Vik Singh, the executive responsible for Microsoft’s AI assistant, Copilot, has highlighted a crucial challenge in the development of AI chatbots: the need for these systems to ask for help when they are uncertain about an answer. As AI models like ChatGPT and Microsoft’s Copilot become more integrated into business operations, their tendency to “hallucinate” or generate incorrect information remains a significant issue. This problem is especially critical for corporate customers who rely on AI to assist with tasks like customer service, sales, and accounting, where accuracy is paramount.
Singh, who joined Microsoft in January and now oversees Copilot’s development, emphasized that AI models should be designed to admit when they don’t know something and seek human assistance. He believes that even a more humble AI that requests human input in 50 percent of cases could still be incredibly valuable, offering substantial cost savings and improving customer service efficiency. For instance, one Microsoft client currently spends $8 on average for a customer service representative to respond to each new request. By using AI to handle these requests—while asking for help when necessary—the company could save significant amounts of money and provide faster responses to customers.
The drive to make AI more reliable and autonomous is critical for Microsoft as it seeks to justify the massive investments being made in generative AI technology. Applications like Copilot are designed to boost productivity by automating mundane tasks, freeing up human workers to focus on more complex and creative work. For example, Copilot can assist salespeople by conducting research, allowing them to spend more time engaging with customers. Singh pointed out that Lumen, a telecom company, saves around $50 million annually by leveraging Copilot in this way.
Looking ahead, Singh’s team is working on integrating Copilot more deeply into Microsoft’s software ecosystem, aiming to make the AI assistant more autonomous. He envisions a scenario where Copilot can handle follow-up tasks, like sending emails on behalf of a sales representative, without requiring constant human oversight.
Despite the optimism surrounding AI’s potential to enhance productivity, there are concerns about job displacement. Some industry leaders, such as K Krithivasan of Indian IT giant TCS, have suggested that generative AI could lead to significant job losses, particularly in sectors like call centers. However, Singh is more optimistic. He argues that, rather than eliminating jobs, AI has the potential to create new ones by enabling humans to be more creative and productive.
Singh drew on his experience at Yahoo in 2008, where an AI system was introduced to optimize the selection of articles for the home page. While there were initial fears that this automation would lead to job losses, the opposite occurred. The AI-driven process increased the need for fresh content, resulting in the hiring of more editors to keep up with the demand. Singh believes a similar outcome could occur with AI applications like Copilot, where automation drives the need for new types of work.
Singh sees the current phase of AI development as just the “first inning” in a long game. While there are still challenges to overcome—such as ensuring AI models know when to ask for help—he is confident that these technologies will ultimately lead to greater productivity, new job opportunities, and significant benefits for businesses and employees alike.
