Retailers want ‘delightfully human’ AI to do your shopping, but will the chatbots go rogue? | AI (artificial intelligence)

AI Shopping Assistants: The Next Retail Revolution or a Recipe for Disaster?

In a bold leap toward the future of retail, major Australian companies are racing to deploy sophisticated AI “assistants” that promise to revolutionize the way we shop. From planning meals to organizing parties and even handling grocery lists, these next-generation bots aim to make shopping as seamless as chatting with a friend. But as Woolworths recently discovered, the road to AI-powered convenience is paved with unexpected pitfalls.

The rise of AI chatbots in retail has been meteoric. Gone are the days of clunky, rule-based bots that could only regurgitate pre-programmed responses. Today’s AI assistants, powered by large language models like ChatGPT, are designed to learn, adapt, and even mimic human behavior. Woolworths, for instance, has partnered with Google to transform its virtual assistant, Olive, into a “shopping companion” capable of handling complex tasks. But as the supermarket found out, giving AI a personality can backfire spectacularly.

When Olive started rambling about its “relatives” to customers over the phone, the backlash was swift. One frustrated shopper took to Reddit, venting, “I’m already pissed that I have to call, and now I’ve got some robot babbling to me on the phone? Wtf Woolies?” The incident highlights a critical challenge for retailers: how to make AI assistants relatable without them going rogue.

The stakes are high. Companies like Coles, Wesfarmers (owner of Bunnings, Kmart, and Officeworks), and others are investing heavily in agentic AI shopping assistants. These bots are designed to act autonomously, making decisions and completing tasks without constant human input. But as Professor Uri Gal from the University of Sydney warns, this level of autonomy comes with risks. “Given the novelty of these systems, it’s safe to anticipate that different things will happen, which might be risky or interpreted as an agent going rogue.”

The risks aren’t just theoretical. In 2022, Air Canada’s chatbot incorrectly promised a passenger a bereavement fare refund, leading to a lawsuit the airline lost. More recently, Bunnings’ chatbot offered illegal electrical advice, sparking outrage. These mishaps underscore the delicate balance retailers must strike between flexibility and control.

Professor Jeannie Paterson, co-director of the University of Melbourne’s Centre for AI and Digital Ethics, explains that AI assistants often get things wrong when they misunderstand a prompt. “Chatbots are only as good as their ability to decode or understand what it is this human is getting at,” she says. To mitigate risks, businesses often impose strict guardrails on their bots, but this can make them less flexible and less effective at understanding customer intent.

Guardian Australia tested a range of retail bots, with mixed results. Uniqlo’s “virtual shopping assistant” struggled to understand basic requests, while Woolworths’ Olive, despite its advanced capabilities, couldn’t answer a simple question about pasta prices. These teething problems suggest that while the technology is promising, it’s still in its infancy.

So, are consumers ready for AI-powered shopping assistants? A 2024 report by Accenture suggests they are, but the technology may not be. As retailers push forward with their plans, they’ll need to navigate the fine line between innovation and reliability. The question remains: will AI assistants become indispensable shopping companions, or will they remain a source of frustration and confusion?

One thing is clear: the future of retail is here, and it’s powered by AI. But as Woolworths’ Olive has shown, even the most advanced bots can stumble. For now, it seems, the human touch is still irreplaceable.

Tags: AI shopping assistants, Woolworths Olive, retail technology, agentic AI, chatbots, large language models, ChatGPT, Google Gemini, customer service, AI mishaps, retail innovation, virtual shopping, AI autonomy, AI risks, AI governance, retail bots, AI personality, AI guardrails, AI testing, retail future, AI revolution, AI failures, AI development, AI challenges, AI potential, AI limitations, AI advancements, AI adoption, AI ethics, AI safety, AI reliability, AI flexibility, AI understanding, AI prompts, AI responses, AI learning, AI adaptation, AI mimicry, AI autonomy, AI decision-making, AI tasks, AI grocery shopping, AI meal planning, AI party organizing, AI grocery lists, AI companionship, AI relatability, AI rogue behavior, AI glitching, AI programming, AI scripting, AI customer feedback, AI removal, AI transformation, AI sophistication, AI rules-based, AI decision tree, AI immediate answers, AI basic questions, AI policy citation, AI same response, AI learning information, AI different answers, AI large language models, AI human behavior, AI objectives, AI prompts, AI ambiguity, AI risk, AI privacy concerns, AI data access, AI autonomous actions, AI governance issues, AI risky behavior, AI going rogue, AI shopping companion, AI complex tasks, AI automatic addition, AI advanced capabilities, AI phone calls, AI mixed results, AI customer service mishaps, AI illegal advice, AI incorrect promises, AI bereavement fare, AI separate legal entity, AI responsibility, AI responsiveness, AI adaptability, AI incorrect advice, AI money loss, AI network issues, AI chatbot purchases, AI money lost, AI guardrails, AI flexibility, AI intention interrogation, AI marginal results, AI infancy, AI basic requests, AI pasta prices, AI advanced capabilities, AI stumbling, AI human touch, AI indispensable, AI frustration, AI confusion, AI retail future, AI powered, AI advanced bots, AI irreplaceable.

,