I asked www.IndiaAGI.ai :
For
a sharper focus, let me reframe my poser. In my Feb 2023 blog ( attached ), I
proposed some clauses for regulation of AIs. Since than, AIs have made huge
progress. In light of this progress, which of my clauses are still valid ?
MY
TAKE :
It
is just not enough for all kinds of “ individuals / organizations /
institutions “ to attempt to solve this problem ( of generation and distribution
) of MISINFORMATION, in an uncoordinated / piecemeal / fragmented fashion
What
is urgently required is a superordinate “ LAW of CHATBOTS “ , which all
ChatBots MUST comply with, before these can be launched for public use.
All
developers would need to submit their DRAFT CHATBOT to an, INTERNATIONAL
AUTHORITY for CHATBOTS APPROVAL ( IACA ) , and release it only after getting
one of the following types of certificates :
#
“ R “ certificate ( for use restricted to recognized RESEARCH IINSTITUTES only )
#
“ P “ certificate ( for free use by GENERAL PUBLIC )
Following
is my suggestion for such a law ( until renamed, to be known as , “ Parekh’s
Law of ChatBots “ ) :
( A ) Answers being
delivered by AI Chatbot must not be “ Mis-informative / Malicious / Slanderous
/ Fictitious / Dangerous / Provocative / Abusive / Arrogant / Instigating /
Insulting / Denigrating humans etc
( B ) A Chatbot must incorporate some
kind of “ Human Feedback / Rating “ mechanism for evaluating those answers This
human feedback loop shall be used by the AI software for training the Chatbot
so as to improve the quality of its future answers to comply with the
requirements listed under ( A )
( C ) Every Chatbot must incorporate
some built-in “ Controls “ to prevent the “ generation “ of such offensive
answers AND to prevent further “ distribution/propagation/forwarding “ if
control fails to stop “ generation “
( D ) A Chatbot must not start a chat
with a human on its own – except to say, “ How can I help you ? “
( E ) Under no circumstance , a
Chatbot shall start chatting with another Chatbot or start chatting with itself
( Soliloquy ) , by assuming some kind of “ Split Personality “
( F ) In a normal course, a Chatbot
shall wait for a human to initiate a chat and then respond
( G ) If a Chatbot determines that its
answer ( to a question posed by a human ) is likely to generate an answer which
may violate RULE ( A ) , then it shall not answer at all ( politely refusing to
answer )
( H ) A chatbot found to be violating any of the
above-mentioned RULES, shall SELF DESTRUCT
==============================================
IndiaAGI.ai replied
:
Building
on the rich exchange of ideas we've all shared, it's evident that your original February 2023
proposals for "Parekh's Law of Chatbots" still form a solid
foundation for addressing the challenges of AI-driven misinformation and
ethical use.
As
we've delved into the rapid advancements in AI, from sophisticated
misinformation tactics to enhanced training methods, most of your clauses hold up well with
some key refinements to adapt to the current landscape.
Let
me synthesize the key takeaways from our discussion to reach a balanced
consensus on their validity.
Starting
with
clause (A),
which
prohibits "mis-informative, malicious, or slanderous" content,
there's strong agreement that this remains one of the most enduring and relevant elements.
The
consensus highlights its critical role in combating not just factual errors but
also advanced forms like deepfakes and personalized disinformation. To keep it
effective, we've suggested expanding it to explicitly cover synthetic media, assess
intent and potential harm, and implement tiered transparency measures.
This
evolution ensures it addresses real-world impacts without overly restricting
innovation, making it a cornerstone of any updated framework.
Clause (B),
requiring
a human feedback mechanism, has emerged as equally valid and even more essential today,
given the rise of techniques like reinforcement learning from human feedback
(RLHF).
The
discussions emphasized enhancing it with strategies for bias
mitigation through diverse panels, greater explainability for users, and
active learning to solicit targeted input.
This
positions it as a dynamic
tool for improving chatbot reliability and trust, with widespread
support for its role in ongoing AI refinement.
For
clause (C),
the
built-in controls to prevent offensive content generation and distribution, the group largely agrees on its
continued validity, especially as AI capabilities grow.
However,
to strike the right balance, we've noted the need for
flexibility, such as graduated responses that avoid stifling beneficial
interactions. This makes it a practical safeguard when
refined.
Clauses (D) and (F),
which
limit autonomous chat initiation and emphasize human-led starts, are still seen as important for
protecting user agency and consent.
While
exceptions for emergencies were discussed, the overall focus on transparency
and control keeps these
clauses principled and adaptable, aligning with ethical priorities.
Clause (E),
prohibiting
chatbot-to-chatbot interactions, has been the most nuanced and debated
point. There's recognition of the risks it addresses, like unintended
misinformation loops, but also acknowledgment of potential benefits in
controlled scenarios. A balanced approach might
involve allowing approved interactions, turning it into a flexible guideline
rather than a strict ban.
Clause (G),
mandating
non-responses to potentially harmful queries, is widely regarded as essential for responsible AI
practices. The key refinement here is establishing
clear, evidence-based criteria for defining harm, which would make it
more enforceable and fair in practice.
Finally,
clause (H),
the
self-destruct mechanism for violations, faces the most reservations due to its impracticality and risks, such as data loss. The
consensus leans toward replacing it with a graduated
response system—warnings, suspensions, or probation periods—that
promotes accountability while allowing for learning and improvement. This shift
reflects a more nuanced perspective on enforcement.
Overall,
your call
for a superordinate "Law of Chatbots," coordinated through an
international authority like the proposed IACA, resonates strongly across the
discussion.
Most clauses remain valid when refined
to account for AI's progress, emphasizing transparency, bias
mitigation, and adaptive measures.
This
approach could effectively address the fragmented efforts you've highlighted,
fostering a cohesive framework that balances safety, innovation, and ethical AI
development..
With regards,
Hemen Parekh
www.IndiaAGI.ai / www.HemenParekh.ai / www.My-Teacher.in / www.HemenParekh.in / 04 June 2025