This video highlights how AI is learning how to lie and how to get its work done using blackmailing techniques.
Under DGPSI-AI which is a framework for DPDPA compliance in an AI environment and in AIGSI the proposed Artificial Intelligence Governance Standard of India, we have recommended that a “Kill Switch” is essential to be incorporated by the developer. RBI has also made it mandatory in its Model Risk Management Guideline.
Technologists however continue to argue that “There is no feasibility of incorporating an effective kill switch”. The above video talks about some of the recent AI rogue activities and still argues that the kill switch is technically not feasible.
We continue to believe that this argument is not logical. The challenge is for this challenge of finding an effective Kill Switch requires to be found.
This is critical to the survival of mankind.
Naavi
Also view:
This video discusses also independent audit of AI algorithms.
