Until recently, discussions around artificial intelligence were mostly associated with the world of technology, algorithms, and academic research. Today, AI is becoming an integral part of everyday decision-making – from recruitment and financial analysis to content personalisation and e-commerce recommendations. Artificial intelligence ethics is now a critical challenge for any organisation aiming to build trust, act responsibly, and avoid missteps that could cost its reputation.
Based on a presentation by Izabela Nowak during a webinar co-organised with Grant Thornton Poland, we demonstrate that the ethical challenges related to AI are far from abstract – they are concrete and measurable. Below, we explore the six most important risks that every company should be aware of and consciously manage.
AI and the role of humans in the decision-making process
AI agents are systems that carry out tasks independently, based on data and algorithms. They are capable of analysing information, making decisions, and operating effectively without constant human oversight. Their purpose is to increase efficiency, speed, and potentially reduce costs.
What matters is where we place the human within the decision-making process:
- Human in the loop: AI analyses data and suggests options, but the final decision is made by a human.
- Human on the loop: AI makes decisions autonomously, but a human is able to intervene at any time.
- Human out of the loop: AI operates fully autonomously, with no human involvement in the decision-making process.
The less human involvement there is in the decision-making process, the greater the risk of unpredictability, unethical outcomes or breaches of the law. AI can have both a positive impact (such as improving workflows and supporting decision-making) and a negative one (such as enabling control, manipulation, or reinforcing biases).
Artificial intelligence ethics is essential for responsible development, especially in situations where legislation has yet to catch up with the pace of technological advancement. We need ethical reflection to ensure that AI serves the common good and does not undermine fundamental values.
Why do we need ethics?
Artificial intelligence ethics has taken on growing importance in today’s public discourse, as AI becomes an integral part of more and more aspects of life and is applied across a wide range of use cases.
AI can serve different purposes: increasing efficiency and profits, strengthening control over people, or supporting positive social and environmental change. Which direction we take depends entirely on us.
While the law defines the limits of what is allowed, ethics of artificial intelligence helps us understand which actions are appropriate and meaningful in striving to create a fairer and more sustainable society.
As AI plays an increasingly central role in our lives, it is crucial to identify potential ethical risks and challenges in order to shape its development and use responsibly.
1. The AI black box – when you don’t know how the decision was made
One of the biggest challenges associated with AI is the so-called black box problem – a situation where the system delivers an outcome, but there is no way to trace how it reached that decision. We can see the input (e.g. a candidate’s CV) and the output (99.9% match), but the path the algorithm took in between remains opaque.
This issue is particularly concerning in the case of deep neural networks, which operate through a multi-layered architecture. Each layer processes data in a different way, creating a complex web of interdependencies that is nearly impossible to break down step by step – even for experts.
Explainability as an ethical condition
There is a lot of talk in the AI community today about so-called explainability. This means the need to design systems in a way that allows us to understand why a given outcome has been generated – especially for decisions that affect people (recruitment, credit, diagnosis).
Imagine that an AI system recommends a candidate for a job with a 99.9% match.
- But why exactly this candidate?
- What specifically about this candidate made this match so high?
- Can we somehow trace how AI arrived at this?
Unfortunately, at present, many decisions made by AI are indeed impossible to understand, and not just because of the complexity of the system. First of all, most algorithms are not publicly disclosed by the companies that own them. Even when they are publicly available, they are too difficult for the average viewer, so we have to rely on experts, who are often employees of these very companies.
According to the 2018 guidelines, the ethos of artificial intelligence is that being able to explain the decisions of AI systems is crucial to building and maintaining user trust. This principle means that processes must be transparent, the capabilities and goals of AI systems openly communicated, and decisions as explainable as possible to those directly and indirectly affected.
Clarification tools
Tools such as SHAP (SHapley Additive exPlanations) or LIME (Local Interpretable Model-agnostic Explanations) are emerging that attempt to “replicate” the impact of specific variables on the final outcome. However, even these methods are only approximations and not full insights. There is still a lot of work ahead of us before we can say with certainty that we understand exactly how each decision was made.
A lack of explainability carries a real risk: we lose control over the logic of the decisions made on our behalf, which undermines user trust and exposes the company to legal risk.
Artificial intelligence ethics are crucial here. We must strive to build systems that not only work effectively, but also enable users to understand their decision-making processes. This is not only a matter of trust, but also of legal and social responsibility.
2. Algorithmic bias – AI can discriminate too
Contrary to appearances, AI is not free of biases. On the contrary, it learns them from us. The training data on which the system is based often contains coded biases resulting from human decisions. And instead of neutralising them, AI can perpetuate and reinforce them.
A real-life example: a company wants to improve recruitment and introduces an AI system to analyse CVs. It inputs data on “good” candidates from the past – that is, those who were hired and promoted. If the company was dominated by white males, AI will recognise this as a pattern. A candidate with a different career path? A person with a disability? The algorithm can exclude them.
This is not fiction. In 2018, Amazon withdrew its AI-based recruitment system because it was found to favour men and lower the ranking of women’s CVs. Why? Because the system trained on data from a male-dominated industry and… learned to discriminate.
Similar problems occurred in a word association system from Google, which displayed a racist bias and associated Mexicans with crime and women with housework. Such cases show that AI can not only reproduce but also reinforce existing inequalities.
Consequences of AI bias
AI biases can lead to discrimination, reduced trust in technology and serious legal consequences. This is why it is important to consciously select training data and apply bias reduction techniques. Artificial intelligence is not free of them therefore it can amplify them. The standard should be to select data for training AI systems in such a way that the data is as exclusion-preventing and representative of different social groups as possible.
3. The loss of privacy – what does AI really know about you?
For AI to work efficiently, it needs to have access to data. This can be very sensitive data: your PESEL number, bank passwords, location, health, call history, calendar. And while we give such consents knowingly, the question is: who else has access to this information?
Imagine an AI assistant that plans our day, reminds us of appointments and helps us analyse our health. It works brilliantly, but it also collects a large amount of data. Is this information secure? Can we really be sure that this data will not end up in the wrong hands – as a result of the manufacturer’s actions, a government decision or an attack by a hacker, which could even be another AI?
According to the Amnesty International Report – Technology giants offer their services to billions of users without charging them. Instead, users pay for the services with their personal data, which is constantly tracked online and offline.
The Cambridge Analytica and Facebook scandal in 2018 revealed how users’ personal data was used without their consent for political purposes.
Cambridge Analytica collected data from millions of Facebook users to create precise psychographic profiles that were used to influence election results. Facebook was accused of mismanaging privacy and failing to have adequate data protection measures in place.
The case has sparked a global debate on the ethics of data use, the responsibility of technology companies and information security.
Limits of user comfort and trust – implications
On the one hand, AI can be a huge convenience. On the other, users need to be confident that their data will not be used in an unauthorised way. Otherwise, the company risks a loss of trust, an image crisis or… a class action lawsuit.
Privacy breaches can lead to serious legal consequences, including financial penalties and reputational damage. This is why it is so important to comply with regulations such as RODO or the AI Act, and to build transparent and secure AI systems.
AI needs data to operate effectively, but this data must be protected. Users need to be confident that their privacy is respected and companies need to ensure security and transparency in data management. Artificial intelligence ethics are key to trust and responsible use of the technology.
4. Lack of accountability – who is responsible for AI error?
Liability for decisions made by artificial intelligence is also a serious question. If the system misjudges a customer, offers the wrong diagnosis or misleads the user – who is responsible?
The company that implemented the tool? Or the model provider? Or perhaps the creators of the training data?
Current regulations do not always clearly resolve this, increasing the need for ethical standards in AI implementation and clear oversight procedures. Central to this is the principle of Human in the Loop – humans should be the ones making the final decisions, not just observing the consequences.
Today, existing regulations (e.g. RODO, AI Act) do not cover all situations in which artificial intelligence makes decisions. In practice, responsibility is blurred between the creator of the algorithm, its owner and the company that implements it.
It is important for companies implementing AI to clearly define who oversees the decision-making process, how the redress path works, and how the user can claim their rights.
The lack of clear accountability rules can lead to abuse, loss of user trust and serious legal consequences.
This is why it is so important to put in place clear regulations and ethics practices for artificial intelligence in management.
5. Misuse of content creation – when AI infringes copyright
The use of AI in the creative industries – graphics, music, film or literature – is particularly controversial. Artists often raise the argument that their style, ideas and originality are being “copied” by AI systems that learn from their work without their knowledge or consent.
On the other hand, AI proponents point out that humans are also inspired by other artists, and that art has always been a process of combination, transformation and reinterpretation.
However, the difference in scale and automation in the case of AI makes this argument take on new importance – especially when the commercial use of generated content starts to compete with human labour.
Imagine a company wants to create graphics to promote a new product. Instead of hiring an artist, it uses AI to generate an image in the style of the famous Studio Ghibli. The effect is impressive, but is it legitimate and honest?
Studio Ghibli is a world-famous brand and its distinctive style is the result of years of work and experience by its artists. Today, thanks to AI, anyone can generate an image in this style with a single prompt. This raises questions about copyright and ethics.
In 2023, there was a lawsuit in the US in which an artist sued a company for using his style in AI-generated graphics. This case shows that the ethics of artificial intelligence cannot overlook such key issues as copyright.
Consequences for companies
Companies that use AI to create content today need to think twice before publishing such content. Copyright infringement can lead to serious legal consequences, including financial penalties and reputational damage.
Check out our approach to automation and AI.
Artificial intelligence can be a powerful creative tool, but it must be used responsibly. Companies must take care to respect copyright laws, and the ethics of AI, should be integral to the use of this technology to avoid legal consequences and loss of user trust. Otherwise, inspiration can turn into theft.
6. Trustworthy AI – can we trust it?
Trust is a relationship between equals in which one party, without knowing for sure, believes that promises will be kept.
Humans should not have to trust AI, because artificial intelligence should be constructed so that we know how it will act in a particular situation. The trust relationship is primarily between the people who use AI and the people who created the AI system in question.
In recent years, researchers, technology companies and governments have begun to focus on setting standards and guidelines for AI. The most important regulations are RODO and the EU AI Act.
Pillars of the AI Act and the artificial intelligence ethics
The AI Act, the legal regulation of the use of artificial intelligence in the European Union, is based on four key pillars:
- Principle of respecting human autonomy – AI should support and not replace humans in decision-making processes. Humans should be able to supervise and control the operation of AI.
- Principle of explainability – AI systems must be designed in such a way that their operation, goals and risks can be understood. This requires transparency and the ability to explain decisions made by AI. The level of explainability should be tailored to the specific application of AI and the potential consequences of any errors.
- Principle of fairness – This is about ensuring both substantive fairness – i.e. avoiding bias, discrimination and stigmatisation – and procedural fairness, i.e. the ability to challenge decisions.
- Principle of harm prevention – AI systems should operate safely and reliably, especially in the context of vulnerable people and where they may exacerbate inequalities arising from information or power advantage. Protection from harm also includes care for the environment.
However, the artificial intelligence ethics should not be limited to formulating general principles – it is equally important to continually respond critically to specific contexts of AI use and to identify necessary changes in regulation and its practical application.
The example of RODO shows what an overly legalistic approach leads to: the focus was mainly on the legal clauses, neglecting the technical and organisational aspects and missing the opportunity to involve managers.
As a result, the regulation became perceived as a bureaucratic obligation, which weakened its positive impact. The same scenario could befall the EU AI regulation (AI Act) if the practical and ethical side of it is not taken into account.
