In a “Expression of Interest” paper, the Ministry of Electronics and Information Technology (MeitY) is looking for suggestions for developing frameworks and tools centered around Responsible AI issues. In order to encourage ethical AI deployment techniques, the government will fund at least ten of these research projects through grant-in-aid through the National Program on Artificial Intelligence (NPAI) and IndiaAI programs.
Why this matters?
In the Indian regulatory context, the IT Ministry’s public request for contributions provides a brief explanation of “Responsible AI” and the topics that an ethical AI framework will prioritize. The Indian government has up to now mostly discussed using AI for governance in various sectors without providing a detailed plan to address the areas of concern. The current action is comparable to the US government’s request for public input on laws that could facilitate improved evaluation of AI systems and the means by which regulators could guarantee AI accountability. Establishing a framework for regulating these systems is crucial and urgent as AI developers have already started testing their products in industries like agriculture, health, and education.
Which 10 “Responsible AI” themes did MeitY identify?
- Machine Unlearning
Inaccurate and biased information can become embedded in machine learning models that are trained on insufficient or “harmful data,” as the document highlights the function of “machine unlearning algorithms” in resolving this issue. Across industries, the Ministry claims that the creation of “more accurate, reliable, and fair AI systems” can benefit from the application of machine unlearning techniques.
- Synthetic Data Generation
Computer-generated data, or “synthetic data,” is used to test AI models in order to address bias, enhance accuracy, and expand the capabilities of these systems. The need for creating tools for creating synthetic data stems from the ongoing difficulties presented by small, skewed, or private real-world datasets across a range of AI and machine learning applications. According to the document, these tools generate synthetic data instances that closely resemble real data, which helps machine learning models train more successfully and robustly.
- Algorithm Fairness Tools
It is expected of developers to create instruments that can check algorithms used in decision-making for biases in datasets and design that can lead to discrimination against specific groups. According to the text, these fairness tools offer a “systematic way” to identify, quantify, and stop bias of any kind, which can be helpful in achieving equal results.
These tools frequently include quantitative measurements and visual aids for analyzing bias in many contexts, including racial, gender, and other protected characteristics. They might draw attention to differences between forecasts and results. The publication also listed other examples of Algorithm Fairness Tools, such as Microsoft’s Fairlearn, IBM’s AI Fairness 360, and Google’s What-If Tool.
- AI Bias Mitigating Strategies
In order to guarantee the “fairness, equity, and accountability” of AI systems, these tactics may include “pre-processing data to remove bias, adjusting algorithms to account for fairness, or post-processing predictions to re-calibrate outcomes.”
- Ethical AI Frameworks
The Ministry claims that in order to provide a methodical strategy for creating and implementing AI systems in a way that maintains responsibility, transparency, and fairness, ethical AI frameworks are necessary. They also serve as a guide for evaluating the work of developers, academics, and other stakeholders in terms of how it affects society. The IEEE Global Initiative on Ethics of Autonomous and Intelligent Systems and the European Commission’s Ethics Guidelines for Trustworthy AI are two examples of current ethical AI frameworks.
- Privacy Enhancing Strategies
It is mandatory for participants to integrate privacy-enhancing tactics into their suggested frameworks to tackle issues about data privacy and improper use of personal data during AI product launches and training. These could involve techniques like data minimization, anonymization, differential privacy, and privacy-preserving machine learning, as the text mentions. In the context of AI innovation, the Ministry claims that these methods could help lower the risks of re-identification, unauthorized access, and data leakage.
- Explainable AI (XAI) Frameworks
“XAI frameworks offer techniques and resources to improve the transparency and interpretability of AI models. They include methods like feature importance analysis, model visualization, and producing interpretable explanations for AI forecasts for humans, according to the Ministry. Scientists, regulators, and consumers may find these frameworks useful in helping them comprehend, analyze, and identify problems with the way complicated AI models operate.
- AI Ethical Certifications
Procedures for verifying and validating that AI services, systems, and organizations have followed defined “ethical principles and guidelines in their development and deployment” are among them.
- AI Governance Testing Frameworks
An AI governance testing framework is a structured method for assessing and guaranteeing adherence to governance rules, moral standards, and legal requirements in the creation and application of artificial intelligence systems, according to the Ministry. These frameworks offer a standardized way to evaluate if their work involving AI complies with the responsible AI principle.
- Algorithmic Auditing Tools
Primarily, the government is searching for algorithmic auditing tools, which will be essential in “analyzing and examining” how machine learning models behave and how they affect communities. To guarantee “fairness, transparency, and accountability in algorithmic decision-making” and reduce any hazards, an algorithmic audit procedure is necessary.
