E-Paper
Create AI governance group led by PSA, build AI incident database: Panel
The panel was led by Balaraman Ravindran, who heads the department of data science and AI, and the Centre for Responsible AI, at IIT Madras
NEW DELHI: A panel set up by the government last year is likely to recommend that India’s information technology ministry and the office of the principal scientific advisor (PSA) should create an inter-ministerial artificial intelligence (AI) coordination committee or governance group to take a whole-of-government approach to AI governance.
According to a draft report of the panel shared with stakeholders on October 7, this governance group should bring together all authorities and institutions that deal with AI governance at the national level so that the government and regulators have a better understanding of the AI ecosystem in the country so that governance is “rooted to the realities of existing and likely risks”.
HT has reviewed the sub-committee’s draft report.
The ministry of electronics and information technology (MeitY) formed the sub-committee on November 9, 2023, to “analyse gaps and offer recommendations for developing a comprehensive framework for governance of Artificial Intelligence”. The panel is headed by Dr Balaraman Ravindran, head of the department of data science and AI and the Centre for Responsible AI, at IIT Madras.
The draft report also said that MeitY should house a technical advisory body and coordination focal point for this governance group, which should build an AI incident database to understand the “actual incidence of AI-related risks in India”, echoing a similar recommendation by NITI Aayog in 2021.
These AI incidents could include “cyber incidents” and “cyber security incidents” and extend to “adverse or dangerous” outcomes from the use of AI that can disadvantage or harm individuals, businesses, and societies”. They could include “malfunctions, unauthorised outcomes, discriminatory outcomes, unforeseeable outcomes and unexpected emergent behaviour, system failures, privacy violations, physical safety problems, etc.”
The draft report also recommended that the technical secretariat try to get “voluntary commitments on transparency across the overall AI ecosystem and on baseline commitments for high capability/widely deployed systems” from the industry. These could include regular transparency reports by AI developers and deployers, internal and external red-teaming of models or systems, peer review by third-party qualified experts, and others. The commitments could vary across sectors.
It said that initially, only public sector organisations using AI systems should mandatorily report to this database while private entities were “encouraged to voluntarily report AI incidents”. “The focus should be on defining reporting protocols to ensure confidentiality and to focus on harm mitigation, not fault finding,” the draft report said.
It added that the technical secretariat should try to get the industry to get “voluntary commitments on transparency across the overall AI ecosystem and on baseline commitments for high capability/widely deployed systems”. These could include regular transparency reports by AI developers and deployers, internal and external red-teaming of models or systems, peer review by third-party qualified experts, and others. It added that transparency and governance measures “may also be adopted by the government and their technology providers” since the government is using AI for law enforcement and citizen welfare.
The draft report also recommended that the technical secretariat could examine the suitability of technology measures to address AI-related risks, such as use of watermarking, labelling and fact-checking to deal with deepfakes.
The draft report recommended that the governance group should meet regularly to suggest measures that strengthen existing laws to minimise “risk of harm due to use of AI”, and issue joint guidance to provide legal clarity and certainty around the development and use of AI, amongst other efforts. This governance group could also “encourage” the creation of “sector-specific datasets” relevant to the Indian context.
“Currently, regulators and government departments may have some visibility on the AI systems developed or deployed by entities who are under sectoral regulation (e.g., finance or health) or where the market is concentrated (e.g., e-commerce, social media, aggregators). However, the level of visibility would need to be adequate to assess potential risks associated with such entities in the context of AI. Further, there are likely to be AI systems developed or being developed and/or deployed by entities who may not have an interface with the government/regulators from a perspective of affording suitable visibility to enable a risk assessment in relation to AI,” the draft report said.
It said that the governance group should have a mix of government and non-government members to bring in “external expertise from industry and academia” and could be headed by the PSA. On the other hand, the technical secretariat located in MeitY could be staffed by MeitY officials and “lateral hires, young professionals, and consultants.
“MeitY may form an AI Sub-Group to suggest the form and structure of the proposed secretariat along with a detailed term of reference,” the draft report said. This AI sub-group could also make detailed recommendations about effective grievance redressal and ease of doing business under a new law such as the Digital India Act.
This sub-committee also noted that providers and deployers of AI systems cannot claim “safe harbour” by default as in many scenarios, they “select or modify the content”, something that intermediaries such as Google, Facebook, Cloudflare and Airtel cannot do to claim protection from liability for third party content.
