Hard Law or Soft Law? –Global AI Regulation Developments and Regulatory Considerations

Hard Law or Soft Law?
–Global AI Regulation Developments and Regulatory Considerations

2023/08/18

Since the launch of ChatGPT on November 30, 2022, the technology has been disrupting industries, shifting the way things used to work, bringing benefits but also problems. Several law suits were filed by artists, writers and voice actors in the US, claiming that the usage of copyright materials in training generative AI violates their copyright.[1] AI deepfake, hallucination and bias has also become the center of discussion, as the generation of fake news, false information, and biased decisions could deeply affect human rights and the society as a whole.[2]

To retain the benefits of AI without causing damage to the society, regulators around the world have been accelerating their pace in establishing AI regulations. However, with the technology evolving at such speed and uncertainty, there is a lack of consensus on which regulation approach can effectively safeguard human rights while promoting innovation. This article will provide an overview of current AI regulation developments around the world, a preliminary analysis of the pros and cons of different regulation approaches, and point out some other elements that regulators should consider.

I. An overview of the current AI regulation landscape around the world

The EU has its lead in legislation, with its parliament adopting its position on the AI ACT in June 2023, heading into trilogue meetings that aim to reach an agreement by the end of this year.[3] China has also announced its draft National AI ACT, scheduled to enter its National People's Congress before the end of 2023.[4] It already has several administration rules in place, such as the 2021 regulation on recommendation algorithms, the 2022 rules for deep synthesis, and the 2023 draft rules on generative AI.[5]

Some other countries have been taking a softer approach, preferring voluntary guidelines and testing schemes. The UK published its AI regulation plans in March, seeking views on its sectoral guideline-based pro-innovation regulation approach.[6] To minimize uncertainty for companies, it proposed a set of regulatory principles to ensure that government bodies develop guidelines in a consistent manner.[7] The US National Institute of Standards and Technology (NIST) released the AI Risk Management Framework in January[8], with a non-binding Blueprint for an AI Bill of Rights published in October 2022, providing guidance on the design and use of AI with a set of principles.[9] It is important to take note that some States have drafted regulations on specific subjects, such as New York City’s Final Regulations on Use of AI in Hiring and Promotion came into force in July 2023.[10] Singapore launched the world’s first AI testing framework and toolkit international pilot in May 2022, with the assistance of AWS, DBS Bank, Google, Meta, Microsoft, Singapore Airlines, etc. After a year of testing, it open-sourced the software toolkit in July 2023, to better develop the system.[11]

There are also some countries still undecided on their regulation approach. Australia commenced a public consultation on its AI regulatory framework proposal in June[12], seeking views on its draft AI risk management approach.[13] Taiwan’s government announced in July 2023 to propose a draft AI basic law by September 2023, covering topics such as AI-related definition, privacy protections, data governance, risk management, ethical principles, and industrial promotion.[14] However, the plan was recently postponed, indicating a possible shift towards voluntary or mandatory government principles and guidance, before establishing the law.[15]

II. Hard law or soft law? The pros and cons of different regulatory approaches

One of the key advantages of hard law in AI regulation is its ability to provide binding legal obligations and legal enforcement mechanisms that ensure accountability and compliance.[16] Hard law also provides greater legal certainty, transparency and remedies for consumers and companies, which is especially important for smaller companies that do not have as many resources to influence and comply with fast-changing soft law.[17] However, the legislative process can be time-consuming, slower to update, and less agile.[18] This poses the risk of stifling innovation, as hard law inevitably cannot keep pace with the rapidly evolving AI technology.[19]

In contrast, soft law represents a more flexible and adaptive approach to AI regulation. As the potential of AI still remains largely mysterious, government bodies can formulate principles and guidelines tailored to the regulatory needs of different industry sectors.[20] In addition, if there are adequate incentives in place for actors to comply, the cost of enforcement could be much lower than hard laws. Governments can also experiment with several different soft law approaches to test their effectiveness.[21] However, the voluntary nature of soft law and the lack of legal enforcement mechanisms could lead to inconsistent adoption and undermine the effectiveness of these guidelines, potentially leaving critical gaps in addressing AI's risks.[22] Additionally, in cases of AI-related harms, soft law could not offer effective protection on consumer rights and human rights, as there is no clear legal obligation to facilitate accountability and remedies.[23]

Carlos Ignacio Gutierrez and Gary Marchant, faculty members at Arizona State University (ASU), analyzed 634 AI soft law programs against 100 criteria and found that two-thirds of the program lack enforcement mechanisms to deliver its anticipated AI governance goals. He pointed out that credible indirect enforcement mechanisms and a perception of legitimacy are two critical elements that could strengthen soft law’s effectiveness.[24] For example, to publish stem cell research in top academic journals, the author needs to demonstrate that the research complies with related research standards.[25] In addition, companies usually have a greater incentive to comply with private standards to avoid regulatory shifts towards hard laws with higher costs and constraints.[26]

III. Other considerations

Apart from understanding the strengths and limitations of soft law and hard law, it is important for governments to consider each country’s unique differences. For example, Singapore has always focused on voluntary approaches as it acknowledges that being a small country, close cooperation with the industry, research organizations, and other governments to formulate a strong AI governance practice is much more important than rushing into legislation.[27] For them, the flexibility and lower cost of soft regulation provide time to learn from industries to prevent forming rules that aren’t addressing real-world issues.[28] This process allows preparation for better legislation at a later stage.

Japan has also shifted towards a softer approach to minimize legal compliance costs, as it recognizes its slower position in the AI race.[29] For them, the EU AI Act is aiming at regulating Giant Tech companies, rather than promoting innovation.[30] That is why Japan considers that hard law does not suit the industry development stage they’re currently in.[31] Therefore, they seek to address legal issues with current laws and draft relevant guidance.[32]

IV. Conclusion

As the global AI regulatory landscape continues to evolve, it is important for governments to consider the pros and cons of hard law and soft law, and also country-specific conditions in deciding what’s suitable for the country. Additionally, a regular review on the effectiveness and impact of their chosen regulatory approach on AI’s development and the society is recommended.

[1] ChatGPT and Deepfake-Creating Apps: A Running List of Key AI-Lawsuits, TFL, https://www.thefashionlaw.com/from-chatgpt-to-deepfake-creating-apps-a-running-list-of-key-ai-lawsuits/ (last visited Aug 10, 2023); Protection for Voice Actors is Artificial in Today’s Artificial Intelligence World, The National Law Review, https://www.natlawreview.com/article/protection-voice-actors-artificial-today-s-artificial-intelligence-world (last visited Aug 10, 2023).

[2] The politics of AI: ChatGPT and political bias, Brookings, https://www.brookings.edu/articles/the-politics-of-ai-chatgpt-and-political-bias/ (last visited Aug 10, 2023); Prospect of AI Producing News Articles Concerns Digital Experts, VOA, https://www.voanews.com/a/prospect-of-ai-producing-news-articles-concerns-digital-experts-/7202519.html (last visited Aug 10, 2023).

[3] EU AI Act: first regulation on artificial intelligence, European Parliament, https://www.europarl.europa.eu/news/en/headlines/society/20230601STO93804/eu-ai-act-first-regulation-on-artificial-intelligence (last visited Aug 10, 2023).

[4] 中國國務院發布立法計畫 年內審議AI法草案，經濟日報(2023/06/09)，https://money.udn.com/money/story/5604/7223533 (last visited Aug 10, 2023).

[5] id

[6] A pro-innovation approach to AI regulation, GOV.UK, https://www.gov.uk/government/publications/ai-regulation-a-pro-innovation-approach/white-paper (last visited Aug 10, 2023).

[7] id

[8] AI RISK MANAGEMENT FRAMEWORK, NIST, https://www.nist.gov/itl/ai-risk-management-framework (last visited Aug 10, 2023).

[9] The White House released an ‘AI Bill of Rights’, CNN, https://edition.cnn.com/2022/10/04/tech/ai-bill-of-rights/index.html (last visited Aug 10, 2023).

[10] New York City Adopts Final Regulations on Use of AI in Hiring and Promotion, Extends Enforcement Date to July 5, 2023, Littler https://www.littler.com/publication-press/publication/new-york-city-adopts-final-regulations-use-ai-hiring-and-promotionv (last visited Aug 10, 2023).

[11] IMDA, Fact sheet - Open-Sourcing of AI Verify and Set Up of AI Verify Foundation (2023), https://www.imda.gov.sg/-/media/imda/files/news-and-events/media-room/media-releases/2023/06/7-jun---ai-annoucements---annex-a.pdf (last visited Aug 10, 2023).

[12] Supporting responsible AI: discussion paper, Australia Government Department of Industry, Science and Resources,https://consult.industry.gov.au/supporting-responsible-ai (last visited Aug 10, 2023).

[13] Australian Government Department of Industry, Science and Resources, Safe and responsible AI in Australia (2023), https://storage.googleapis.com/converlens-au-industry/industry/p/prj2452c8e24d7a400c72429/public_assets/Safe-and-responsible-AI-in-Australia-discussion-paper.pdf (last visited Aug 10, 2023).

[14] 張璦，中央通訊社，AI基本法草案聚焦隱私保護、應用合法性等7面向 擬設打假中心，https://www.cna.com.tw/news/ait/202307040329.aspx (最後瀏覽日：2023/08/10)。

[15] 蘇思云，中央通訊社，2023/08/01，鄭文燦：考量技術發展快應用廣 AI基本法延後提出，https://www.cna.com.tw/news/afe/202308010228.aspx (最後瀏覽日：2023/08/10)。

[16] supra, note 13, at 27.

[17] id.

[18] id., at 28.

[19] Soft law as a complement to AI regulation, Brookings, https://www.brookings.edu/articles/soft-law-as-a-complement-to-ai-regulation/ (last visited Aug 10, 2023).

[20] supra, note 5.

[21] Gary Marchant, “Soft Law” Governance of Artificial Intelligence (2019), https://escholarship.org/uc/item/0jq252ks (last visited Aug 10, 2023).

[22] How soft law is used in AI governance, Brookings,https://www.brookings.edu/articles/how-soft-law-is-used-in-ai-governance/ (last visited Aug 10, 2023).

[23] supra, note 13, at 27.

[24] Why Soft Law is the Best Way to Approach the Pacing Problem in AI, Carnegie Council for Ethics in International Affairs,https://www.carnegiecouncil.org/media/article/why-soft-law-is-the-best-way-to-approach-the-pacing-problem-in-ai (last visited Aug 10, 2023).

[25] id.

[26] id.

[27] Singapore is not looking to regulate A.I. just yet, says the city-state’s authority, CNBC,https://www.cnbc.com/2023/06/19/singapore-is-not-looking-to-regulate-ai-just-yet-says-the-city-state.html#:~:text=Singapore%20is%20not%20rushing%20to,Media%20Development%20Authority%2C%20told%20CNBC (last visited Aug 10, 2023).

[28] id.

[29] Japan leaning toward softer AI rules than EU, official close to deliberations says, Reuters, https://www.reuters.com/technology/japan-leaning-toward-softer-ai-rules-than-eu-source-2023-07-03/ (last visited Aug 10, 2023).

[30] id.

[31] id.

[32] id.