Foundational models of generative AI not good enough for enterprises: Genpact CEO

Tyagarajan believes that the first step to regulating AI needs to be 'self-regulation' in every industry across the world

Genpact is working on 100 generative artificial intelligence (AI) proof of concepts (PoC) across 10-12 use cases, said CEO NV ‘Tiger’ Tyagarajan, coming a month after he announced that the Business Process Management (BPM) firm will be investing around $600 million over the next three years to build out its AI capabilities, during its Q2 2023 earnings conference.

Speaking to Moneycontrol, Tyagarajan said that these 10-12 specific use cases are being developed with the objective of repeating the same use case for multiple clients.

He added that though the company has been working on OpenAI’s large language model, GPT, one has to be careful as the LLM could give some wrong answers without the user being able to decipher it. The LLM needs to be fine-tuned further for enterprise use cases.

Tyagarajan explained, “The foundational models (of LLMs) are based on public information. But that's not good enough, if you want to tell a customer who has asked you what happened to the order I placed.”

“Public information doesn't help you. You use the technology that public information has helped build the language models on and then you fine tune it for the specific information like an interaction between a retailer and a food and beverage company,” he added.

Through its ongoing $600 million investment, Genpact will be “investing into its own innovation and R&D teams, client co-innovation programs, data, tech and AI skills training and creating deep expert groups as well as acquisitions focused on data analytics and IP and frameworks in the use of data models.”

According to Tyagarajan, AI will not be able to completely replace humans as he sees the technology as augmented intelligence with human in the loop.

Regulating AI

As the governments of countries like India and the US among others are gearing up to regulate AI, Tyagarajan believes that the first step needs to be “self-regulation in every industry across the world.” This can then become a template for the governments to build their regulations on.

“Historically, if you look at the best regulations, they've been when the industry has self-regulated and hopefully in AI, various players in the industry will come together to self-regulate. And this will be around data privacy, data security, explainability etc. It's about being able to have audit trails and ensuring that for critical decisions, there's always a human in the loop to avoid biases,” he said.

“It's about making sure that the data that is used to actually train the engine has no bias. It's about making sure that a particular country's citizen data doesn't leave the country,” he added.

Demand outlook

While Tyagarajan cautioned that inflation, high interest rates still continue and won’t go away anytime soon, customers are opting for more cost take out and efficiency projects, integrating its with AI and generative AI use cases.

“Digital transformation cost is a big agenda. Still, I think the world needs cost take out deals…some of these digital journeys will ultimately help with cost programmes of businesses, and then they’ll use digital for growth programmes,” he said.

Tyagarajan added, “In an environment where growth is not easy, those two areas are sharply intersecting with each other. A lot of cost agendas are also driven by the need for digital, AI and generative AI.”

Meanwhile, top IT services firms including TCS, Infosys and HCLTech have started seeing recovery in large deal wins and renewals after a period of delayed deal closure timelines. A lot of these deals are AI and cloud themed, or were digital transformation deals with an element of AI.

TCS, among its several mega deals in 2023, last won a $1 billion deal from Jaguar Land Rover earlier this month.

Infosys bagged a $454-million from Danske Bank and a $1.6-billion deal from Liberty Global recently, which had AI components as a part of the contract. HCLTech won a $2.1-billion Verizon deal.

Internally too, IT services companies are seeing nearly 30 percent increase in productivity using generative AI for code generation and developers.