AI Models Need to Admit Uncertainty: Microsoft VP

AFP/APP

San Francisco: Generative AI tools like Microsoft’s Copilot promise significant time and cost savings, but they still need to learn when to admit they don’t know something, according to Vik Singh, a Microsoft vice president.

Singh, who oversees the development of Copilot, emphasized that AI models currently lack the ability to recognize when they are uncertain and need human assistance. This is a critical issue as corporate clients cannot afford errors from their AI systems.

Despite the challenge, Singh believes that even if AI has to defer to humans in 50 percent of cases, it would still result in substantial savings. He cited an example where a client spends $8 per customer service query, implying that AI could save money and improve customer experience by providing quicker responses.

Singh, who joined Microsoft in January, highlighted that while generative AI tools are primarily focused on boosting productivity, they hold the potential for broader applications in the future. He argued that rather than replacing jobs, AI will likely enhance creativity and generate new employment opportunities, drawing on his past experiences at Yahoo where AI led to the hiring of more editors.

As AI technology continues to evolve, Singh acknowledges that we’re still in the early stages but remains optimistic about its potential to transform productivity and create new opportunities.