Genpact's Agentic AI Officer on why she dislikes 'human in the loop'

Jinsook Han, the company's chief strategy, corporate development, and global agentic AI officer, states that AI agents are what truly distinguish the technology as a significant advancement. While generative AI was groundbreaking, Han notes that its value was not fully realized, whereas agents can bring about substantial improvements for both the company and its clientele.

“From our client service and capabilities perspective, the previous sort of permutations of AI were very limiting,” she said. “Yes, there was analytics, digital transformation, AI, and ML. But it was not until GenAI started coming up and now, agentic AI—now it’s truly a gift for an industry like ours.”

How does Han's position contribute to the broader leadership framework within organizations? Furthermore, how does she prioritize strategy in the company's self-directed initiatives?

Han, who started at Genpact twelve months ago, sees her role as encompassing both broad and deep responsibilities, collaborating with company leaders on agentic AI while also managing the creation of its agentic AI offerings. 

One of her closest counterparts is chief technology and innovation officer Sanjeev Vohra, who defines the company’s architectural direction, making her both “his peer and customer.” Together, they also co-chair the Genpact Architectural Review Board, where they classify all of the company’s agentic solutions and review them for adherence to internal standards from design through the execution.

She also collaborates closely with Vidya Rao, the firm's chief information and transformation officer. Their joint efforts focus on implementing agentic AI initiatives internally, which involves using it as “client zero” to evaluate its capabilities. 

“So basically, I work with everyone in the organization to make sure it’s both horizontal and vertical. And if people say, ‘Jinsook, does this work with strategy. Is it okay?’ Let me actually take off my agentic AI hat and put on the strategy hat, and I can give you the same answer coming from the same person.”

The Pokémon approach to AI agents

Han's dual responsibilities for strategy and agentic AI effectively merge when she delves into the practicalities of developing and implementing the technology. This is precisely why she dislikes the term “human in the loop,”, a common expression used by executives to claim they aren't allowing AI technology to function autonomously without human oversight. 

“It makes it sound like, oh, you enter any human in any part of the loop, and it’s going to work out. Whereas, do you really need a human in this loop? When the human actually goes in to look at it, what is the human doing? Is the human reshaping, creating, proctoring? What type of a human do you need? How’s the human from yesterday to today different?” Han said, going on to list more strategic questions about how such human involvement should actually play out.

She'd rather discuss agent deployment using a Pokémon card analogy, a concept inspired by her son. She explained that managing a fleet of agents is similar to managing Pokémon, as each individual possesses unique strengths and weaknesses. Consequently, they're most effective in specific tasks and situations, and their combined use can yield varied results. This perspective adds a layer of complexity to the decision-making process.

“The reason why I use this in a metaphor is because you gotta know what the battle count of each agent is,” she said. “And it’s not static.”