Artificial Intelligence, Circuit Breakers, Partitions and Knowledge Management

From Abhivardhan, our Chairperson

This is a post authored by Abhivardhan, our Founder and Chairperson.

I recently came across a fantastic paper co-authored by Sanjeev Sanyal from the Economic Advisory Council to the Prime Minister (EAC-PM) on artificial intelligence and complex adaptive systems (read the complete paper at https://eacpm.gov.in/wp-content/uploads/2024/01/EACPM_AI_WP-1.pdf).

This excerpt from the paper kind of proposes some excellent ideas, such as:

"Governing Principles for AI:
1. Guardrails and Partitions/ Fire Breaks & Choke Points
2. Circuit Breakers/ Manual Overrides
3. Transparency & Explainability
4. Distinct Accountability
5. Specialized, Agile Regulatory Body"

National Registry for AI:
"A separate national registry of unforeseen or dynamic non-standard events during the functioning of AI models can be made to create necessary feedback loops for the regulator."

"Equip the regulator with tools and methodologies to continuously scan the AI horizon, identifying potential pitfalls or areas requiring regulatory attention."

"Encourage a feedback-driven approach, where the regulator engages with industry and academia to refine its directives continually."

Liability Protocols:
"Clearly delineate responsibilities among developers, operators, and end-users of AI systems."

"Embedded traceability mechanisms. "

While these proposals are noteworthy, I would like to highlight that knowledge management is a core aspect of ensuring better AI regulation, especially in India.

This is why in the AIACT.IN draft proposal I had authored for Indic Pacific Legal Research LLP - we had recommended Section 8 to enable Knowledge Management Practices for AI entities. Here is an excerpt:

"(4) Entities already operating AI systems within the jurisdiction of India shall comply with the prescribed model standards within a reasonable timeframe, as determined by the IRDC. The compliance timeline may vary based on the complexity and risk levels associated with AI systems.
(5) The Central Government shall empower the IRDC or agencies to establish a knowledge management certification process that verifies the compliance of AI entities with the model standards outlined in this section.
(6) AI entities, upon fulfilling the compliance requirements, may be awarded a knowledge management certification, signifying their adherence to industry best practices in data handling, model governance, and ethical decision-making in AI technology.
(7) AI entities shall be required to submit regular reports to the designated authorities, outlining their adherence to the model standards for knowledge management and decision-making."

Still, an overarching regulation on every aspect of artificial intelligence may not be a great start, because a lot of what Sanjeev Sanyal's paper suggested is covered in cybersecurity principles.

However, I wholeheartedly support AI standardisation not for stringent compliances, but for effective documentation purposes, which can inform regulators to create better regulatory standards, across industry sectors.

Read the complete LinkedIn Post here.