In recent years, AI has slowly entered many decision-making scenarios in daily life, and many times we are not even aware of its existence. It will affect which news we see first, which content will be marked as suspicious, and will also affect recommendations, recruitment, and even trust and risk judgment in some systems. Because of this, more and more people begin to rely on AI. The reason is actually very simple: it runs fast, costs relatively low, and can handle far more information than humans. In a digital society that pursues efficiency, it seems to be a reasonable choice to involve or even dominate decision-making.But the problem is that making decisions is not only about speed or data, but also about power, responsibility and value judgment. When more and more decisions begin to be made by AI rather than by people, a more important question arises: what will happen if something goes wrong? More importantly, who should be responsible for this mistake?
I think AI can be an auxiliary tool for human decision-making, but it should not replace human judgement in high-risk situations. The reason is not complicated: efficiency is not equal to fairness, automation is not equal to neutrality, and data-driven is not equal to true transparency. Even if a system can give an answer in a few seconds, it does not mean that the answer is reasonable, legitimate, or can be questioned and corrected. UNESCO’s proposal on artificial intelligence ethics emphasises that the development of AI must take human rights, dignity, transparency, fairness and human supervision as the core, which in itself shows that AI cannot be understood as a “perfect referee” that can operate independently from human responsibility (UNESCO, 2021).
The most common sentence of people who support AI decision-making is: “Machines are at least less emotional and more stable than people.” This sentence sounds very convincing, because human beings are indeed tired, eccentric, negligent, and will also make wrong judgements under pressure. In contrast, AI always seems to be an orderly and efficient, and can process a large amount of information in a very short time. From platform governance to enterprise management to government services, more and more institutions want to use AI to improve efficiency and reduce costs. Especially in the digital platform environment, with a huge amount of content and extremely fast updates, enterprises tend to regard AI as a “must-rely” solution. Th problem is that it is easy for us to mistake “fast handling” for “right handling”. This is the most dangerous part of AI decision-making: it often covers up the injustice in the process with speed and the value choices behind it with automation.
First of all, AI is not neutral. Many people have the habit of seeing it as a pure technical tool, as if it has no position in itself and does not have prejudice. However, the reality is not that simple. The data used by AI comes from humans, and the goal of the model is set by people. How much risk the system can withstand depends on the judgment of the institution, and even which scenario it ultimately applies to is essentially the result of artificial choice. In other words, AI will never operate in a “vacuum environment”.If the dataitself that the system learns is biased, these inequalities can be further widened. Similarly, if a platform emphasizes efficiency and traffic more and underestimates risks and user rights, the AI developed or used by the platform is more likely to be biased towards “convenient management” rather than substantial fairness. It is. The main reason why many people think AI is neutral is that these man-made factors are hidden in complex technical processes, making it difficult for ordinary users to notice and question them.
Secondly, AI decision-making often lacks sufficient transparency. Many people have experienced similar situations: the account is suddenly restricted, the content is inexplicably suppressed, and an application is rejected by the system, but you don’t know why. The platform may tell you that this is the result of the system’s automatic detection, but it will not really explain the basis of judgement. For users, this is a very typical experience of “being decided but unable to appeal”. One of the most basic requirements of digital governance is to make the operation of power understandable, supervised and corrected. But when AI decision-making is packaged into an inexplicable “black box”, this basic requirement is weakened. Transparency is not only a technical issue, but also a democratic and rights issue. Because a decision that cannot be explained, no matter how efficient it is, it is difficult to be considered legitimate (UNESCO, 2021).
Third, the most serious problem of AI decision-making is that it will dilute the responsibility. Traditionally, if a person makes a wrong decision, we at least knowho to ask: whether the manager’s judgement is wrong, or there is something wrong with the institutional rules. However, if a decision is blamed on the “algorithm”, the responsibility becomes ambiguous. Developers may say that they are simply providing tools, the platform may say that the system is running automatically, and administrators may say that they are simply referring to the output of the model. . As a result, what is actually affected is not the subject to be held accountable, but a series of technical processes that have no clear responsibility. For companies, such an arrangement is rather very convenient. Because “system error” sounds easier to avoid responsibility than “we made a mistake”. However, from a governance perspective, this is very dangerous. In a system that affects the rights, reputation, and safety of users, if the boundaries of responsibility are not clear, the harm becomes normal and accountability is peripheral.
Of course, some people will argue that human beings themselves are not perfect. In reality, people will make judgements with emotions, and will also be affected by prejudice, lack of experience and institutional pressure. In this case, why can’t AI gradually replace human beings? This rebuttal seems reasonable, but it ignores a key difference: human mistakes can be discussed in the social and institutional context, while AI’s mistakes are often technicalised and naturalised, and finally become irresponsible results. When a person makes a wrong judgement, we can at least ask him what he is based on, whether he violates the procedure, and whether he should take responsibility; but when the error comes from AI, many institutions will package it as “system problem”, “model misjudgement” or “automatic process abnormality”, making it more difficult for those who are really affected to seek explanation. And remedy. That is to say, the biggest risk of AI is not only that it will make mistakes, but also that it will make mistakes seem more normal and difficult to challenge.
In addition, we should be wary of expressing AI decision-making as “more objective”. The so-called “objectivity” is often because the machine cannot express emotions, but that does not mean that it is truly fair. In many cases, AI only converts social prejudices that already existed into a more covert and institutionalized form. Systems trained based on historical data often learn the unequal patterns that already exist in real society, rather than learning ideal fair rules. In this way, rather than eliminating prejudice, AI may repackage prejudice in the name of “technology neutrality”. For general users, this prejudice becomes even more difficult to find. Because the output of a machine often appears to be a precisely calculated “scientific result”. This is why “algorithm neutrality” is a myth that should always be repeatedly questioned in the debate on digital policy and governance.
Recent examples illustrate this point well. According to a study published by the Australian Institute of Criminology in 2026, many Australians are concerned about becoming victims of AI-related crimes and harm, and will be impersonated, scammed, tracked, or seriously counterfeited. It includes problems such as encountering. This shows that the general public does not see AI risk as a problem in the distant future, but as a safety and governance issue in the real world (Australian Institute of Crimino Logy [AIC], 2026). In other words, AI is not just a tool for improving efficiency, but is also changing people’s understanding of risks, trust, and responsibility in the digital environment. As more and more people start worrying about whether AI will hurt me, it won’t be easy for us to leave the decision-making power entirely to it.
A relatively intuitive example, there is a controversy over Grok on the X platform in early 2026. The media points out that this generative AI tool can generate explicit images without permission and may even involve women and minors. What is really noteworthy here is not only the technology itself “what can do”, but also the platform, despite the lack of sufficient protection, it has introduced its function into the actual social environment, and only after the problem has already had an impact. It is to start dealing with itThis example shows that AI is not a “neutral system” that can exist independently of social relations. When it is incorporated into the platform, market, and traffic-oriented operational logic, the results directly affect real people, especially vulnerable people. If the platform simply attributes the problem to “user tool abuse” afterwards, it is obviously insufficient. A more reasonable way is to set clear boundaries, restrictions, and protection mechanisms in advance at the system design stage, rather than responding passively after a problem occurs (ABC News, 2026).
Therefore, the key question is not “whether AI is useful”, but “in what situation it should play a role and how much it plays”. In some low-risk scenarios, such as sorting information, improving efficiency, providing advice or assisting screening, AI can indeed bring convenience. But when decision-making involves security, rights, identity, reputation, and even brings practical consequences, the final decision cannot be simply handed over to AI.The reason is also clear: high-risk decision-making not only depends on data processing ability, but also on interpretation ability, understanding of specific situations, ethical judgment, and the ability to be responsible for results. These are the most core parts of human social governance. If we leave these to AI, we will lose not only human judgment itself, but also the transparency and accountability of the power operation.
This does not mean that we must reject AI. A more reasonable view is that AI can participate in decision-making, but it should not replace human beings; it can help, but it cannot be a tool to cover up responsibility; it can improve efficiency, but it cannot override fairness and rights.
From the perspective of digital governance, a more responsible framework should include at least several basic principles. First of all, in high-risk decision-making scenarios, the mechanism of human supervision and manual review must be retained. Secondly, platforms and related institutions need to be able to clearly explain how AI affects decision-making results, instead of hiding the process in technology. Thirdly, when AI causes harm, the responsibility must be clearly implemented to a specific organization or institution, rather than simply attributed to “system problems”. Finally, when formulating governance rules, priority should be given to the protection of user rights, rather than the reasonable default technology expansion itself.Only under the premise that these conditions are really implemented can AI become a tool to serve the society, instead of allowing society to adapt to the limitations of technology itself.
In the end, the reason why AI is fascinating is that it promises a faster, smoother and more trouble-free future. But governance never only pursues “saving trouble”. In a truly mature digital society, it is not to give judgement to the machine to complete the task, but to know when to put people back to the centre of decision-making. AI may be able to get answers faster than people, but speed is not justice, and automation is not responsibility. In the face of the intelligent system that is more and more deeply embedded in social life, what we need to be most alert about is not whether machines will become too smart, but whether human beings will give up their responsibilities too easily.
Reference list
ABC News. (2026, January 3). AI chatbot Grok under fire over complaints it let users generate explicit images. ABC News.
Australian Institute of Criminology. (2026). Perceived risk of victimisation by artificial intelligence enabled crimes (Statistical Bulletin 51). AIC.
UNESCO. (2021). Recommendation on the ethics of artificial intelligence. UNESCO.
Image source: Unsplash (free stock image)
Be the first to comment