“In-Flight Transponder” Code

Per recent EU AI Regs, transparency is a recognized requirement for high risk AI applications. Methods for model self identification and broadcasting will ensure centralized data disclosure and regulator transparency.

By Pamela Jasper

What is your proposal? AI models are built by algorithms to replicate human intelligence. But unlike humans AI models are not self referencing and generally do not have methods of self identification. Smart Models is a method of centralized management and disclosure of AI models wherein each AI model will contain a unique model id within its API code. Upon live use the model will broadcast to a central server live intra day messages of the models usage , status and other vital information, including explainable decisions. The models will act like an aircraft transponder during flight. The model broadcast receiver will exist either inside or external to the developing firm. External receivers at regulatory bodies can monitor real time model usage, bypassing faulty self assessments or audits. Based on the work of NYU Professor Jon Hill.

What problem does your proposal address? A major problem for high risk AI (those used in health care, criminal justice, lending, housing, education, etc.) is the need for reg oversight of model activity and end consumer user knowledge of when AI is being used. This proposal would create a required central repository of model usage for live models, thus serving as a means of informing and verifying end user exposure to the AI model, and facilitating reg oversight and insight into its usage. This is a data intensive proposal, similar to FINRA oversight of all stock exchange trades.

How does this policy proposal relate to artificial intelligence? Smart models can be used for all AI model types AI, ml, dl. Facilitates transparency, governance and provides industry data to compare usage stats. Right now the regulators rely upon self reported data from AI tech firms if they provide it. It would transform regulatory technology and foster USA consumer confidence in AI.