Grounding RAI principles: Case Studies

Gokul S Krishnan , Sivaramakrishnan Guruvayur , Srinath K R

Collaborators:

Description:

Partner:

Open Loop is a global program supported by Meta. They partner with governments, tech companies, academia and civil society to co-create and test new governance frameworks through policy prototyping programs, and to support the evaluation of existing legal frameworks through regulatory sandbox exercises.

Aim: Mapping of generic RAI principles into grounded policies, governance models and regulations for sector-specific deployment of AI.

When AI models are deployed in various domains, the policies and guidelines recommended for each domain will be different. They will also be multi-dimensional in accordance to the level and means of governance(top-down regulation, self-regulation, and co-regulation). Thus there is the requirement to learn, understand and accordingly map the use cases, performances and deployment of AI models/systems to these policies tailored to the sector. This is further illustrated in the following examples.

  1. Finance

    a. AI-based credit assessment and Lending

    b. Determining Fraudulent practices in financial markets

    c. Banking customer services through chat-bots

  2. Healthcare

    a. Disease diagnosis and decision making

    b. Medical chatbots and virtual nursing assistants

  3. Education: AI-based Learning Management systems

    a. Equality of Assessment

    b. Educational material allocation

  4. Transportation

    a. Autonomous cars and accidents

    b. Emergency transport allocation

    c. Equity in the transport sector

  5. Law enforcement

    a. Criminal behavior and predictive policing

  6. HR

    a. CV selection and Job advertisements

  7. Manufacturing

    a. Defect detection and quality control

Output - Report 1 Chapters:

  1. Survey of Open Source Explainability Toolkits for Fraud Detection in the Finance Sector (1.a)

  2. Ensuring Ethical Implementation of Large Language Models in EdTech: Mitigating Cheating and Plagiarism Risks through Grounding and Responsible AI Practices (3.a)

  3. Responsible AI + Human Collaboration in Social Media Moderation

  4. Evaluating Deployability of LLMs: Responsible AI of LLMs in Healthcare & Biomedical sector (2.a and 2.b)

  5. Responsible AI in Applications of Recommender Systems (adapt for 6.a and 7.a)

  6. Grounding Explainable AI Principles in Medical Imaging (2.a)

For each case/chapter, we elucidate the following dimensions:

# Section
1 Application domain description
2 AI tech description review
3 Motivation of RAI in this case through survey of problems, controversies, media coverage, court cases, etc
4 Survey of state-of-the-art in RAI in this case in tech and governance
5 Recommendations in tech and governance

Links:

  1. Open Loop