Microsoft Executive Urges AI Chatbots

To Learn the Art of Asking for Help

AI’s Limitations: Acknowledging Uncertainty

Microsoft Vice President Vik Singh believes generative AI tools could save companies significant time and money. However, for this to happen, the models must learn to admit when they don’t know the answer.

“To be honest, what’s missing today is a model raising its hand and saying, ‘I’m not sure. I need help,” Singh said during an AFP interview.

Generative AI tools like ChatGPT have been widely adopted by Microsoft, Google, and others. These systems produce text, code, and content on demand, often giving users the impression that they are always right. However, they still suffer from “hallucinations,” where they generate incorrect or fictional responses.

 

 

The Business Case for Honest AI

For Singh, who leads the Copilot team at Microsoft, this flaw is a serious issue. His clients cannot afford unpredictable or misleading AI behavior, especially in critical business operations.

Salesforce CEO Marc Benioff recently pointed out growing customer dissatisfaction with Microsoft’s Copilot, noting its tendency to ramble or misfire.

“Really smart people are trying to solve the challenge of making a chatbot admit it doesn’t know the answer and ask for help,” Singh emphasized.

 

 

Cost Savings and Customer Experience

According to Singh, there is real value in having a model that knows its limits. Even if a chatbot escalates half of its cases to a human, companies can still save substantial resources.

“One Microsoft client spends $8 for every new customer service inquiry. So, even partial automation delivers real savings,” Singh explained. “And customers benefit from faster responses.”

Singh joined Microsoft in January and took over the Copilot AI assistant project in July. The project focuses on automating services across online platforms, sales, and accounting. It’s a key part of Microsoft’s strategy to monetize generative AI while managing its high infrastructure costs.

 

 

Productivity Over Perfection

While AI firms once promised systems that would “uplift humanity,” Singh sees current applications as more pragmatic. “Right now, it’s about boosting productivity and generating profits,” he said.

Copilot helps salespeople by conducting research, allowing them to spend more time talking to customers. For instance, telecom firm Lumen reportedly saves $50 million annually by using Copilot.

Singh’s team is working to make Copilot more autonomous. One goal is to embed it directly into Microsoft’s software suite.

“Imagine you’re a sales rep,” Singh said. “The model can follow up with a customer after two weeks or even send an email on your behalf if pre-approved.”

 

 

We’re Just Getting Started

“We’re in the first inning,” Singh stated. “These are productivity-based applications, but they have huge upside.”

Still, this increased efficiency raises concerns about job loss. Leaders like K Krithivasan of Indian IT giant TCS have predicted that contact centers will soon be fully automated by generative AI.

However, Singh and other Silicon Valley leaders see AI as a way to unlock human creativity and create new types of jobs.

He recalled his 2008 experience at Yahoo, where AI was used to select homepage news stories. Initially, people feared layoffs. Instead, click-through rates increased, requiring more content and leading to additional hires.

“In the end, we needed to hire more editors,” Singh concluded.