A Clamor for Generative AI (Even If Something Else Works Better)

Since the debut of OpenAI’s ChatGPT in November 2022, generative AI—a form of AI that uses prompts to generate text and other content—has captivated corporate boardrooms and Wall Street, catapulting AI chip maker Nvidia to a trillion-dollar valuation.

Companies that mention generative AI in their earnings reports have seen bumps in their stock prices, fueling a race among firms to position themselves as front-runners in adopting the technology. 

But as a result, some CIOs say they are aware of pressure to shoehorn the technology into areas best addressed by either older forms of predictive AI, or even something as simple as a spreadsheet. 

“Generative AI is not a silver bullet for every single use case,” Sharma said. “In fact, in some cases it’s not as effective, and sometimes it’s dangerous.” Engineering generative AI solutions can be more complex, more expensive and more prone to hallucinations than other forms of AI, he added.

For example, Sharma said, for three years Mr. Cooper has been using nongenerative AI to spot whether information, such as a signature, stamp or notary date, is missing from a given document. The company has also tested whether generative AI might have solved the problem more effectively—and found that it couldn’t.

“Right now, we see no point in trying to reinvent the wheel just because GenAI can do it,” he said. 

Luke Gee, TD Bank’s head of analytics and AI, said he regularly fields inquiries and tips from business-side colleagues on generative AI. “Not day goes by where I don’t get a new link of a new article about something sent to me,” he said. 

When he does, Gee unpacks the problems with co-workers and helps determine whether generative AI is the right solution. “And maybe it is, and maybe it isn’t,” he said.

Generative AI is great for generating content, but of less compelling value for things like demand forecasting, anomaly detection, predictive maintenance and churn prediction, according to Sameer Maskey, founder and chief executive of enterprise AI company Fusemachines.

“Companies should first look at the problem and say, ‘What kind of algorithm makes sense?’” he said. If they don’t, CIOs might be simply placating higher-ups by using generative AI, and likely lose time and money, he added.

With a technology as hyped as generative AI, educating the workforce—from the C-Suite to the trenches—on its proper use and application can be a full-time job.

“The hype is taking all the oxygen out of the conversation and not allowing the appropriate attention on other types of AI models that can really generate value,” said Robert Blumofe, chief technology officer at Akamai Technologies. “The goal is not to solve the business problem. The goal is to adopt AI.” 

Blumofe said he’s trying to educate the rest of the company on what generative AI can actually do, where to use it—and where not to. Akamai, a cybersecurity, content delivery and cloud computing company, is using more conventional forms of nongenerative AI to identify anomalies in network traffic and flag cybersecurity incidents, and to recommend products to customers, he said.

These nongenerative algorithms have the advantage of being less expensive, less energy-guzzling and more accurate than nascent generative AI models, he said. That is in part because of their smaller size, he said. “Using GenAI can mean expending megawatts to solve problems that can be solved with milliwatts,” he said.

The company does use some forms of generative AI in its security portfolio. The problem, however, is the tendency today to throw a bunch of data into large generative AI models for even simple things, like calculating a median or the 95th percentile of a set of numbers, he said, questions easily answered by the humble spreadsheet.

Open communication with the board is critical, some CIOs say.

Jacky Wright, McKinsey’s chief technology and platform officer, said she’d advise CIOs who counsel boards to help them figure out whether core capabilities can be truly advanced by generative AI.

“If so, let’s dig in. If not, what are the things that run your business that we can focus on that would really help?” she summed up.

In those conversations, she suggests CIOs guide the board away from areas where generative AI might not be mature enough to deliver full value today, including AI agents that complete a full end-to-end workflow.

“So why focus on that when we can find higher value to go faster looking at something else? This is an ongoing dialogue,” she said. 

Eric Emerson, head of technology at Fidelity Wealth, said the financial services company has been using other, nongenerative forms of AI for almost a decade. But when it comes to some of the biggest problems the company is looking to solve, generative AI simply isn’t the answer right now.

“I do wish that AI could solve our mainframe modernization problem. That would be awesome,” he said. “But so far, the answer to that is no.”

Write to Isabelle Bousquette at isabelle.bousquette@wsj.com