Page 20 - Winter 2025 - 2.pub

integrating language models into their businesses.

Multimodal AI is notably being developed or is already in use
in sectors including:
• Healthcare: where medical imaging can be combined with
patient history and lab results to improve diagnosis and
treatment, for example, by systems developed by Merative
(formerly part of IBM Watson Health) and Bayer (see AI In
Healthcare: A Path To Long-Term Immunity? June 25, 2024)

• Automotive: where inputs, including from cameras, GPS,
and LiDAR, are being combined to improve autonomous
driving, emergency response, and navigation systems at
companies including Waymo and Li Auto.
• Finance: where multimodal AI is being applied to analyze
customers' phone queries to support contact center
employees' response and resolution activities.
the growth in AI are already being felt across a host of sectors-- • Retailing: where the combination of visual images, product
notably those with exposure to the surge in demand for data reviews, product information, customer interaction data,
center capacity, including real estate companies, the power and and warehouse data is used to optimize marketing
utilities sector, and midstream gas suppliers (see Data Centers: campaigns, inventory management, and minimize
Rapid Growth Creates Opportunities And Issues, Oct. 30, 2024). packaging and shipping costs. Amazon, for example, uses its
"Just Walk Out" multimodal AI to enable checkout-less
AI's increasing applications will also entangle the systems in
shopping at its physical stores.
which it operates, with the potential for effects to reach far
beyond the immediate users. That possibility is perhaps most Edge AI, which moves AI computation to end users' machines.
evident in the financial sector, where the adoption of AI Though not a new concept, we expect that the increasing use of
exacerbates the risk of operational disruption leading to small language models (SLMs) by companies will support the
systemic instability that could have wide-ranging implications scalability of edge AI. That shift offers numerous advantages,
(see Your Three Minutes In AI: Financial Systems Will Face New including, notably, a reduced risk of mass outages associated
Systemic Risks, Oct. 4, 2024). with failures at major infrastructure or technology providers--
known as centralization risk. The combination of SLMs with
Those risks could be mitigated through effective governance, crypto technologies should further mitigate centralization risk,
regulation, and decentralization (see section on "edge AI,"
below). Yet the pace of development in AI and its expanding use paving the way for more secure and decentralized computing
and data storage (see AI & DeFi: Can Crypto Innovations Offset
exposes such efforts to the challenges of containing not only
Artificial Intelligence Concentration Risks, Dec. 4, 2025). And
existing and potential risks but also to uncovering unrecognized because SLMs at the edge require less computational power,
risks (a variation on the "Rumsfeld matrix" of known knowns, they will also offer energy efficiency advantages compared to
known unknowns, and unknown unknowns). Predicting risk is data and resource-hungry LLMs (see Can AI become net positive
core to our efforts to understand the evolution of AI's effects on
for net-zero? Nov. 14, 2024). However, meaningful application
organizational credit quality and will continue to drive our
of scalable edge AI will have to overcome the sizable challenge
research (see Crypto and AI: Shaping The Future Of The of ensuring that nascent technologies make their way from the
Internet, Oct. 1, 2024).
lab to the field and are deployed in sufficient numbers to be
What we will be watching in 2025 impactful.
Over 80% of organizations forecast their AI workflows will
Causal AI, which can reason and make choices like humans do,
increase in the next two years, while about two-thirds expect
uses causal inference to fundamentally understand why
pressure to upgrade IT infrastructure (see Generative AI: From processes and decisions are made. This goes beyond the pattern
Hype To Value, Nov. 25, 2024). The breadth and nuance of AI's recognition and correlation that underpins much of the current
development, its effects (and potential), are evident in the slate
ability of AI (particularly discriminative AI and generative AI).
of subjects that we have tasked ourselves to explore over 2025.
We expect that capability will become increasingly important
They include: Multimodal AI, which facilitates more human-like with increased AI adoption and the corresponding demand for
applications of generative AI models by enabling the integration data (both real and synthetic), as a means to identify true
and processing of diverse inputs such as images, audio, text, causality from spurious correlations. It should also enable AI
and video. The technology has gained traction in 2024 after
models to deliver more reliable query responses, make more
OpenAI included the functionality in its models, Meta released a
informed decisions, and potentially aid in the discovery of
multimodal version of its LLaMA model, and NVIDIA launched
underlying causes--which could prove a powerful tool across
an open-source multimodal foundation model (NVLM 1.0). We sectors including healthcare and finance. Causal AI is generally
expect this technology will gradually gain relevance in industrial regarded as one of the missing links to create artificial general
applications, particularly for companies that are already
intelligence (which matches or outperforms human cognitive

Arkansas Community Banker | 20 | Winter 2025