In today’s era of explosive information and data, AI is a very practical and efficient tool for us. Personally, as someone who is highly embedded in digital platforms in daily life, I am interacting with AI almost every moment, but many times I do not realise this. AI has gradually integrated into many different fields of our information life, including recommendation systems, search engines, chatbots, and facial recognition. For companies, AI is a symbol of innovation and convenience, while governments often regard it as an important part of economic growth and national competitiveness. As AI becomes highly integrated into our daily life, a very important question appears in front of me: how should AI be regulated so that it can retain its advantages of efficiency and innovation as much as possible, while also providing users with a high level of rights protection?
In recent years, many public discussions around AI have focused on the concept of “ethical AI.” From governments to companies, from conference themes to technology promotion, it can be seen in many different fields. Because of the popularisation of AI, fairness, accountability, transparency and safety have also become common contents in discussions about artificial intelligence. The more people care about these contents, the more it shows that developers, companies and policymakers have already realised that AI will have many impacts on society. However, I also think that ethical language sometimes makes people feel too reassured. We always say that AI should be “ethical,” but this does not mean that we have truly established rules and systems for how to correctly design and use AI.
Ethical principles are an important part of AI regulation, but only considering principles in this aspect is not comprehensive. Since in today’s society AI systems have already been able to affect rights, opportunities, public life and many other aspects, AI regulation needs to jump out of broad value claims and further develop into forms of regulation and supervision that are more practical, highly enforceable and able to serve internationalisation, further serving the public’s needs for AI regulation and the actual promotion of the development of “ethical AI,” rather than completely becoming a slogan with no chance to be truly analysed and improved.
One of the important reasons why I feel strongly about this is: I find that AI is not neutral. People often think it is objective because it relies on data and calculation. But I think this statement is not completely correct. We all know that AI is designed by human beings, trained with historical data, provided with resources by institutions, and finally placed into a society that already has unequal power. Therefore, AI will not only copy existing bias and inequality, but may even further strengthen them. UNESCO warns that the rapid development of AI has brought “profound ethical concerns,” because these systems may “embed biases, contribute to climate degradation, threaten human rights and more.” It also points out that these risks will add to “existing inequalities,” and cause “further harm to already marginalised groups” (UNESCO, 2021, para. 2).
UNESCO states that in November 2021 it developed “the first-ever global standard on AI ethics,” and that this standard is “applicable to all 194 member states of UNESCO” (UNESCO, 2021, para. 3). It also points out that “the protection of human rights and dignity is the cornerstone of the Recommendation,” while emphasising “transparency and fairness” and “the importance of human oversight of AI systems” (UNESCO, 2021, para. 3). I think these viewpoints are very important because they emphasise that we must regard AI as something that needs to be examined within the framework of rights, dignity and social harm, rather than only a technological innovation. In my view, the most valuable point of UNESCO’s framework is that it realises a problem often ignored in public discussion: high-level value principles do not automatically change technical practice. UNESCO believes that although “values and principles are crucial” for establishing an ethical AI framework, recent discussions have gradually emphasised that it is necessary to “move beyond high-level principles and toward practical strategies” (UNESCO, 2021, para. 4). This Recommendation “does just this by setting out eleven key areas for policy actions” (UNESCO, 2021, para. 4). In my view, this is almost one of the clearest statements showing that “ethics alone is far from enough,” and ethical language only has practical meaning when it truly becomes governance tools, public policy and institutional practice.
In this technological era, AI is gradually going deeper into our daily life. For example, recommendation systems will affect what content we see first, or in search engines some information sources will be ranked first, and some moderation tools will automatically label or delete certain content and so on. Therefore, the gap between principles and practice needs even more attention. In these situations, AI often subtly influences the visibility of some information, influences our attention, or our judgment of some things. As one of many users, I personally care about this point very much. I come into contact with AI almost every day, but especially when I scroll through platforms, search for information or use digital tools to study, I often do not know why certain content appears or disappears. After I became aware of these problems, I found that this situation will affect the content I see on social media, or affect the ranking of content in my search results. I may notice that some content suddenly becomes visible, certain topics appear repeatedly, and articles and videos are also recommended to me. I often do not understand the logic behind these results.
AI governance should not only be an abstract issue that governments or companies need to pay attention to. It is even more closely related to the daily experience of ordinary users. From the perspective of a user, a platform can of course claim that its system is fair, open, just and transparent, but these statements are difficult to verify in real life. This is what I truly feel as a user — what I actually feel is a kind of uncertainty. In daily life, I constantly interact with these systems that subtly influence my behaviour, but I almost do not get explanations of how they cause these influences. People are constantly classified and guided by these systems, but have no way to truly see the internal logic of these systems. Therefore, I think accountability is so important: if AI will shape people’s daily digital life in subtle but powerful ways, then people should not blindly believe these so-called ethical principles, but should build a practical mechanism to make these systems actually explainable, challengeable and accountable.
If in daily digital life, the filtered content only limits us inside an “information cocoon,” then what about when AI is used in more important scenarios? After AI leaves the daily life usage scenario, these problems often become more severe: imagine that once AI enters decision-making scenarios such as employment, education, law enforcement and healthcare, its errors and biases are no longer small troubles, but may directly affect people’s life choices, economic security, and may even produce legal consequences. In these situations, one fact we have to face is: AI is making judgments about people who are in existing unequal social structures. In these more important scenarios, people’s idea of using AI is often “as long as the system is efficient enough, it can operate with minimal human control.” But UNESCO emphasises that AI systems “should be auditable and traceable,” and that “oversight, impact assessment, audit and due diligence mechanisms” need to be established (UNESCO, 2021, para. 6).
Many people assume that as long as everyone realises ethical principles, then actual governance problems will be seen and solved. But this is often exactly the biggest problem in discussions about AI. Abdala et al. (2020) point out that if AI wants to “reduce existing inequalities without creating new divides,” then it “must be governable and interoperable.” This brief provides “a roadmap for the practical implementation of AI regulation” (Abdala et al., 2020, p. 1). I think this is a very valuable reference, because it shifts the discussion from “good intentions” to “how to implement.”
At the same time, it also raises another issue that I think is very important: fragmentation. It warns that “lack of coordination” will lead to “a fragmented governance landscape,” and this landscape will “exacerbate pre-existing inequities,” prevent people from obtaining “equal rights” across different jurisdictions, and create “new divides” between countries and regions (Abdala et al., 2020, p. 1). Because AI is clearly a cross-border technology, its models, platforms, datasets and infrastructure will not stay inside one country. A recommendation system designed in one country may affect the daily experiences of users in many other countries; a company may collect data globally but only be regulated at the local level. Therefore, we can see that when fragmented methods are used to govern AI, we cannot achieve complete protection measures.
Therefore, I think AI governance must also be understood as an international issue. The key to the problem is not only whether a certain country can effectively regulate AI within its own country, but whether governments and institutions of various countries can establish enough common standards to avoid a “race to the bottom.” Abdala et al. (2020) believe that common rules and standards can reduce risks at the international and market levels while supporting broader policy goals (Abdala et al., 2020). I am not saying that the world needs a unified global law, but that we can choose a more realistic and more feasible interoperability, meaning the establishment of governance frameworks that can cooperate across borders — governments and institutions of various countries sharing standards, sharing definitions and establishing international cooperation can still establish a minimum accountability foundation even when local practices are different.
In addition, governance cannot only be understood as the responsibility of government. Many of the most influential AI systems are developed, owned and deployed by private companies. In the Harvard Business Review, it clearly points out that AI regulation is coming, and organisations need to “prepare for the inevitable” (Candelon et al., 2021). The article also mentions that “for most of the past decade, public concerns about digital technology have focused on the potential abuse of personal data” (Candelon et al., 2021, para. 1). In my view, this point is important because today’s discussion around AI regulation did not appear suddenly. It actually developed from earlier discussions about data exploitation, platform power and the crisis of trust in digital systems.
From this, it can be seen that AI governance is even more a problem that organisations must face. Those companies that regard governance as a burden may not be prepared to deal with legal changes or declining trust. More importantly, they often take the attitude of “waiting until regulation comes.” But these systems are already shaping people’s daily life, and the most important thing in AI governance is to protect the public from harm. If organisations realise that regulation is inevitable, then the more responsible approach should be to establish stronger governance capacity as early as possible, such as internal review procedures, impact assessments and so on, and to establish truly effective handling channels when harm occurs.
For most people, ethics sounds flexible, positive, and seems not to hinder innovation. For organisations, it can also let organisations shape themselves into an image of responsibility, while not necessarily needing to undertake binding obligations. Therefore, the phrase “ethical AI,” which easily causes people to misunderstand, needs to be thought about: governance must go beyond ethics. Ethical commitments are easy to announce, but difficult to measure through anything; they can be selectively interpreted, inconsistently implemented, and may also be strategically used for public relations promotion. By comparison, governance raises questions that are more difficult, but also more important for us: Who has power? What standards must it meet? How should the harm caused be assessed? Who has the right to question a system? When institutions fail to comply with requirements, what consequences will arise?
I am not completely denying the concept of “ethical AI.” In fact, “ethical AI” itself is a very important principle. It provides us with direction: fairness, non-discrimination, dignity and so on. These concepts help us define what kind of AI is responsible. But the problem is that we only remain at this “concept,” instead of truly implementing them. “Ethical AI” should only be a starting point. Under the combined effects of commercial pressure, national interests, security demands and so on, its existence cannot play a constraining role.
In today’s competitive society, AI is also a field full of competition. It even involves international competition: governments of various countries all hope to attract investment and want to become leaders in technological development. In this situation, the AI field always puts the pursuit of growth first, rather than oversight. Abdala et al. (2020) warn that competition between countries to attract AI industries may weaken the supervision truly needed to reduce risks (Abdala et al., 2020). AI governance is definitely not only a technical problem. It is also a political, economic and geopolitical problem. If governments worry about losing competitive advantage, they may weaken or delay truly meaningful regulation; if companies can freely choose the loosest regulatory environment, then public protection will become even weaker.
Therefore, I think a serious AI governance plan needs at least three things. First, it needs a clear framework based on rights, dignity, fairness and accountability. We need to connect AI governance with human rights, inclusion and the protection of vulnerable groups. Second, it needs specific mechanisms that can implement principles. Third, it needs a certain degree of international coordination to reduce fragmentation and prevent a “race to the bottom.” Perhaps each of these elements is not sufficient when seen separately, but when combined together, they can form a solid AI governance foundation and can protect the public’s rights well while fully using the convenience of AI.
Perhaps someone will ask: since there are so many hidden risks, is AI good or bad? Actually, I have heard similar questions many times in discussions about AI, but I think this question is too simple. The influence of AI depends on how it is designed, who controls it, and the usage scenario, so we cannot simply judge whether it is good or bad. But imagine this: a recommendation system that amplifies false information, an automated hiring system that copies biased data. Once we continue to ignore the implementation problem of “ethical AI,” these often no longer belong to technical accidents.
AI has already become part of the infrastructure of contemporary life, and we can no longer rely only on voluntary commitments and vague ideals to govern it. I believe that when asking every person who comes into contact with daily digital life, most people can speak ethical language about AI. But the most important and difficult question for us now is: is society willing to establish a governance system that matches the current power of AI?
If we continue to allow AI to subtly influence people’s daily digital experience, influence the visibility of content on different platforms, and influence people’s judgment and ways of accessing information resources without strong oversight, then the problems that AI will bring will not be accidental events. The problems will become common and increasingly difficult to correct. AI is a very convenient and advanced technology, but its future depends on whether we can truly transform this “ethical AI” into enforceable practice.
References
Abdala, M. B., Ortega, A., & Pomares, J. (2020). Managing the transition to a multi-stakeholder artificial intelligence governance. Global Solutions Initiative. https://www.global-solutions-initiative.org/publication/managing-the-transition-to-a-multi-stakeholder-artificial-intelligence-governance/
Candelon, F., Charme di Carlo, R., De Bondt, M., & Evgeniou, T. (2021, September–October). AI regulation is coming. Harvard Business Review. https://hbr.org/2021/09/ai-regulation-is-coming
UNESCO. (2021). Recommendation on the ethics of artificial intelligence. UNESCO. https://www.unesco.org/en/artificial-intelligence/recommendation-ethics
AI Declaration
I used ChatGPT to support the development of this blog post. The AI tool was used to help me understand the topic, organise my ideas, translate and refine some sentences, and improve the clarity of my writing. I also used it to help check APA 7 referencing style and to better understand the sources used in this blog.
The main argument, final structure, source selection, and personal reflections were reviewed and edited by me. I take responsibility for the final content, interpretation of sources, and any errors in this submission.
Be the first to comment