E-Handbook: Enterprise machine learning and AI: Use cases and challenges Article 4 of 4

FotolEdhar - Fotolia

Critical elements CIOs need to build an AI team

Top-notch recruiters, organizational structures that engender trust, and attention to ethics are key to ensuring your AI team succeeds.

CIOs looking to take advantage of advances in artificial intelligence often have a thicket of technical issues to get through. But this is just the starting point. CIOs also need to consider how to ensure their AI teams will deliver business results, said experts at the recent ReWork Deep Learning Summit in San Francisco.

"AI is not just a data science problem; it is a business problem," said Amy Gershkoff, former chief data officer at Ancestry.com LLC, a genealogy service.

Two areas in particular need attention from CIOs when building an AI team, according to AI experts at the event. The first is to build an organizational structure that engenders trust in AI across the enterprise and does not run afoul of regulatory requirements. The second, less obvious, prerequisite is having knowledgeable, top-notch AI recruiters.

Organizational alignment

A critical first step in developing the right AI organizational structure, in Gershkoff's view, is to place data scientists in departments aligned with their goals.

If the AI team's goal is to drive more revenues, for example, putting its data scientists in the office of the CFO may conflict with cost-cutting goals of the group. Instead, Gershkoff said it might make more sense to embed data scientists in the product management teams or marketing teams, where revenue growth is a high priority. "You need to situate them in a place where they are most effective."

Amy GershkoffAmy Gershkoff

Some organizations are starting to move to a hybrid model for distributing data scientists across different departments. They are embedded into a variety of departments including marketing, product engineering and finance. The AI team members throughout the enterprise then report to a chief data officer as well as to their department managers. This can help align data scientists' goals with each department. "The disadvantage is that they may feel like they have two bosses," Gershkoff said.

Select knowledgeable recruiters

Hiring good recruiters is often an afterthought in many AI efforts. As a result, the interview process is not as rigorous and the company actually appears less attractive to potential recruits. Data scientists and machine learning experts are in demand. "So, for organizations to compete on talent, they need recruiters who speak the data scientist's language," Gershkoff said.

For organizations to compete on talent, they need recruiters who speak the data scientist's language.
Amy Gershkoffformer chief data officer, Ancestry.com

Savvy AI candidates will ask about the company's infrastructure, tech stacks and the kinds of problems they will be working on. If the recruiter does not know the answer, it will be hard to get the best talent.

Get infrastructure ready

New data scientists will also expect the basic data science infrastructure to be in place. CIOs must create an architecture to support data warehousing and flexible data management methods. "This is the basic block and tackle that needs to happen before an organization hires machine learning experts," Gershkoff said.

In addition to that, CIOs need to deploy technologies that support data governance, compliance, privacy and security concerns. This will allow the data scientists to do great work while ensuring controls are in place.

Making sense of the black box

Manuel ProisslManuel Proissl

A good AI strategy also requires finding a balance between more accuracy and transparency. At their core, AI systems can automate many decision-making processes. But the AI algorithms often evolve from complex machine learning rules, which can be difficult to decipher. "It is a black box to some extent," said Manuel Proissl, senior quant and data scientist at Ernst & Young.

Even if these algorithms end up being more accurate than people, stakeholders will not always trust them if the logic is not transparent. Embedding AI into the enterprise requires looking beyond the data science to establish trust. This transparency is also important to address new regulatory requirements like the EU's General Data Protection Regulation that mandates enterprises must be able to provide meaningful information about the logic involved in decisions made by AI.

Create a culture of ethics

Rumman ChowdhuryRumman Chowdhury

An AI team that is applying advanced machine learning techniques to solve business problems might also be faced with making decisions that run counter to the enterprise's culture of ethics. CIOs will need to work with stakeholders across the organization to carefully define the shared values and create a culture of ethics. Pithy mottos won't necessarily align AI efforts with enterprise goals, said Rumman Chowdhury, global lead for responsible AI at Accenture PLC. "If you don't have a culture around any process, you cannot implement it."

For example, a social media enterprise or department might establish the goal of social engagement. Product managers may use AI to increase social engagement. It may be the case that angry people post more often on the site than others do. But optimizing for angry exchanges would be out of alignment with the enterprise's ultimate goal of connecting with people in a meaningful way. If product managers are so focused on getting a new product out the door, ethics may be the first thing to go. But this can lead to a situation that gives the company a black eye. "Good AI is invisible, but bad AI is noticeable," Chowdhury said. "The paradox of responsible AI is that when it works well, you don't notice it."

Dig Deeper on Digital transformation

Cloud Computing
Mobile Computing
Data Center
and ESG