Emerging software tools regularly challenge CIOs to weigh the desires of the business against the pitfalls of hasty adoption. Generative AI is just the latest technology that's almost too tempting for organizations to resist.
Companies have already seen a surge in employees using off-the-books generative AI tools, according to Productiv.
The software portfolio management company crowned ChatGPT the king of shadow IT in April, based on an analysis of billions of app usage data points across millions of SaaS licenses dating back to 2021.
“Shadow IT exists for a reason,” Chris Bedi, chief digital information officer at ServiceNow, said last week at the MIT Sloan CIO Symposium. “Somebody needs to drive the business forward and they don't have the tech to do it.”
Mojgan Lefebvre, EVP and chief technology and operations officer at Travelers, agreed, speaking on the panel. “You’ve got to be aware of what’s going on before you can get in front of it,” she said.
Travelers keeps shadow IT in check with an open yet prudent procurement process, ensuring that “as demand arises there is a path for it to get to the right people,” Lefebvre said.
Telegraphing a clear tech procurement policy across the enterprise is a step in the right direction. But a robust vetting process for new technologies is a must.
“We've set up a partnership between procurement, finance and our technology team,” Lefebvre said, adding that approval begins with a compelling business case and progresses through a vendor evaluation and cyber, compliance and financial reviews.
Cause for caution
CIOs are motivated to put guardrails in place. For all the impressive capabilities the technology exhibits, data-hungry large language models remain vulnerable to jailbreaks, according to a recent analysis by the U.K. AI Safety Institute.
Major tech providers, including Amazon, Google and Microsoft, agreed this week to a set of responsible AI guidelines, but are still in the early stages of defining a model safety risk threshold.
As the fog of uncertainty around the technology lifts, CIOs will have to grapple with adoption cost, workforce AI training and a chaotic vendor ecosphere rife with startups.
“I don't mean to diminish them, but the majority of companies getting funded right now are a UI on top of OpenAI,” Bedi said. “We’re just getting started on small language models and domain-specific language models that provide unique differentiation.”
A level of risk tolerance is part of the adoption process.
“With cloud, there’s so much maturity you can play with fewer vendors,” Lefebvre said. “With Gen AI, we don’t know who the winners are and right now I’m not ready [to] say we should just deal with one or two vendors.”
