This article was originally published by George Lawton, Journalist and contributor at TechTarget. George attended the Deep Learning Summit in San Francisco as a member of the press.
As the use of AI across industries grows, some nontechnology enterprises like Aramark are hiring a chief AI officer to bring governance and common sense -- and value -- to the hype around AI. Other organizations like Kohl's, Square and eBay have strong AI teams, but have not elevated AI governance to the same level as a CIO position or chief human resources officer -- yet.
At the RE•WORK Deep Learning Summit in San Francisco, senior IT executives at these companies elaborated on their current strategy. All agreed that leading companies need strong executive support to take advantage of AI opportunities. But whether this is managed by a federated team or centralized office headed by a chief AI officer (CAIO) is up to individual boards.
One person who's convinced of the need for an AI executive in the C-suite is Pavan Arora, who was named chief AI officer at Aramark in May, following a three-year stint as chief data officer and director at IBM Watson.
"Part of the reason [for having an AI chief in the C-suite] is rebranding the power of a data," Arora told SearchCIO in an exclusive interview.
Other C-suite roles are focused on driving value for the company, but the data management role has long been associated with governance and compliance -- and not on the value data generates, Arora said. Even the relatively recent creation of the chief data officer has done little to rebrand the power of data, he contended, arguing that most people in this role are treating data management as an expense, not an asset.
"So, a chief AI officer is necessary, particularly for nontech companies that have that latent data they have not used yet," Arora said. "If you want to turn data off from an expense into an asset, you need a CAIO."
Aramark is a nontechnical business compared to internet startups; its 270,000-strong workforce delivers food to hotels, hospitals and prisons. Arora, who said he believes he is the first chief AI officer of a Fortune 500 company, said his position was created by board members when it became clear the company was sitting on an untapped goldmine of data.
One early project Arora worked on was to use AI to improve Aramark's food planning. The general managers of each operation are pretty good at guessing how many burgers to defrost for a day, but they also tended to overproduce, because no one wants that last piece of meatloaf. Arora's team created a forecasting algorithm that uses historical data and the calendar to figure out how many burgers to defrost. The result: a significant reduction in food wastage, he said.
"My goal is how do I bring the company to the next level and create shareholder value from a single view. The tooling now exists to make that data into a leverageable asset," Arora said.
Companies need somebody who understands whether the data is suited for a specific purpose, and somebody who can advise other C-level roles on the risks of implementation, or whether this pie is actually worth going after, or whether the ROI is justifiable. (Alan Lee, VP of Data Science Engineering, Kohl's)
In a panel discussion, Alan Lee, vice president of data science and data science engineering at Kohl's, said the department store chain has not yet created a chief artificial intelligence officer role. Lee's team is focused on improving e-commerce operations, working with the company's marketing, finance and merchandising teams.
But he said he believes the CAIO position will evolve similarly to that of the CTO role. Today, few doubt that companies need a CTO, but in the early days when technology was first coming to the forefront, the CTO did not have as large a role.
"Now we're starting to merge everything into this AI because, like it or not, systems are actually getting really good at predicting a large variety of things, including customer behavior," Lee said.
"Companies need somebody who understands whether the data is suited for a specific purpose, and somebody who can advise other C-level roles on the risks of implementation, or whether this pie is actually worth going after, or whether the ROI is justifiable," he added.
An important aspect of the chief AI officer role lies in level setting bad ideas before they propel the enterprise down the wrong path, Lee said. The chief AI officer has to be able to understand and explain the dangers and compliance issues of edge cases that might ruin the business.
Indeed, the person in charge has to understand the AI well enough to stay ahead of any unintended consequences. "You can't just let a computer make a decision without some kind of an intervention," Lee said.
Lee has heard of companies that have asked their data scientists to train models on fake data for compliance reasons. The resulting models did not work well in practice, but they were compliant.
What companies need is a senior executive -- whether the head of machine learning or chief data officer or chief AI officer -- who has the ear and respect of senior executives.
"It doesn't actually need to be a title in order for them to convey these risks to the executives who make the decision," Lee said.
Tech companies like Square already have a strong data science culture and have less need for a centralized office of AI.
"We have not found the need for a chief AI officer, although we have some members of our board, like Naveen Rao or Anna Patterson who are hardcore ML people and can help guide the overall strategy, but not in a day-to-day manner," said Marsal Gavalda, head of machine learning at Square.
The company is well known for its payment service, but also has several other business units each run by independent executives. In fact, Square doesn't have a CTO or chief product officer either.
"We have a very strong culture and some of the decisions about the tech stack are more of a bottom-up decision. That's also the approach that we have for adopting a machine learning mindset," Gavalda said.
At the same time, as they put AI and machine learning into production, there is a need for principles and practices and a common platform that all the different products run on. Individual product managers are responsible for publishing data into a common stream, which is accessible by other teams.
Gavalda's team helps ensure that this data is both useful and shared across the organization in a compliant way. They're also working on a unified semantic model of the data that helps improve the way data is shared and consumed across the different teams. He said the magic happens when there is alignment between the data science, engineering and design.
Another strategy explored by eBay is to create an AI steering committee that meets with the CEO and CTO once a month. The commerce giant has federated its AI governance across teams focused on AI infrastructure, natural language processing and computer vision, said Robinson Piramuthu, chief scientist for computer vision at eBay.
George Lawton asks: "How would your company benefit from having a chief AI officer?
Experts from these teams work with business managers on specific initiatives, like improving customer interaction or visual search.
The steering committee works with executives to debate different initiatives and make sense of the current state of AI tooling and eBay data sets that might be relevant to a project. This committee tracks the progress of advances in specialized AI hardware and improvements to deep learning frameworks. The hardware and software are changing so much, they need to find ways to be adaptable as part of eBay's AI infrastructure design, Piramuthu said.