Abstract
From assisting in writing school essays to frequently making headlines –Mark Zuckerberg bought GPU- Artificial Intelligence is raising questions about whether it will be as transformative as penicillin or as addictive and abusive as OxyContin. The European Union Artificial Intelligence Act makes the very first attempt to regulate Artificial Intelligence. This Act, being the first of its kind, sets an example for many countries and organizations across the world. This paper delves into the intricacies of the European Union Artificial Intelligence Act and the core features that can pave a way for regulating it by other counterparts.
Key Words: Artificial Intelligence, European Union Artificial Intelligence Act, European Union, Classification of AI based on Risk,
Introduction:
The term “Artificial Intelligence” was first coined by John McCarthy in 1956. Intelligentsia across the world, various organizations have tried to define AI to a certain. However, there was no specific definition that was agreed upon either to define AI (hereon referred to as Artificial Intelligence) or regulate AI. On March 14, 2024, the groundbreaking European Union Artificial Act was enforced throughout European Union countries, which regulated AI on different levels and at the same time defined AI in its parlance.
The paper put forth the key features of the act such as the purpose of the Act, the definition of AI, the risk-based approach categorization, the regulation laid down to crave AI to be human- centric, and the GPAI. The research paper aims to look into how these laws can used globally and what are the things that can also be specified in their respective Acts or regulations. The paper in the later part highlights the AI ecosystem in terms of India.
Research Methodology:
The paper is descriptive and is a collation of both primary and secondary resources. The primary resource for the research paper is the European Act and the secondary resources used for the research paper are articles and research papers.
Review of Literature:
This part highlights the journal/ new articles/ research papers, blogs and websites.
The European parliament has provided with the brief articles provide compressive understanding in relation to various provision of the European Union Act.
Report on Artificial Intelligence Policy and the European Union A Look Across the Atlanic by Sabine Neschke and John Babak Soroushian explored the lacuna in the European Union Act.
The European Union AI: Act Next steps and issues for building international cooperation by Josh Meltzer and Aaron Tielemans has elucidated how the different stakeholders, organization, states are working in regulating AI at the nascent stage.
Niti Ayog’s National Strategy for Artificial Intelligence pressed the need for an AI regulation in India and how the AI Ecosystem can be improved.
Aim of the European Union Artificial Intelligence Act:
Firstly the main purpose of the Act as stated in Act –
“The purpose of this Regulation is to improve the functioning of the internal market by laying down a uniform legal framework in particular for the development, the placing on the market, the putting into service and the use of artificial intelligence systems (AI systems) in the Union, in accordance with Union values, to promote the uptake of human-centric and trustworthy artificial intelligence (AI) while ensuring a high level of protection of health, safety, fundamental rights as enshrined in the Charter of fundamental rights of the European Union (the ‘Charter’), including democracy, the rule of law and environmental protection, against the harmful effects of AI systems in the Union, and to support innovation. This Regulation ensures the free movement, crossborder, of AI-based goods and services, thus preventing Member States from imposing restrictions on the development, marketing and use of AI systems, unless explicitly authorised by this Regulation.”
In simple terms, it is set forth to regulate Artificial Intelligence systems across the 27 European Union member countries. It laid down the rules as to how AI can be sold, and used and at the same time ensure safety along with safeguarding fundamental rights. The Act allows services of AI systems to move freely without each country making its own rules unless it is allowed by the regulation.
Key features of the AI Act:
- DEFINITION OF AI :
The Act defined the AI as-
“An AI system is a machine-based system designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments.”
The Act details one big thing that sets AI systems apart from other traditional software is the capability of figuring out things. It specifies that AI systems can learn from data to obtain the goal or use logic and knowledge to solve problems. This goes beyond simply figuring out information- it associates learning, thinking and modeling. The AI system is a machine-based system that works on its own and knows how it adapts over time. It configures things based on the information it receives from either its users or creators. It uses this information to create content, make predictions, or assist in making decisions that can affect virtual or real scenarios.
- CLASSIFYING AI BASED ON RISK INVOLVED:
One of the underscoring features of the AI Act is the classification of AI systems based on their risk-causing factors. To make it easy to understand the AI systems are categorized into 4 levels, the primary one being the one with the slightest risk or negligible risk which has slight involvement by the regulations of the AI Act. The prohibited ones are those having the highest involvement with legal regime and they are classified as unacceptable AI systems which are completely prohibited by the Act. The mid-level are the ones having high risk and limited risks are made to comply extensively and limitedly respectively with the Act.
Source: Europe fit for Digital Age: Artificial Intelligence
- Prohibition of certain AI Practices: To break it down the unacceptable risked AIs or prohibited AI systems are those that are highly likely to pose a threat to a person’s fundamental right or exploit their vulnerability. The AIs that use social scoring that is using AI to evaluate or group individuals based on their social behavior that is inferred from their online activity, leading to detrimental treatment are prohibited under this Act. In addition to this, the AIs that collect biometrics to get sensitive data such as race, religious beliefs, or sexual orientation and AIs that intrude on people’s privacy and protective policing used for forecasting individual behaviors are prohibited under this Act and some exceptions do apply.
- AI systems included in High-risk :AI systems that are categorized as high-risk include those used in critical infrastructures such as transportation, educational and vocational training that can effect access to education and careers, safety components like robust-assisted surgery, employment and management of workers like CV-sorting software, vital public and private services such as credit- scoring for loans, law enforcement activities like evaluating evidence, migration and border control management, and administration of justice and democratic process like using AI to search court ruling.
Before High-risk AI systems can be marketed, they should comply with the strict conditions set out by the Act alongside the providing assurance of:
- Using high-quality data to minimize discrimination
- Keeping records of activities for accountability
- Assessing and reducing the risks
- Having human oversight to minimize the risks
- High levels of security, reliability, and accuracy.
- Providing exhaustive, precise, and detailed documents to authorities and users.
Furthermore, all the remote biometric identification AIs are considered under this category and they must adhere to strict requirements. They are prohibited from using in public places for law enforcement purposes, except in certain exemplary conditions like used for finding missing children or preventing terrorist threats, which can only be done by the approval of a competent judicial or independent body and with limits on time, location, and the data searched.
- The limited or transparency risk: The AI systems that interact directly with people pose risks specifically in cases of deep-fake which is a system that generates or manipulates images, audio, or videos, according to this Act must disclose the contents generated or manipulated with the help of AI, other than in limited cases. Providers of AI who supply in mass quantity synthetic content must implement and ingrain the most robust techniques like watermarks to mark the output that has been generated by the AI and not by the human.
- The minimum risk categorized AI systems are those the negligible effect by the Act and which cause minimal effect to people.
- General Purpose of AI (GPAI) – The other key feature of the Act is it set out the guidelines for GPAI which include Generate AI. Generate AI systems are not traditional AI systems. The Generate AI systems have the capacity to generate new AI systems and it is similar to the black box where the sender (the one who gives input/ prompt) will also not be in a position to accurately state to what extent and how the output can be used. Even the GPAI models are categorized into two segments – GPAI which do not pose significant are allowed to be freely with their parameters for public accessibility. The latter are those with high- impact GPAI that are likely to pose risk should include model parameters, dataset quality, or use base.
- Setting out the principle of Trustworthy AI
As stated in the purpose of the Act, the AI systems should preserve fundamental right and should be ensure that they are human-centric in nature. To attain these the European Union Act listed out the requirements for AI to be addressed as trustworthy AI. Before AI can be said to comply with the requirements to be trustworthy AI it should ensure it reflects the 4 ethical principles-
- Respect for human autonomy
- Prevention of harm
- Fairness
- Explicability
Primarily these ethical principles are to be followed by an AI system.
The requirements enlisted should be followed by developers, deployers, end-users, and society at large. Here-
Developers are creators of AI systems
Deployers are the users, and
The end-users are those who interact with them.
The Act further stated that the developers must integrate requirements into design and development process. Deployers must ensure systems meet requirements. The end-users should be informed and should be able to acquire compliance.
The 7 requirements which are stated by the Act are listed in the below picture for better understanding:
Suggestions:
-Regarding the definition AI according to the Act still needs to be widened from its narrow definition as many stakeholders argued that it should mediate between the “software-based automated system” to that of those that “ might have an impact on fundamental rights”.
-The risk categorization of AI strikes a balancing effort between soft law (laws that are flexible and not stringent) in the matter of the minimum-risk, limited-risk AI and hard law (stringent laws) in the matter of High-risk AI system setting a new example for tailor-made laws instead of setting for the one size fit for all approach.
The Potential Global Impact of the European Union Artificial Intelligence Act:
-Prior to the enforcement of the EU AI Act, there were many talks regarding creating a global regulating AI body. Canada made a mark by becoming the first country to adopt an AI strategy. But still, there was a need for global corporations among nations for regulations, divergent approaches for AI governance, and effective cooperation among countries this Act will probably set an example for AI governance among countries for building an international corporation for AI.
-Antecedent the EU AI Act there hasn’t been an apt definition for AI that was globally accepted. But the Act tried to define AI, this can give leeway to other countries to improvise or stick to the definition of AI.
The AI Ecosystem in India-
In India as of now there is no separate statue to regulate AI. There are plans to regulate AI.There are still many strategies in pipeline. The NITI Aayog has published National Strategy Artificial Intelligence which is the current guiding light for the AI regulating modals in India.
The Niti Aayog’s National strategy for Artificial Intelligence has come up with definition of AI, a new classification which has broadened the scope compared to EU AI Act’s risk based approach. It has included various sectors such as health, agriculture, education, smart cities and infrastructure, smart mobility and transportation.
– In the Niti Aayog’s National Strategy for Artificial Intelligence it was enunciated that there is a dire need for incentivizing in AI research and AI ecosystem. The key point to be adopted by India in governance of AI is to invest in AI-based innovations and projects and to work on digital governance. It was well forecasted by the EU much before the governance of AI when it started with the Civil Law Rules on Robotics, European Parliament, 2017(albeit it did not mention the AI) to the enforcement EU Act it is clear that there has been a lot of resources investment in developing a regulatory framework for AI.
– In India parlance if there is any dispute related to AI, the IT Act or the data protection related Acts are used which has limited powers overAI. This also creates ground for the need of effective regulation of AI ecosystem for making a better use of AI and preventing abuse or harmful effects that can be caused by AI.
As pointed out Niti Aayog guidelines India needs to strengthen the ecosystem by collaborating with research organizations, investing in human resources and AI Industry economically are to be set in action.
Conclusion:
In the fast-growing technological word, AI innovation is groundbreaking such as the invention of percilline. But any greatest thing can’t be set free without regulation. The European Union Artificial Intelligence Act is like San Marino. Still there’s a long way to go. For sure this Act is a stepping stone. But there’s still a need for more comprehensive regulations. The AI systems are to be regulated from time to time and other nations should adopt them in a tailored made manner based on their aspirations and needs.
- RAJEEV HARSHA LAKKA FROM UNIVERSITY COLLEGE OF LAW, OSMANIA UNIVERSITY