Organizations are becoming more aware of the potential value of artificial intelligence and machine learning (AI/ML). But since widespread adoption of the latest AI/ML technologies is still in early days, you don’t often hear about how business and IT leaders are building strategies, scaling adoption, learning and evolving, or using their data to measure success.
At this point in the technology’s evolution, bringing these milestones to light and sharing best practices can benefit the business community at large since responsible AI/ML has the potential to unearth tremendous new opportunities.
In this Argyle virtual event, Eightfold Chief Product Officer Sachit Kamat joined a panel of industry leaders to share how organizations are using AI/ML to drive new opportunities and make the most of their data.
Here are four key takeaways from this panel discussion:
RELATED CONTENT: Welcome to the future of talent planning using a skills-based approach.
Identifying the business value of AI/ML
Many organizations are just beginning to realize the business value of AI/ML. The majority of organizations are using AI to automate tasks that machines can do better, more accurately, and faster than people. For example, our customers automate the talent acquisition process by taking the manual work out of sourcing, screening, and scheduling to speed up the process.
The broader business value of AI/ML depends on where organizations are on their journey toward adoption. Like with any new technology, in the early stages many organizations and IT departments see the technology itself as providing the bottom line value.
Panelist Matthew Versaggi, Senior Director of AI and Cognitive Technology for Optum Technology, used the analogy of the smartphone when it comes to adoption: “Most people aren’t concerned about the technology of the phone or what’s on the phone, they just care that it works,” he said.
The further along organizations move in the adoption process is where scale comes into play. As AI/ML is integrated into more business processes the value begins to reveal itself. Infrastructure becomes important here to support scaling. Then as the organization incorporates AI into the entire business and customer journey it permeates the rest of the organization including HR, finance, and marketing.
Developing data discipline to fuel AI/ML innovation
At the heart of AI/ML is data, and the importance of data strategies in using the technology. There could be a volume of data sources built into an AI solution of any type. Data sources can be streamed, batched, and then used for ML training.
Kamat shared an example from his former employer, Uber, of how data can be used to find innovative solutions to business challenges.
“During the pandemic, Uber, which is in the business of moving people from point A to point B, had its core business come to a literal halt,” Kamat said. “Despite the fact that the company shifted in response to try and quickly grow Uber Eats, it wasn’t able to figure out how to navigate the people movement within the company to support this new initiative. This is a great example of how having the right data on employee capabilities and skills could have prevented layoffs in trying to fuel the new business.”
Data and AI can help organizations navigate these types of changes at scale — especially when it comes to helping an organization launch a new line of business, enter a new industry, or move employees around. There is a convergence that’s happening where data maturity and the adoption of AI/ML enables organizations to withstand — and even foresee — rapid shifts and changes in economic cycles that could impact the business.
The challenge is ensuring that the data is accurate and putting proper controls in place. Panelist Balaji Veeramani, Director, Cloud Data Platform Engineering & ML/AI Ops for Union Pacific Railroad, talked about whether this should be a top-down or bottom-up approach in terms of creating a data-driven culture.
“Every organization across the globe is fully aware that data is the key cornerstone,” Veeramani said. “If data is not there, and you don’t organize the data in a way that is fundamental, we can’t even talk about ML and AI. Data is the fuel for whatever we want to do in this space. It’s both a top-down and bottom-up approach. Many questions need to be answered from both levels: What are the new cloud technologies? Do our people have the skills for these new technologies? How do we curate this data for the next level? How can we make a path for new ML/AI use cases? And so on.”
Managing the skills gap with AI/ML initiatives
One of the challenges the panel addressed is ensuring that the people initiating and scaling AI/ML have the necessary skills and competencies to execute on them. Because of the fast pace of change, organizations have to create and roll out aggressive reskilling and upskilling programs that sometimes are not able to keep up.
“Thinking about AI/ML initiatives from a very basic point of view, every time a manual process becomes automated, it starts to emit data,” Kamat said. “The more data you collect, the more complex management and analysis becomes. So you may solve one problem, but then you realize that there’s a whole other set of problems that you need to solve. Companies need to look at the skills required to solve these problems as they continue to evolve.”
The acceleration of technology advances with AI/ML can cause fatigue, fear, or even uncertainty at the leadership level on how to best adopt the technology and enable their workforces with the right skills at the right time. The panelists shared what they are seeing organizations do to fill these skill gaps, including identifying the skills that are needed for the organization to be successful with an AI/ML initiative, and hiring people with those skill sets or developing people for a specific set of skills.
Overall, a move toward a skill-based approach to talent planning will be necessary for organizations to not only bridge the gaps, but to also keep talent up with the pace of change in the AI/ML space.
Mitigating AI-related risks
A hot topic area around AI/ML is the potential for unforeseen risks. It’s no surprise that any new technology causes a level of disruption to the way things were done before, but as more leaders take a look at the business value of AI/ML, mapping out the potential risks is a necessary strategy.
A potential area for risk is bias in AI/ML algorithms. One of ML’s flaws is that it can sometimes mimic the decisions made by people. If the data being used to train AI contains decisions made with bias built into it, then ML becomes a tool to amplify that bias. For example, if an organization is hiring a particular candidate and happens to mention “golf” in the job description. Building ML algorithms using that information would favor a particular group of people and not consider others based on this data point.
Another potential area to be aware of involves anomalies in the data. Data is the foundation for AI/ML so ensuring that there is a plan in place to curate the data is critical.
“Start with what you are trying to accomplish with your solution, have a measurable KPI, and assume your data isn’t perfect,” said panelist Martin Miller, former Global Director of AI/ML Production Operations for Levi Strauss & Co. “Then ask, ‘What are we going to do with the anomalies in the data? How are we going to handle anomalous decision making?’ ”
Versaggi added that some organizations develop synthetic data to correct any anomalies.
“The assumption is that you have to get data from the outside world,” Versaggi said. “You have to go and observe some phenomena that you’re interested in, grab all the data you can, and then train an algorithm on that. That’s not true. You can start developing synthetic data to train an algorithm with the way you want that algorithm to behave. You can effectively artificially create it to cover all the use cases, and eliminate any concern with bias or anomalies.”
Another risk when using AI/ML is when algorithms don’t function as expected, or become irrelevant or counterproductive. Perhaps the learning algorithm is not working as expected, or things have changed outside of its scope, and it doesn’t produce results. Like with any new technology, rigorous testing is important.
Veeramani emphasized the importance of supervised machine learning and continual check and balances. As more organizations begin to widely adopt AI/ML into all business functions, it’s important to keep up with trends, challenges, and opportunities.
“Start with supervised training and get the model to the point where it’s working using the Turing Test or some other rigor,” Veeramani said. “Then slowly move to unsupervised learning, see how the model is performing. Also keep updating it. So once we develop the model, it’s not like we are done with that. It is constantly evolving, and we are frequently learning along with the algorithm. Having said that, the checks and balances and maintaining the ops is as important as we go.”
Watch the full panel conversation with Eightfold Chief Product Officer Sachit Kamat and other industry leaders on “Data-Driven Business Value: AM/ML Opportunities and Challenges,” hosted by Argyle Executive Forum, now on demand.