The Aakhya Weekly #89 | Machine Learnings: AI Regulation in the EU and India
In Focus: AI Governance: EU’s AI Act & India’s Regulatory Prospects
By Kushal Mital
Last week was a tempest for the world of technology and policy. The world saw its very first piece of legislation on AI regulation enacted, as the European Union voted overwhelmingly in favour of the proposed AI Regulation Act, 2023. It is said to be the world’s first extensive regulatory framework on artificial intelligence development. It establishes a risk-based approach to governing AI systems and is a point of reference for global actors to take action.
In a similar light, India’s own long-brewing concerns over misinformation, the misuse of generative AI and the exploitative nature of an unregulated tech sphere, motivated the Ministry of Electronics and Information Technology to issue an AI advisory last week. The advisory outlined guidelines for AI developers such as large tech firms to seek government approval before initiating development of AI programs and software. Backlash and a frenzy of responses from the tech industry prompted the release of a subsequent clarification, that only “significant” companies and AI industry stakeholders would be subject to the advisory. This is our first insight into what Indian AI regulations could look like and can be considered a precursor to a comprehensive artificial intelligence governance policy, which can be expected after the general elections.
Evidently, governments and civil societies across the globe are clobbering to frame and implement AI regulation with a growing sense of urgency as these systems become increasingly sophisticated and possibly, more dangerous. Given this, it’s worth exploring the origins and motives of the EU regulators as well as how India can gain key insights on shaping policy over shared interests with the EU.
Insights from the EU
The EU has long been in process to enact a legal framework for the use, development and dissemination of AI, as it introduced the very first draft of the Act in 2021. Their motives have remained clear and consistent; to harness AI in a manner that provides for innovation, public good and private industry, without compromising on fundamental rights of users. The EU’s Vice President, Vera Joruva reiterated this sentiment by praising the vast potential of AI across industries but cautioned against prioritising innovation before preservation of fundamental rights and freedom. This translated into the risk-based framework which aims to differentiate the need for compliance upon AI stakeholders, to reduce oversight on AI tools designed to further research, innovation and public good while ensuring strict accountability for riskier softwares. The ‘regulatory sandbox’ is also an interesting and extremely useful example of a methodology that India too might pursue. Avenues such as these would help understand the plausible policy elements which could help shape Indian laws.
The four categories under the EU’s risk-based approach to AI regulation range from- “unacceptable risk” where AI which erroneously influences the user; to minimal risk criterion for AI software such as a spam filter. Such a model could succeed in an Indian context, where indigenous AI development would require minimal bureaucratic pressure to succeed as a technological hub. This is more suitable as compared to a sector specific Act, as it allows each industry to take steps to shape regulation and implement it without being bound by arbitrary criterion. For instance, sectors such as healthcare and medicine may have different concerns about AI as compared to the entertainment industry. Such a system could potentially address a wider range of AI-based disputes and concerns.
Furthering the notion of a “case-based” approach, the EU introduced the regulatory sandbox. The Act proposed creating a “sandbox”, i.e a framework for early development and monitoring of AI before it is deployed on the open market to ensure compliance, as it’s initially trained and tweaked. In an Indian context, with a little tweaking to focus more on incubating these AI startups and helping build an AI ecosystem, Indian developers could benefit from regulation rather than feel its burden. The recent advisory too, has hinted at such a system, since it also suggests consulting with the government prior to development of AI tools and software.
In essence, India and the EU have a shared interest in developing policies that can curb harmful aspects of AI, promote a competitive AI industry and provide mechanisms for widespread implementation. Beyond this, what are some unique Indian problem statements an AI act must address?
Some Questions for India’s AI Regulation
India’s social and demographic structure have led to certain unique risks and concerns regarding AI. The most notable being the challenge of deep fakes, the question of self regulation, and scale.
First, this is relevant for India due to a higher vulnerability to misinformation in the media and the difficulty in distinguishing between artificially altered media and original content. The EU’s stance on deepfakes cannot simply be mimicked by India. Deep fakes are only considered a tier-2, transparency risk AI, where the EU requires the creator to disclose the artificial origins of the media while sharing it. However, with the percolation of misinformation in India so significant, if creators find methods to dodge liability, this policy would be futile in India. Given how the government expressed AI-specific concerns, policymakers could outline limited uses of deepfakes to act as a deterrence against widespread misuse.
Secondly, how will India tackle self regulation?
India could employ a self regulatory framework for governance, where a risk-based categorisation has differentiated levels of liability. Possible tools involve labels on the outputs of AI programs, consistent auditing of training data to check for privacy and training biases, while limiting the use of private digital information. A self regulatory environment has even been highlighted in India’s 2018 National Strategy for AI by the NITI Aayog, where it was recognised that given the nature of AI, any initial policy will become outdated if it is not implemented in an adaptive manner.
Finally, how will the scale and size of an AI program and developer be defined?
The EU defined all GPAIs (General Purpose Artificial Intelligence) on the basis of a set level of computing power. This has witnessed backlash from stakeholders, with the industry claiming that arbitrary limits would derail progress for startups and punish companies for growing. Instead, India must definitively clarify what distinguishes an established, high impact AI-based company or tech player as compared to a startup. This segregation is essential if we are to ensure a balance of regulation and digital safety while supporting homegrown tech to compete on a global frontier.
Policy blindspots
While the EU regulation is a welcome start, and our own policy questions allow us to form a basis for designing the best regulations suited to India’s priorities, AI will remain an area of significant concern as development and increased consumption of AI-related products and services rakes in. A lot remains unanswered by the European Union’s AI Act. Even in the present Indian context, there are several grey areas and sectors which need to have well established laws and implementation rules as AI becomes more ingrained within these industries.
For instance, how should regulation differ for each sector’s risks, be it financial markets, services, healthcare, or the military? To what extent would the Indian military and police be exempt from using live biometric data and social scoring as compared to private players? How will IP disputes be settled in the AI domain, and what qualifies as original IP in this space? These are important questions to be answered. As artificial intelligence becomes our present, all stakeholders need to protect the future, or risk the repercussions of a dysregulated AI ecosystem.
Top Stories of the Week
MCC Enforced, ECI Announces Elections
On March 17, 2024, the Election Commission of India (ECI) declared the commencement of elections for the 18th Lok Sabha and legislative elections in Andhra Pradesh, Arunachal Pradesh, Odisha, and Sikkim. Following the announcement, the Model Code of Conduct (MCC) was enforced nationwide, mandating that ruling parties, both at the central and state levels, refrain from utilizing their official capacity for campaign purposes. The MCC also prohibits incumbent governments from announcing policies, projects, or schemes that could potentially influence voting behavior.
According to the directives issued by the Election Commission, India will go to the polls from April 19 to June 1, in a marathon seven-phase exercise. Subsequently, the results of the elections will be officially declared on June 4, 2024, while the term of the 17th Lok Sabha is set to expire on June 16, 2024. In Bihar, West Bengal, and Uttar Pradesh, voting will take place in all seven phases. As the population continues to grow, the forthcoming parliamentary election is poised to become the second-longest polling exercise in India's electoral history. The longest duration recorded thus far was during the country's inaugural general election, which spanned over a five-month period from September 1951 to February 1952.
A week before the announcement of the elections and the MCC, the Union Government swiftly sanctioned several key initiatives. Some of the measures taken by the government include providing further extension to electric vehicle incentives and approvals for mega infrastructure projects worth Rs 2.3 trillion. The project approvals pertain to national highways, railway infrastructure, power generation and transmission, airports, and upgrades to State-owned ports.
Startup Mahakumbh Wraps Up with Prime Minister Modi's Address, Celebrating the Growth of India's Startup Ecosystem
Startup Mahakumbh, a first-of-its-kind event, brought India's startup ecosystem under one roof, with the central theme of 'Bharat Innovates'. The event featured mentorship clinics, pitch competitions, and a multi-track conference comprising leadership talks, panel discussions, workshops, and exciting activities for startups and future entrepreneurs.
The inaugural day saw government officials and industry veterans engage in insightful discussions on topics ranging from artificial intelligence (AI) to omnichannel strategy for Direct-to-Consumer (D2C) brands. Throughout the event, various thematic pavilions showcased innovations from startups across sectors like AI, agrotech, and gaming, to name a few. Day two featured sessions and keynote addresses of industry experts, aimed at empowering startups for a thriving future, and exploring the role of regulatory sandboxes in FinTech. Further to this, fireside chats with seasoned entrepreneurs provided insights to delegates on navigating success and failures in the startup journey. The final day culminated with Prime Minister Narendra Modi's address, who acknowledged the role of events like the Startup Mahakumbh in bolstering India’s startup ecosystem. He highlighted India’s exponential startup growth, and its phenomenal success in positioning the country as a global leader in innovation. Sessions on groundbreaking technologies like Natural Language Processing (NLP) & AI-based tool-Bhashini, esports, etc., enriched the event, fostering collaborations and showcasing India's entrepreneurial spirit.
In line with its aim of connecting startups with a spectrum of investors such as VCs, angel investors, family offices, HNIs, and potential corporate partners, the event attracted over 1000 startups, featuring 10+ thematic tracks, 1000+ investors, 500+ incubators & accelerators, 5000+ conference delegates, 10+ country delegations, and 40,000+ business visitors over three days. With various thematic pavilions covering various sectors, the event provided a robust platform for startups to engage with stakeholders.
A Few Good Reads
Arya Roy Bardhan and Soumya Bhowmick write about how China's influence in Maldives prompts concerns over debt and sovereignty for the latter.
Mathew Brooker points out the UK's economic downturn since Brexit, contrasting it with the EU's stability.
Sayatan Ghosh contends that the revelation of electoral bonds implies a quid pro quo situation, offering opposition an opportunity to disrupt the status quo.
Rana Banerji writes about how Shehbaz Sharif must tackle economic woes amidst civil unrest and military controversies in Pakistan.
China’s next agenda: modernizing military capabilities in cyberspace, space, and AI, say Harsh V Pant and Kalpit Mankika.