Tamil Nadu Chief Minister Edappadi K. Palaniswami on September 19 announced an ambitious Safe & Ethical Artificial Intelligence Policy. Like the Blockchain Policy announced on the same day, the Tamil Nadu e-Governance Agency (TNeGA) will play a key role in the policy’s implementations. The policy puts in place recommendations on how the ethicality of AI programs can be evaluated by government bodies. It also advocates for training government officials — including the judiciary — on AI, and calls for an AI sandbox for testing, much like the blockchain policy.

Introduction: Tamil Nadu in January 2019 launched the Center of Excellence in Emerging Technologies under the TNeGA to work with the government on emerging technologies like AI, IoT, VR, etc. CEET also works with startups.

AI as an engine for growth: AI is estimated by McKinsey to contribute US$13 trillion to the global economy by 2030. The police noted that while AI has become a “buzzword”, the term has been evolving continuously, with applications in self-driving cars, drones, voice assistants, and medical image diagnostics. In general, AI refers to computer-made decisions that mimic that of intelligent life. AI generally has these characteristics: i) learning from experience; ii) knowledge- and rule-based reasoning; iii) image recognition; iv) complex problem solving skills, v) natural language understanding, and vi) problem solving with alternate prospective.

Machine learning is a key subset of AI. Artificial Intelligence-based systems have led to miniaturisation of computer power, networking of sensors, and affordable internet access. AI is now advancing rapidly thanks to cheap storage and computing power.

Goals and objectives

Goals: This policy has the following goals: a) make AI fair, ethical, transparent, inclusive and free of bias; b) encourage public service delivery with AI; c) augment existing governance workflows with AI; d) build awareness on AI in all three branches of government; e) encourage research in AI by startups, public and private companies, and academia; and f) nurture an ecosystem of open data access for AI innovation.

Objectives: The policy’s objectives are: a) to provide a framework for AI that is fair, inclusive, and reliable; b) provide guidelines for evaluating AI systems before public rollout; c) build a mature, self-sustaining AI community and train and skill people in the state on AI; d) provide access to open data, data models, and computing resources; e) build a regulatory sandbox for testing AI applications; and f) promote AI R&D investments in Tamil Nadu.

Addressing concerns around AI: Why is this policy needed 

“AI Systems have often been criticized for having gender and racial bias problems, primarily when used in Law Enforcement,” the policy stated.

The policy notes that from education and governance to healthcare, AI touches many aspects of life. AI has a few distinct features that raise concerns that need to be addressed, like a) discreetness, as all AI systems are created and consumed collaboratively by many entities with attribution being almost impossible; b) diffusiveness, as each system or algorithm can be diffused across several geographies; c) discreteness, as each component of AI systems can be designed at different times for different purposes; and d) opacity, as the way an AI system works may be a black box by design or kept secret for proprietary reasons.

Since the AI market is not well organised, managing risks is challenging, especially for policymakers. In areas like banking and law enforcement, poor AI design may interfere with fundamental rights. For example, a bad assessment of creditworthiness may unfairly disadvantage a loan seeker. AI has also been criticised for gender and racial bias.

What is more challenging, the policy notes, is that the government has to be both a user of AI to deliver better services, as well as a regulator of AI to keep the technology harmless. As such, the government needs policies and a framework to make sure that AI is compatible with human values and is inclusive in nature.

Who does this policy apply to? The policy is applicable to all state-controlled or state-owned authorities, as well as organisations whose composition and administration is controlled by the state government. It also applies to joint ventures with the state government.

Framework for pre-rollout AI evaluation

The Tamil Nadu government will use the TAM-DEF framework and DEEP-MAX Scorecard to evaluate AI systems before they are rolled out for public use. TAM-DEF and DEEP-MAX were proposed on August 3 by an international, multidisciplinary team of 35 researchers. TAM-DEF consists of the following concerns.

  • Transparency & audit, to make sure that AI systems interacting with humans in fields like finance, education, healthcare, law enforcement, and elderly care are accountable to their decisions, and are not a black box, by having high standards for explainability and auditing;
  • Accountability and legal issues, as accountability of a system that takes decisions autonomously is challenging;
  • Misuse protection, to balance innovation without excessive regulation while reducing potential for misuse;
  • Digital divide & data deficit, where people and communities with access to data and better information benefit disproportionately more than others who lack access, and even harm the unconnected;
  • Fairness and equity, to make sure that the social fabric is not damaged and human labour is not commoditised.
  • Ethics, as explained below

AI Ethics: AI ethics is split into Privacy and Data Protection, and Human and Environmental Values. AI systems must make sure that highly granular information on humans that they are trusted with remains private, and that social values like respect, dignity, fairness, kindness and compassion are upheld; it should also be clear if the system has a “preferential duty” towards vulnerable groups like children and the elderly.

DEEP-MAX scorecard: The DEEP-MAX scorecard will evaluate an AI system on the following principles: a) diversity, for how well the system is trained for diversity in race, religion, accent, etc; b) equity & fairness, i.e., whether the system treats everyone fairly, c) ethics, i.e., how the system preserves values like dignity, respect, compassion, etc; d) privacy and data protection; e) misuse protection, f) audit & transparency, to evaluate whether decisions can be explained; and g) cross-geography and society, i.e., how well the AI system works across societies and geographies. The DEEP-MAX score has to be updated frequently as AI evolves upon use.

Using blockchain for safe and ethical AI: An AI Certification, Transparency & Scorecard Blockchain mechanism would ensure the safety and social desirability of AI solutions, for example by making sure the training material for an AI solution is diverse and equitable. ACTS-Blockchain would help in rating and understanding AI solutions before they are put to use. They would do this by i) providing a trusted mechanism to certify training data quality; b) assign standardised and tamper-proof DEEP-MAX scores; and c) building in misuse prevention, with a log of all changes.

Guidelines to procure an AI solution: Any government or government-aided body that procures AI systems will have to follow these guidelines before procuring an AI system: a) consult the Center of Excellence in Emerging Technologies (CEET) under the TN e-Governance Agency before making any procurement; b) evaluate the AI system on the DEEP-MAX scorecard as detailed above; c) assign an ethics score with the DEEP-MAX scorecard after weeding out any irrelevant parameters if needed; d) update the DEEP-MAX score periodically, perhaps once in six months; e) use the ACTS-Blockchain, to be developed by TNeGA, for safe and ethical use of AI; and f) reject AI systems that don’t meet the criteria.

Oversight: Implementation of these guidelines will be overseen by the Ethical AI Monitoring Committee, to be headed by the Chief Secretary of Tamil Nadu and composed of secretaries and senior officers of some departments, including law enforcement. AI/policy experts representing academia and research institutions will also participate.

AI training for all branches of government

Building capacity in all three branches of government — legislature, the judiciary and the executive — is essential as AI is likely to be all-encompassing in its reach. The police notes that this must also be done for citizens, as they need to know how AI will impact their lives; skill building for youth is also necessary, on top of updating school/college curriculums.

In addition to training sessions for legislators and senior officers of the government, the judiciary should also be a part of this process — in consultation with the Madras High Court, a program for educating judicial officers in AI and its implications would be drawn up, along with short training courses for judicial officers in training.

Promoting AI in Tamil Nadu

AI would be promoted among all the applicable bodies discussed above. To do this, the following steps will be undertaken:

  • Expert groups will be formed to identify application opportunities, with workshops, conclaves and networking events. These groups will facilitate knowledge sharing, policy development, and dissemination of best practices, in addition to recognising significant contributors to the field.
  • Open AI challenges will be held for enthusiasts, students, teachers and startups.
  • An AI Research Centre would be constituted.

To implement this policy, the following are needed: a) open data that is high quality and reliable, with shareability in the form of APIs; b) a data platform that the government would create to host information and encourage interoperability and secure data management practices; and c) an AI sandbox for experimenting with new ideas without a risk of failure.

Implementation: The policy’s implementation will be overseen by the Tamil Nadu Department of Information Technology, which will issue guidelines and revisions as needed from time to time.

*This is a summary and does not necessarily reflect MediaNama’s views on the subjects discussed above.