New Delhi: Government institutions should exercise caution before adopting AI-based solutions and ensure they are thoroughly tested and reliable, a senior official said on Friday.

Rohit Bhardwaj, Deputy Director General of the Data Informatics & Innovation Division at the Ministry of Statistics and Programme Implementation (MoSPI), also urged government departments to prepare their data for AI use by making it machine-readable.

"There should be context files, there should be semantics, and there should be metadata," he said, emphasising the storage of information in a structured format so that they become AI-ready.

Also read | India joins US-led ‘Pax Silica’ alliance to boost AI and semiconductor supply chains

Flagging the importance of testing an AI solution before putting it into use, Bhardwaj cited a report by a group of researchers from a Canadian university, which shows that AI can analyse a particular data set in multiple ways even if given similar prompts.

"I just want to flag that we should not be just gung-ho about things which is still untested," he said at the AI Impact Summit session during a session on 'AI-Ready Data: Shared Infrastructure for Innovation'.

"I would be the first person to adopt AI for my work, but it needs to be trustworthy, he said, adding, "People are not aware what it takes to make data AI-ready, and it is the responsibility of institutions like MOSPI to make people aware about what AI readiness is all about."

Also read | 19-year-old brutally beaten by live-in partner in Gurugram; rushed to AIIMS

He suggested that ministries or government departments should have a catalogue and that all files should not be in PDF form, as it should be machine-readable.

"I plan to create a slide deck, what AI can see and what AI can't see. So my folder, if it has 10 versions, some answers will come from Version 1 and some from Version 2. Unlike humans, AI is designed to scan the entire thing (data) available," he said.

Prem Ramaswami from Google said that his company tries to bring multiple data sets globally together in a common knowledge graph and put a date searching engine on top of that. "So that you can quickly access that data. We open-source that entire stack."

He pointed out that the idea of data being centralised at one source is dangerous.

Instead, he suggested, the data should be located at every organisation and governed locally by the organisations, making them affordable and accessible to business enterprises.

"If you have 74 million MSMEs in India, you cannot afford to have data scientists or computer scientists that you can hire," he said.

He further explained, "If you are a policymaker, thinking about poverty, climate change, education and health, these are holistic problems. It is no longer ...I can go to one ministry, pull one spreadsheet and solve the poverty. We should approach AI as a tool we can use to derive the answer," he said. PTI