial institutions — those large, complex creatures that in many cases are the products of mergers and acquisitions — there is a dearth of labeled, “clean” data to go around. Think about it: How much data is really “clean?” How much data lacks typos and other relatively small mistakes?
“You cannot design something with only clean data,” Adjaoute told Webster. “Unless financial institutions free themselves from the requirement to use labeled data, they’ll never be able to put the power of true AI to work and reap the benefits from it.”
Solving for Now, Not Yesterday
One of the other problems with feeding labeled data to AI systems is that it perpetuates the myth that AI must live in a database of static, stale data that reflects the past, not the “now” when actions must be taken.
“Databases do not store intelligence,” Adjaoute told Webster during the discussion, which took place Wednesday (Jan. 30).
Adjaoute explained that among the main appeals of AI, if not the main appeal, is that the technology adapts to what is happening in the present. One of the biggest values of unsupervised learning and true AI is its ability to recognize patterns in the data in real time and deliver insights that can quickly help financial institutions make better decisions.
That said, AI should not be thought of as some sort of “black box,” Adjaoute said. “Transparency is extremely important.” And that’s not all: Despite mainstream views, AI is more than just a high-level algorithm, he noted. As well, people behind AI deployments and operations, including the executives in charge, need to retain a constant interest and curiosity about the technology. After all, failure can often be attributed (and this goes for any field) to people not asking the right questions and not addressing those misconceptions about data and data storage, along with data accessibility and the ability for an AI system to adopt, flex and scale.
With 94.5 percent of FIs absent from the true AI playing field, artificial intelligence still has a long way to go to reach mainstream – or anywhere even close – when it comes to banks and credit unions. Accelerating that progress is possible, Adjaoute said, once FIs get past the misconceptions.