Responsible AI: from principles to practice

Artificial Intelligence is developing rapidly. New models appear almost every week, and more and more organisations are experimenting with AI. At the same time, an important question is emerging: how do we ensure that AI not only adds practical value to an organisation, but is also developed and used safely and responsibly?  

In practice, we see many organisations facing the same questions:  

• Can you use all data for AI? 

• How do you prevent a model from discriminating? 

• How do you ensure AI complies with new legislation such as the EU AI Act and GDPR/AVG? 

• Who is responsible if the AI makes a mistake? 

• What data are used to train this system? 

Many AI projects start with enthusiasm. Questions about risks, compliance and ethics only arise later. That is often where things go wrong. 

Without clear guidelines, the risk of legal problems increases, results can become unreliable, and customer trust can come under pressure. Responsible AI is therefore not a theoretical concept, but a practical requirement for organisations that want to use AI sustainably. 

At Data Science Lab, we turn Responsible AI into a concrete framework built on three principles: ethical responsibility, legal compliance, and technical safety and reliability. Together, these principles form the basis for AI solutions that make an impact while remaining responsible.  

Our approach to Responsible AI 

Responsible AI is an umbrella term covering a range of topics. These include respecting human rights and developing AI systems that are safe, reliable and explainable, as well as fair, inclusive and developed with sustainability in mind.  

1. Ethically responsible AI

AI systems affect people. They support decisions, automate processes and help shape policy choices. That is why you need to assess the societal impact of an AI solution critically.  

For example, we look at where the data comes from, investigate whether a model disadvantages certain groups and make sure stakeholders can understand how the results were reached. 

From principle to practice 

To address these questions systematically, we have a Responsible AI team that acts as an internal sounding board.  

We discuss potential ethical risks from the first project phase. We also encourage knowledge sharing within the organisation through: 

• Responsible AI workshops 

•  Internal research projects 

• Developing sustainable AI solutions 

One example is our own LessGPT, a sustainable alternative to ChatGPT that uses less energy, water and CO₂.  

2. Legally compliant AI 

AI systems must not only be ethically responsible; they must also comply with laws and regulations.  

In Europe, the EU AI Act applies. This legislation sets requirements for organisations that develop or use AI systems. The obligations vary depending on the risk posed by an AI application. 

Examples of obligations that may apply to organisations (depending on their role and the system’s risks) include: 

• How the AI system is designed and documented (technical documentation) 

• Ensuring the data used is of sufficient quality and bias is minimised as far as possible 

• Systematically identifying and mitigating risks (risk management) 

• Enabling human oversight (human oversight) 

• Ensuring the system operates safely, robustly and accurately 

• Clearly informing users about the use of AI (transparency)  

From principle to practice 

To assess AI projects against these requirements, we developed an internal EU AI Act compliance check. We use it to assess new projects on: 

• The risk category of the AI system 

• Required documentation 

• Transparency requirements  

• Developers’ responsibilities. 

This way, we take regulations into account from the outset.  

3. Technically secure and reliable AI 

Alongside ethics and legislation, technology also plays a crucial role. AI systems must work reliably and handle data securely. That requires control over infrastructure and models.  

From principle to practice 

Not every organisation wants to share sensitive data with public AI platforms. That is why, where necessary, we run AI models on our own GPU infrastructure. This allows organisations to retain control over their data and meet security and compliance requirements more easily. 

Responsible AI is an ongoing process 

Responsible AI is not a checklist you complete once.  

AI technology is constantly evolving. New applications bring new risks and questions. That is why we see Responsible AI as an ongoing process of learning, evaluating and improving. By combining clear principles with practical ways of working, we ensure that AI solutions not only create value, but are also developed safely, transparently and responsibly.   

Organisations that consider Responsible AI from the outset do more than avoid risks. They also build trust more quickly with employees, customers and regulators. That is what makes Responsible AI more than a box-ticking exercise: it is an important part of successful AI implementations. 

Get in touch with us 
Want to know how your organisation can use AI safely, responsibly and in compliance with regulations? We would be happy to discuss it with you. In a no-obligation conversation, we will identify the opportunities, risks and next steps for your organisation.