Walmart is going all out on generative AI — but with guardrails

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

The company announced plans to build a conversational AI platform over three years ago, and has been experimenting with generative learning and other forms of machine learning for several years. Converse’s Converse Shopping His Assistant is debuting in his 2020, along with similar applications for use by in-store employees and buyers.

The company announced GenAI Playground last month. It’s a tool that allows employees to safely experiment with different generative AI models without the risk of data breaches or intellectual property infringements that plague public services.

Playgrounds are connected to the internet but do not share internal data. “From an information security perspective, instead of having public instances such as ChatGPT used by employees, there should be a controlled environment where they can see what comes in, what comes out, identify hallucinations, and work. This helps me sleep better at night because there is something there,” said Rob Duhart, deputy chief information security officer at Walmart. Hallucinations occur when a generative system fabricates a response when it fails to find the correct answer.

Risk factor

There’s no question that Walmart is fully committed to AI, but a company that deals with nearly 20 lawsuits every day is inevitably sensitive to governance issues. An internal AI governance steering committee tracks technology development and greenlights promising projects.

“We are hearing about potential new partnerships, vendors and use cases to move quickly and ensure good news spreads across the company. Chairman of the Steering Committee.

The purpose of this group is not to put the brakes on AI development, but to champion promising use cases. Her mission, she said, is to “go and explore, research and think about what works for you.”

Executives meet regularly to demonstrate the group’s activities and share ideas. Subgroups focus on areas such as data governance, digital identity, emerging technologies, and ethical information management.

trust issues

O’Connor said Walmart’s approach to AI is governed by high levels of transparency, fairness and trust. “We owe our employees and customers transparency and accountability regarding the use of these technologies,” she said. For example, not making it clear that a chatbot is a computer rather than a human “will reduce trust. Our team’s goal is to increase trust,” she said. “I have encouraged the team to let customers know this technology exists and provide context on how it works so they can learn in real time.”

Customer-facing AI usage is subject to ongoing pre- and post-deployment testing to guard against model drift, which occurs when weaknesses in algorithms or datasets cause predictions to become less accurate over time. becomes. “Policies are iterative, processes are iterative, and the entire program needs to remain closely aligned with the design, development, and deployment teams,” said O’Connor.

Adherence to policy and governance standards is scrutinized particularly carefully in situations involving people. Like most corporate HR departments, Walmart uses automation to screen job candidates. Model drift in that scenario can have disastrous consequences if unintended biases are introduced.

human factor

“The policy dictates that there should be no form of prejudice against humans,” O’Connor said. “When human decisions are made, my team works with engineers to ensure that attempts are made to prevent bias.”

Applications that affect humans are reviewed again during the development process and before launch, and “we are also building a continuous monitoring process right now,” she said. Anti-stigma “is now part of the conversation at all levels of technology organizations.”

O’Connor and Duhart said they believe Walmart is using AI responsibly and putting processes in place to protect against emerging threats such as machine-based spear-phishing attacks. Large language models haven’t changed the fundamentals of good cybersecurity practice.

“We have seen machine learning and potentially AI-driven campaigns for years,” Duhart said. “This is not a new venue or a new path, it’s just a new arena for us to sparr and hone our skills.”

The retail giant sees near limitless use of AI in both internal and customer-facing scenarios. A generative AI tool called Ask Claude, named after Walmart’s first corporate buyer, was recently launched to help buyers with repetitive tasks, conduct research, and advise on merchandising best practices. I was.

O’Connor is particularly interested in AI’s potential to help people with disabilities. “I am very interested in how these technologies can help people who are physically or neurologically diverse,” she said. “This helps our employee base as well as our customer base.”

Photo: Walmart