Industry’s take on the chief artificial intelligence officer role

In 2012, the Harvard Business Review famously declared the role of data scientist “the sexiest job of the 21st Century.”

In 2024, a similar claim could be made for the role of the chief artificial intelligence officer, or CAIO, especially as it relates to technology jobs. Over the past 24 months, companies from Amazon to Zendesk — and as different in the services they provide as Hinge and Tractor Supply Company — have hired or designated positions to oversee their AI efforts.

Dozens of federal agencies have quickly followed suit, naming CAIOs in response to President Joe Biden’s October 2023 executive order. The departments of Veterans Affairs, State and Energy and other agencies have installed CAIOs in the past year to oversee AI efforts and balance risk and innovation.

Yet the chief AI role originated in the private sector, so companies — especially those in tech — have had a little more time to flesh out the role. According to a survey by Foundry, one in four enterprise companies either have an AI chief or are seeking candidates to fill the position. How are industry leaders deploying AI chiefs and what recommendations do they have for their public sector counterparts?

Nextgov/FCW sat down with several experts to find out.

A successful chief AI officer needs to understand the business

Jared Coyle, Head of AI for SAP North America, said many organizations “immediately want to go out and hire a data scientist” to fill the chief AI role. Yet hiring skilled PHDs to architect amazing algorithms does not necessarily translate to business success. It’s easy to imagine, Coyle said, brilliant algorithms that sales divisions don’t know what to do with and thus don’t provide much of a return or value to the organization.   

“The truth is, what you actually need are people who understand the business and business process experts,” Coyle said.

Many organizations — and federal agencies — have hired their chief AI officers internally. Accenture Federal Services, for example, promoted Denise Zheng to be its first chief AI officer in April after she’d served as the company’s global generative AI and ecosystem lead.

“You need people in these roles who know how to operate within the organization,” Zheng said. “It isn’t just, ‘How many academic articles have you published on the topics of AI and machine learning?’”

Many chief AI officers thus far wear multiple hats. Zheng, for example, also serves as the lead for Accenture Federal Services’ Federal Data and AI practice, and many federal CAIOs are also chief information officers or chief data officers. But that might be an advantage.

“I actually think [being dual-hatted] enables them to be more effective in the early stage,” Zheng said. “They’re a known quantity in the organization, and they have relationships. They know what data and analytic capabilities have already been established and how they can evolve that to drive AI.”

While Zheng said organizations and agencies should always seek to attract and retain the best talent, “institutional knowledge and respect and getting [stuff] done is just as important.”

Finally, the government itself presents additional challenges for the chief AI role. Amy Jones, U.S. Public Sector AI lead at EY, said she’s seen examples where agencies hire talent from outside the government but that talent does not “know how government works.” Budgets are different, and priorities are mission-driven rather than profit-driven.

Establish and drive governance

Karen Dahut, chief executive officer of Google Public Sector, said it’s important to establish appropriate governance at the leadership level. Many organizations, she said, are federated and enable all sorts of different platforms and technologies.

“I think that what they do need is somebody that is driving the governance, the selection of the technologies, the way they’re going to approach it, and then let it flourish inside the organization,” Dahut said. “But somebody has to drive the governance and policy.”

Dahut added that CAIOs ought to report directly to senior leadership, rather than another C-suite peer, like the chief information officer.

“I do think the CAIO and the CIO have different fundamental risks and opportunities,” Dahut said. “A CIO is about providing broad-based technology at a reasonable cost that’s secure, and a CAIO is all about experimenting with new technologies and trying new things to affect the mission. A CAIO needs to be reporting to leadership in some way because, at the end of the day, you’re really going to be able to transform the mission through this technology.”

Coyle recommended an internal governance approach as well as an external one.

“You would think at various agencies and companies that it’d magically be there already, but it’s about actual buy-in at the leadership level,” Coyle said.

Balance AI innovation and risk

Jones used a racing analogy to describe how a chief AI officer needs to operate regarding two key areas: AI innovation and risk.  

In the analogy, the guardrails or gutters on the side of the road roughly translate to compliance and security, while the pavement — and road itself — is akin to the CIO-developed infrastructure. The AI chief is essentially driving the car.

“You’ve got to keep the car going as fast as possible, but stay on the road,” Jones said.

Jones added that the AI chief’s role is to remove the friction between innovation and risk. One of those early friction points, she said, is “taking things out of the sandbox and getting them into the enterprise.” In government, that’s even tougher to do with additional security requirements and standards.

“CAIOs will have to navigate big challenges, first to market and then in helping agencies balance controls,” Jones said.

In terms of risk management, Zheng said chief AI officers ought to safely experiment with AI tech through sandboxes — especially if some budget dollars come through.

“If you get new budget money, I think what’s most important is to create that safe and secure sandbox for experimentation,” Zheng said. “You have to be able to, in a contained way, experiment with models, test out models, build models and deploy models into applications. Otherwise, you’re just going to be waiting for ‘name-your-vendor’ to hopefully develop a solution that maybe you can deploy that fits your needs. And even then, there’s a lot of customization of those things that’s needed.”