AI regulator with broad mandate mooted
Sanjeev Sanyal, a member of the Prime Minister’s economic advisory council (PMEAC), says India should have a specialist AI regulator with a broad mandate, along with a national registry of algorithms and a “repository of national algorithms for innovation of AI”.
There was a need for such a regulatory framework amid “extreme approaches” being taken by global economies, he said in a research paper published by PMEAC that suggests ways to regulate AI.
Sanyal said traditional methods fall short due to the nonlinear and unpredictable nature of AI. Current regulatory approaches typically rely on ex-ante impact analysis and risk assessment and therefore face challenges in effectively governing AI.
The paper, titled ‘A Complex Adaptive System Framework to Regulate Artificial Intelligence’ and written by Sanyal, Pranav Sharma and Chirag Dudani, proposes a framework based on CAS (Complex Adaptive System) thinking, consisting of five key principles.
These include establishing guardrails and partitions to limit undesirable AI behaviour, mandating manual overrides and authorization chokepoints where critical infrastructure will remain in human controls at key stakes for active intervention.
The principles also include open licencing of core algorithms and continuous monitoring of AI systems for ensuring transparency, accountability and explainability, while mandating incident reporting protocols to document system aberrations or failures, that will define clear lines of AI accountability and ensure ‘skin in the game’ by holding individuals or developers responsible.
The key pillars have been suggested after considering approaches taken by other countries. The US and UK, for instance, have taken a hands-off or selfregulatory approach, the paper notes, as opposed to the heavily state-regulated approach adopted by China.
India has offered to lead the development of a draft global artificial intelligence (AI) regulatory framework, which will be discussed and debated at the GPAI (Global Partnership on Artificial Intelligence) Summit, sometime in June or July.
The GPAI is a grouping of 29 nations including the European Union that in December last year adopted the New Delhi Declaration where countries agreed to use the GPAI platform to create a global framework on AI trust and safety, within six months.
Against that backdrop, the research paper by the PM-EAC member said open licencing of core algorithms for external audits, AI factsheets, and continuous monitoring of AI systems, are crucial for accountability, apart from periodic mandatory audits for transparency and explainability.
Kazim Rizvi, founder of one of India’s leading tech policy think tanks, The Dialogue, said the formulation of AI regulation in India will be a complex endeavour which will require careful consideration to ensure responsible and ethical development and deployment of AI technologies.
A spokesperson of the electronics and IT ministry didn’t respond to emailed queries.
Current regulatory approaches typically rely on ex-ante impact analysis and risk assessment