It may seem counterintuitive, but for AI assistants to work well together, they don’t necessarily have to get along with each other, says Babak Hodjat, chief technology officer of AI at Cognizant.
The next stage of this technology, Hodjat said, will be a world in which different AI assistants, each an expert in their own field, work together to solve people’s everyday questions. . During a live interview at luckCOO Summit. (Disclosure: Cognizant sponsored a conference in Middleburg, Virginia last week and received coverage as part of the sponsorship deal.. )
However, if each of these assistants turns into a yes-man, consensus is built too quickly and users are at risk of getting the wrong information. That’s the current state of AI chatbots.
“They’re too good a match,” said Hodjat, who invented the natural language technology that contributed to Apple’s Siri. “You could say they’re too soft.”
Hodjat envisions a world where chatbots specialize in different workplace functions. For example, an HR AI agent that can answer questions about paid time off policies, a finance AI agent that can see the latest supplier contract prices, and a legal AI agent that can see the latest supplier contract prices. Answering your compliance questions as told by Terra Terwilliger, Director of Strategic Initiatives at Google DeepMind luck It is already being piloted at the COO Summit).
Hodjat gave a real-world example of how employees can use a network of AI assistants. When a life-changing event occurs, such as the death of a spouse, employees should consult their human resources department about changes to their benefits, their payroll department about changes to deductions from their paychecks, and in some cases may have legal obligations. Procedures will also be required. advice.
These AI agents work best through healthy discussions, much like thoughtful colleagues exchanging ideas to solve complex problems. But for now, AI agents are still unable to replicate human-specific dynamics.
“That’s one of the frustrations when you set up these systems in a multi-agent setup; you don’t want them to talk to each other, and in some cases argue with each other. [for] Something has to come from that,” Hodjat said.
There are good reasons why AI systems are programmed to be comfortable, especially in the very early stages of development.
“Right out of the box, these models are finely tuned to protect against bad behavior, bias, and weirdness,” Hodjat said.
Employees believe that workplace discussions, rather than arguments, improve collaboration and make them feel more energized at work. They also, believe it or not, higher job satisfaction Because employees feel as if they are allowed to express their opinions without fear of repercussions. Without positive interactions, companies risk becoming echo chambers.
“It’s harder to do that with today’s state of generative AI, because it’s so comfortable,” Hodjat said. “They quickly unite, but that’s not the reality we see on social networks.”
