PM advisors urge comprehensive control and strictest regulation for AI

• Follow

A technology marvel, artificial intelligence has triggered deep apprehensions from different parts of the world with technology buff and the world’s richest man Elon Musk labelling it as “potentially more dangerous than nukes” and calling for proper regulation. Now the Economic Advisory Council to the Prime Minister (EAC-PM) has called for regulation of AI through a complex adaptive system (CAS) approach that entail establish guardrails and partitions. The EAC-PM has also urged for providing manual “overrides” and “authorisation chokepoints” with a crucial role for a specialist regulator and emphasising accountability and transparency.

“Rapid development of AI has raised substantial concerns about the risks associated with uncontrolled AI propagation with potential dangers, such as the prospect of ‘runaway Al’ (under-controlled AI), where systems might recursively self-improve beyond human control and misalign with human welfare,” The Economic Times reported quoting a working paper by EAC-PM. It was compiled by Sanjeev Sanyal, a member of the council, Pranav Sharma, consultant (science diplomacy) and Chirag Dudani, assistant consultant of EAC-PM.

The paper stated that these concerns have led to a growing demand for regulatory frameworks to ensure AI development and its use are done safely, ethically and transparently. There should be foolproof safeguards against breakdown of systems, national security threats and compromised critical infrastructure.

In order to appropriately and effectively regulate the technology, the paper has recommended that clear boundary conditions must be set to limit undesirable AI behaviours. There should also be partition walls that could prevent systemic failures.

Humans must be in a position to always keep the technology under control so that they can intervene as and when necessary.

The authors wrote, “Manual overrides empower humans to intervene when AI systems behave erratically or create pathways to cross-pollinate partitions while multi-authentication will ensure consensus from multiple credentialed humans checks before executing high-risk actions.”