Ethical issues in the use of artificial intelligence: between public policy, innovation and necessary risks

Artificial intelligence has quickly moved from a futuristic concept to a tool used daily in companies, universities and public institutions. In less than a decade, genetic algorithms, decision automation, data analytics and LLMs have changed the way we work, communicate and make decisions. This technological wave has reignited old discussions about responsibility, social justice, and the impact of progress on people.

In fact, this is not the first time that technological progress has raised ethical questions. From the Industrial Revolution to the emergence of the Internet, new technologies have brought innovation, but also labor exploitation, pollution, surveillance with the violation of the right to privacy, information manipulation or economic inequality. The difference is that artificial intelligence moves these dilemmas into an area difficult to anticipate: we no longer automate only physical work, but also intellectual processes, which leads us to a complex question: what ethical issues are relevant in the age of AI? Currently, the central challenges focus on several major themes:

Responsibility and control

A recent incident in Australia highlighted the risks of uncontrolled use of artificial intelligence in critical processes. It all started with an official report, made by a consulting firm for the Australian authorities, which contained errors generated by the use of an LLM (Large Language Model). Although these were later corrected, the situation showed how easily wrong information can be integrated into official decisions when human oversight and transparency are lacking. The incident highlighted vulnerabilities related to quality control, internal auditing and accountability in the adoption of emerging technologies.

If a report of such magnitude can be affected by the process of hallucinating of an LLM (that is, the generation of false, inaccurate or invented information but which seems plausible) suspicions may also arise related to documents with a limited relevance: audit, press, internal reports, etc. Thus, the issues raised by this case remain relevant in the future:

• Who is legally responsible for the error: the consultant, the algorithm, the company?
• How do we verify the source of information in an automatically generated document?
• Is mandatory tagging required for AI-generated content?

Labor market impact and mixed productivity results

In many companies, AI tools are introduced as work efficiency solutions, but they also carry a not inconsiderable risk. Some organizations limit the use of LLMs for privacy or security reasons, while others integrate access to them in all business processes. The difference between these practices creates unfair competitive advantages: where AI is enabled, work time decreases and activities become more efficient.

In a report carried out by McKinsey&Company in 2025, the idea is outlined that the AI ​​revolution is of the same caliber as the invention of the steam engine and that 90% of working people between the ages of 35 and 44 declare themselves ready to integrate AI tools in their work. Although increasing work efficiency in the long term is indisputable, in the short term the benefits brought by artificial intelligence are still unclear.

The same register includes the study carried out by IBM this year, which shows that there is a trend of acceptance (between 24.5% and 34.6% of respondents) of using LLM-type tools, which have proven especially valuable in solving repetitive tasks. A critical approach to the integration of AI in software development can be found in the study published by Joel Becker et al. in 2025 which reveals a decrease of almost a fifth in productivity among experienced software developers due to AI's incomplete understanding of the work context.

While for employers the dilemma is no ifbut cȃwhen and ȋat what pace to integrate AI technologies into the current activity, for employees there is a natural concern: what happens to the professions that can be automated?

Bias and discrimination

Algorithms learn from real data, which contain the stereotypes, prejudices and inequalities existing in society. Without human intervention, AI can reproduce or amplify inequity, from rejecting job candidates, providing incorrect financial or legal recommendations, to misidentifying people on racial grounds.

In this sense, the position of a technological colossus is well known, which declared in 2020 the cessation of the development of facial recognition software used for surveillance, in response to the wave of protests in the United States triggered by the death of George Floyd. Although not directly related to a racially motivated misidentification, the case has sparked a heated debate about systemic racism and racially biased police technologies.

Examples of misidentification for racial reasons, with potentially very serious consequences, can be found all over the world. In Xinjiang (China), facial recognition is used for identification and tracking Uyghura group subject to systemic repression. In Rio de Janeiro (Brazil), the police used facial recognition in armed operations in the favelas, and cases were documented where innocent people were targeted or put in danger because they were mistaken for dangerous suspects. In the USA, the cases of Niijer Parks (New Jersey) and Robert Williams (Detroit) have remained emblematic of the situations in which African-American people were knocked to the ground and handcuffed brutally for reasons of wrong identification through facial recognition.

Information manipulation

The spread of disinformation, the generation of fake content and technology deepfake make it difficult to separate the real from the manipulation. The shocking claim of the scientist Neil deGrasse Tyson, who appears in a deepfake video saying that Groundingȃthe end is flat was debunked in his show, StarTalk. However, many other such video materials remain accessible to the public and undisputed, in which public figures are associated without their consent with recommending a product, for example, or with adopting a controversial position on a sensitive matter.

A McAfee report from February 2023 to February 2024 shows that 43% of respondents reported deepfake content, 26% were exposed to deepfake scam attempts and 9% actually fell victim to them. Scam operations based on identity theft are becoming more frequent and difficult to combat, and there is concern that the phenomenon will gain an increasing scope.

In politics, especially in electoral campaigns, the vulnerabilities created by deepfake or fake news tools at the voter level can be decisive. However, there are exceptions: In the case of Ireland, the deepfake video material in which Catherine Conolly announced her withdrawal from this year's presidential race did not prevent her from winning.

In order to protect against the malicious use of deepfake material, Denmark has proposed an innovative, unique in Europe, amendment to intellectual property law that stipulates a right of ownership over a person's voice, face and physical appearance. If passed, it will allow civil compensation to be awarded to people whose identity has been used without consent in deepfake material.

Despite these major risks, the decisive role that deepfake technology can have in education and personal development is undeniable. From the virtual recreation of some historical personalities, to assist the teacher in capturing the students' attention, to the correction of homework and the provision of personalized feedback to the simulation of some patients to help the students of Medicine or Psychology, the technology remains at the service of the intention behind its use.

Dependence on technology

AI tools promise efficiency, but there is a risk that some of the human skills like critical analysis, writing, argumentation and others will erode over time by delegating them. Some recent studies (MIT 2025, MDPI 2025) warn against cognitive duty (the accumulation of unfinished or unprocessed mental tasks that consume cognitive resources without the person in question realizing it) created when we use AI for tasks that previously required intellectual effort. There is no clear evidence of long-term neurological effects yet, but it is an area of ​​growing scientific interest.

Europe, between caution and losing the race

The context of the globalized society accentuates differences in approach related to AI technologies. The United States and China invest aggressively in technology and patents, while the European Union has so far preferred a protectionist strategy: firm regulations, anti-abuse legislation, banning real-time surveillance with biometric applications in public spaces, limitations on the collection, storage and use of personal data.

This caution holds back economic development: European AI companies cannot compete on the international market with American and Chinese ones, which increasingly leads to an exodus of them to the West. The domino effect appears in all fields of activity in Europe that could have benefited from innovation through AI. However, the same caution can prevent a negative impact caused by the uncontrolled adoption of emerging technologies.

Although apocalyptic scenarios often appear in public discourse, the most relevant concerns are not related to o super intelligence that takes control, but of already visible problems: errors that end up in official documents, the loss of traceability of automatic decisions, mass-generated content without verification, the vulnerability to digital manipulation and the decrease in trust in information.

In addition, more and more scientific articles and professional documents are written with the help of AI. The efficiency is remarkable, but a delicate question arises: how do we assess authorship, originality and fairness when content becomes automated?

The need for rules and dialogue

The answer can't just come from AI professionals. Ethicists, lawyers, teachers, doctors, engineers, legislators and the general public must be involved in defining the line between useful and dangerous. In Romania, the exercise of this type of dialogue is still at the beginning, but debates on this topic appear more and more often in the academic environment, in podcasts and in professional organizations.

The workshop organized in September 2025 at the Research Institute of the University of Bucharest brought exactly this perspective: AI is not just technology, but a social phenomenon. Emotions, education, language and culture are already being influenced by AI-powered applications.

Artificial intelligence is not a nemesis of man, but a powerful tool that can improve life and create opportunities. However, without clear rules, transparency and responsibility, the side effects can be serious: from information manipulation to the decrease in the quality of work or the loss of people's trust in technology and its products.

It is not yet predictable where this process is headed, but an honest and informed debate is essential. Technological progress continues, and the relevant question is not whether we can stop it, but how we can use it for the benefit of society.

Pages of [cod] – Article supported by ING Hubs