[Dong-cheol Kim (김동철), a professor in the Department of AI Applications at Hansung University] When I lecture university students on ethics in the AI era, I end up talking only about ethics from the user’s perspective. Since those attending are users, there is no choice but to cover users’ ethics. Bias, limitations and functional convenience in AI developed so far mean results vary depending on how it is used, and it can cause unintended harm.

A prevailing view is that users generally have to bear responsibility. This is likely because there is still a large gap in understanding of AI between developers and users. The impact of generative AI based on massive language models spread around the world in an instant without being completed. It showed a record phenomenon of taking less than a week to surpass 100 million users.

The P in ChatGPT is an English abbreviation for Pre-Trained. It means that, in the first stage, it was trained by collecting data that could be learned. It is therefore an intermediate-goods type of program for which answers that fit every situation cannot be expected. Recent cases known to the public show that problems related to AI tend to concentrate harm on vulnerable groups.

This is compiled from multiple angles in Empire Power, Capital Labor (2026), a book by Karen Hao, a journalist who reports on AI’s impact on society. Beyond individual issues, serious social problems are also being caused. For example, automated software purchased and used by police or financial institutions was found to entrench discrimination by race, gender and class. It was even reported by an independent investigative media outlet that an algorithm used in the U.S. criminal justice system classifies an innocent Black person as higher risk than a white person with many criminal records.

What happens between vendors providing AI services and users is bound to create problems at some point, and once it happens it becomes known to people in a relatively short time. But it is difficult to know what kinds of logical and ethical clashes happen inside companies that build AI services, and what OpenAI, Google and Microsoft, among others, are doing behind the scenes to improve the completeness of AI services. There is also no way to know whether such matters demand accountability from vendors with the same weight as the responsibility users must bear for ethical use today.

Looking back, in the last century it emerged that children’s labour was exploited to mine diamonds in Sierra Leone, Angola and Congo, raising ethical issues across the entire diamond supply chain. TikTok is being designed to increase addictiveness through algorithms, drawing excessive user time and attention and raising concerns it could cause mental health problems among teenagers. Britain’s consulting firm Cambridge Analytica collected data on tens of millions of Facebook users without consent and used it for political ad targeting, and it emerged that this involved unauthorised use of personal information and affected democratic processes. In related court rulings, Cambridge Analytica went bankrupt and Facebook was fined 500,000 pounds. Such things are happening in fields related to AI as well, almost as if by copy and paste.

To develop AI competitively and operate it profitably, data centres made up of GPU servers are absolutely necessary. High-performance GPU servers are as noisy as their name suggests and require cooling water to lower the enormous heat they generate. AI-specialised data centres also require power on a scale comparable to the electricity use of an average city. Carbon emitted from these data centres can have global influence.

Businesses in Silicon Valley in the United States are trying to build data centres in countries such as Chile and Bolivia. For leaders in these countries in the global South, attracting such projects becomes a way to show political capability. For people who have suffered drought for years, even drinking water could end up being used as cooling water for data centres, raising fears of a man-made environmental disaster.

In Kenya and Venezuela, another countries in the global South, citizens driven by unemployment and hyperinflation are working in content moderation jobs that create data used by European car companies and U.S. AI firms for self-driving and AI training. They receive low pay of about $2 an hour and suffer heavy mental strain from repeatedly filtering sexual, violent and insulting content. But their non-regular worker status makes it difficult even to get proper treatment.

Situations are unfolding that run counter to OpenAI’s early philosophy that it would benefit the whole world with AI. Ethical problems across the entire AI supply chain are still ongoing, and developing countries and vulnerable groups in particular are bearing the damage as it is. It may seem like someone else’s problem, but because the ethical collapse of the AI ecosystem will ultimately come back to harm all of us, more attention is needed.