The generative artificial intelligence feeding frenzy is just a few months old, but retail giant Walmart Corp. has been using the technology for years.

The company unveiled its plans to build a conversational AI platform more than three years ago and was experimenting with generative and other forms of machine learning years before that. Its Converse conversational shopping assistant debuted in 2020 and there are similar applications for use by in-store employees and buyers.

Last month the company announced the GenAI Playground, a tool that lets employees safely experiment with various generative AI models without the risks of data exposure or intellectual property compromise that bedevil public services.

The Playground is connected to the internet but doesn’t share internal data. “From an information security perspective this helps us sleep better at night because instead of having associates using public instances of ChatGPT or others we have a controlled environment where we get to see what goes in, what comes out, we can identify hallucinations and can work more collaboratively,” said Rob Duhart, Walmart’s deputy chief information security officer. Hallucinations occur when a generative system fabricates a response if it can’t find the correct answer.

Risk factors

There’s no question Walmart is all in on AI, but a company that grapples with nearly 20 lawsuits each day is necessarily sensitive to governance concerns. An internal AI governance steering committee keeps track of technology developments and greenlights projects that show promise.

“We hear about new potential partnerships, vendors, and use cases both to run fast and make sure that the good news is getting out across the company,” said Nuala O’Connor, senior vice president and chief counsel for digital citizenship and co-chair of the steering committee.

The group’s purpose isn’t to put the brakes on AI development but to champion promising use cases. The mission is to “go do, go explore, go find out, go think about what might be useful to you in your line of work,” she said.

Executives meet regularly to demonstrate what their groups are doing and share ideas. Subgroups focus on areas like data governance, digital identity, emerging technologies and ethical information management.

A matter of trust

At a high level, Walmart’s approach to AI is governed by transparency, fairness and trust, O’Connor said. “We have a duty to our associates and customers to be transparent and accountable in our use of these technologies,” she said. Failing to disclose that a chatbot is a computer rather than a human, for example, “diminishes trust and the goal of our team is to increase trust,” she said. “I’ve encouraged our teams to inform customers that the technology is present and give context about how it works so the customer can learn in real time.”

Any customer-facing uses of AI are subject to continuous pre- and post-deployment testing to protect against model drift, which occurs when the accuracy of predictions deteriorates over time because of weaknesses in the algorithm or data set. “The policy is iterative, the process is iterative and the entire program has to continue to be really close with the design, development and deployment teams,” O’Connor said.

Compliance with policies and governance standards is scrutinized with particular care when it comes to situations involving people. Like most corporate human resources departments Walmart uses automation to screen job candidates. Model drift in that scenario can have disastrous consequences if unintended bias is introduced.

Human factors

“The policy says we should not be biased against humans in any way,” O’Connor said. “If there are decisions being made about humans, my team works with the engineers to make sure that they have attempted to prevent bias.”

Any application affecting humans is screened during the development process and again before launch “and now we’re building an ongoing monitoring process as well,” she said. Bias prevention “is now part of the dialogue at all levels of the technology organization.”

Both O’Connor and Duhart said they’re confident Walmart has put the processes in place to ensure that AI is used responsibly and to protect against new threats such as machine-fed spear phishing attacks. Large language models haven’t changed the basics of good cybersecurity practices.

“We’ve been seeing machine learning and potentially AI-driven attack activity for years,” Duhart said. “This isn’t a new venue or a new avenue, just a new playing field for us to spar in and to perfect our craft.”

The retail giant sees almost unlimited uses for AI in both internal and customer-facing scenarios. O’Connor is particularly interested in the potential of AI to assist people with disabilities. “I’m very curious about how these technologies can help folks who are physically or neurologically diverse,” she said. “That helps our workforce as well as our customer base.”

Photo: Walmart