Agenda Digitale: AI Ethics Starts with Data – The Role of Training

Source:

• Agenda Digitale, published on May 16th, 2025

By Riccardo Ocleppo, Founder and Director of OPIT – Open Institute of Technology

AI ethics requires ongoing commitment. Organizations must integrate guidelines and a corporate culture geared towards responsibility and inclusiveness, preventing negative consequences for individuals and society.

In the world of artificial intelligence, concerns about algorithmic bias are coming to the forefront, calling for a collective effort to promote ethical practices in the development and use of AI.

This implies the need to understand the multiple causes and potential consequences of the biases themselves, identify concrete solutions and recognize the key role of academic institutions in this process.

Bias in AI is a form of injustice, often systemic, that can be embedded in algorithms. Its origins are many, but the main culprit is almost always the data set used to train the models. If this data reflects inequalities or prejudices present in society, the risk is that AI will absorb and reproduce them, consolidating these distortions.

But bias can also manifest itself in the opposite direction. This is what happened some time ago with Google Gemini. The generative AI system developed by Google, in an attempt to ensure greater inclusivity, ended up generating content and images completely disconnected from the reality it was supposed to represent.

Further complicating the picture is the very nature of AI models, which are often characterized by complex algorithms and opaque decision-making processes. This complexity makes it difficult to identify, and therefore correct, biases inherent in the systems.

Ethical Data Management to Reduce Bias in AI

Adopting good data management practices is essential to address these issues. The first step is to ensure that the datasets used for training are diverse and representative. This means actively seeking data that includes a wide variety of demographic, cultural, and social contexts, so as to avoid AI exclusively reproducing existing and potentially biased models.

Alongside data diversification, it is equally important to test models on different demographic groups. Only in this way can latent biases that would otherwise remain invisible be highlighted. Furthermore, promoting transparency in algorithms and decision-making processes is crucial. Transparency allows for critical control and makes all actors involved in the design and use of AI accountable.

Strategies for ethical and responsible artificial intelligence

Building ethical AI is not an isolated action, but an ongoing journey that requires constant attention and updating. This commitment is divided into several fundamental steps. First, ethical guidelines must be defined. Organizations must clearly establish the ethical standards to follow in the development and use of AI, inspired by fundamental values ​​such as fairness, responsibility and transparency. These principles serve as a compass to guide all projects.

It is also essential to include a plurality of perspectives in the development of AI. Multidisciplinary teams, composed of technologists, ethicists, sociologists and representatives of the potentially involved communities, can help prevent and correct biases thanks to the variety of approaches. Last but not least, promote an ethical culture : in addition to establishing rules and composing diverse teams, it is essential to cultivate a corporate culture that places ethics at the center of every project. Only by integrating these values ​​​​in the DNA of the organization can we ensure that ethics is a founding element of the development of AI.

The consequences of biased artificial intelligence

Ignoring the problem of bias can have serious and unpredictable consequences, with profound impacts on different areas of our lives. From the reinforcement of social inequalities to the loss of trust in AI-based systems, the risk is to fuel skepticism and resistance towards technological innovation. AI, if distorted, can negatively influence crucial decisions in sectors such as healthcare, employment and justice. Think, for example, of loan selection algorithms that unfairly penalize certain categories, or facial recognition software that incorrectly identifies people, with possible legal consequences. These are just some of the situations in which an unethical use of AI can worsen existing inequalities.

University training and research to counter bias in AI

Universities and higher education institutions have a crucial responsibility to address bias and promote ethical practices in AI development. Ethics must certainly be integrated into educational curricula. By including ethics modules in AI and computer science courses, universities can provide new generations of developers with the tools to recognize and address bias, contributing to more equitable and inclusive design. Universities can also be protagonists through research.

Academic institutions, with their autonomy and expertise, can explore the complexities of bias in depth, developing innovative solutions for detecting and mitigating bias. Since the topic of bias is multidimensional in nature, a collaborative approach is needed, thus fostering interdisciplinary collaboration. Universities can create spaces where computer scientists, ethicists, lawyers, and social scientists work together, offering more comprehensive and innovative solutions.

But that’s not all. As places of critical thinking and debate, universities can foster dialogue between developers, policy makers, and citizens through events, workshops, and conferences. This engagement is essential to raise awareness and promote responsible use of AI.

In this direction, several universities have already activated degree courses in artificial intelligence that combine advanced technical skills (in areas such as machine learning, computer vision and natural language processing) with training that is attentive to ethical and human implications.

Academic Opportunities for an Equitable AI Future

More and more universities around the world – including Yale and Oxford – are also creating research departments dedicated to AI and ethics.

The path to ethical AI is complex, but it also represents an opportunity to build a future where technology truly serves the common good.

By recognizing the root causes of bias , adopting responsible data practices, and engaging in ongoing and vigilant development, we can reduce the unintended effects of biased algorithms. In this process, academic institutions – thanks to their expertise and authority – are at the forefront, helping to shape a more equitable and inclusive digital age.

• Agenda Digitale