1 of 10

Authentic AI Solution Proposal

Wenjing Chu @ May 9, 2024

2 of 10

Authentic AI: use cases

  • Securing elections
  • Authentic GAI in social media systems
  • Healthcare
  • Data collection/contribution
  • Creator's assertion of authorship and distribution rights
  • Customer services
  • Travel planning and execution
  • AI based fraud and fraud prevention
  • Agentic AI
  • Verifiable Model Cards for AI Models and Apps
  • Wearable AI Pins

3 of 10

A Social-Technical Model for AI Systems

4 of 10

A Social-Technical Model for AI Systems

Model developer

System deployer

User

High level roles as defined by the OpenAI paper

5 of 10

An Expanded Socio-technical Model for AI Systems

Learned Model

Service Software

Aug.Data

Training System

Compute

Datasets

Code

Text, audio, image, video…

User(s)

Model developer

Service deployer

Result

External Services

Deployed Services

Collected data

Log

Applied rules

6 of 10

Another Architectural Risk Analysis

7 of 10

NIST AI Risk Management Framework (RMF)

https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.100-1.pdf

8 of 10

OpenAI news

https://openai.com/index/understanding-the-source-of-what-we-see-and-hear-online/

  • Provenance meta-data (via C2PA)
  • Voice watermarking
  • AI classifier
  • CALLS: help in distribution

9 of 10

Authentic AI Content in New Media

TSP carries the authenticity info of the distributor/creator

Distributor

Creator

Consumers

New media

10 of 10

TSP + C2PA outline

  1. Specify how to use TSP to share/consume C2PA labeled content

TSP Envelope

TSP Payload

TSP Signature

C2PA Meta-data

Creator’s ID

Consumers

Media content

Creator

Distributor