BackCompanies' AI adoption will hinge on safety standards: Deepak Sharma of Suryoday Small Finance Bank
He added that conducting audits will be key to promoting transparency in AI in financial services, which in turn will be key to building consumer trust.
PremiumMumbai: The adoption of artificial intelligence (AI) by enterprises will hinge on safety standards, Deepak Sharma, CXO advisor and independent director, Suryoday Small Finance Bank, said at Mint Digital Innovation Summit on Friday.
Speaking about the implementation of AI and generative AI services responsibly in the all-important banking and financial services industry, Sharma said the adoption of standards for AI will define benchmarks for the quality of datasets, transparent audits of these datasets, referencing of data and, most importantly, building an understanding of AI models for the right implementation.
“There are key themes when one is looking at mass adoption of AI, including whether the data is voluminous enough. If the answers to such questions is yes, you’ll have a checklist to move ahead in AI adoption. Be it AI or generative AI, they can only do as much as is provided. If the quality of the data is not pristine, don’t blame the LLMs – they won’t be able to make a difference. Auditability and traceability are crucial in AI, too," Sharma said.
He added that conducting audits will be key to promoting transparency in AI in financial services, which in turn will be key to building consumer trust. “During audits, one should be able to demonstrate open end-to-end auditability. Enterprises also need to gain an understanding of content and context as AI models are deployed," he said.
Sharma underlined that consumer-facing AI will be depolyed only after AI ‘hallucinations’ are dealt with satisfactorily. “A second pillar is about AI control, which involves hallucination protection as we look to build enterprise scale. One way is to see how a model’s output can be reckoned with the raw data, which is crucial for businesses. Referencing of data is key to handle hallucinations."
He said that there were no substitutes for specific safety-first measures. “There is also the need for confidence-scoring to affirm datasets. There are no substitutes for human review. Partners working on AI also need to ensure there’s no data retention," Sharma added.
3.6 Crore Indians visited in a single day choosing us as India's undisputed platform for General Election Results. Explore the latest updates here!
