Why we developed our own AI chat 

AI tools like ChatGPT have become a part of everyday life, including at the UvA. We can see that generative AI tools are increasingly put to use. This raises questions about privacy, academic integrity and reliability. In response, the UvA decided to develop its own AI Chat, one that is built in-house, based on public technology and rooted in the UvA’s academic values. 

With UvA AI Chat, the university aims to provide a safe, ethical and transparent alternative to commercial AI tools, one that contributes to improving AI literacy. Students and staff are free to experiment with it within clear boundaries. In this way, we can explore if and how this technology can enrich education and support the learning process. 

UvA AI Chat is safer than commercial alternatives 

UvA AI Chat is designed to ensure that you can use it safely and responsibly: 

• Data stays within the UvA: all data is stored within the UvA’s technical infrastructure. We maintain control over what happens to the input. With commercial AI tools, this is out of our control. 
• Privacy first: all user queries are anonymised. What you do within UvA AI Chat is visible only to you, the user. 
• No commercial use or selling of data: The UvA has no commercial interests in UvA AI Chat. Your input is not used to train new models or for advertising purposes. 
• Transparency about the models used: the UvA is open about which underlying models have been used and how UvA AI Chat uses them to generate answers. 
• Access for everyone: all students and staff have free and equal access to the tool. There are no hidden costs or premium subscription tiers. 

Autonomy and academic freedom 

With UvA AI Chat, the UvA is also taking important steps to preserve and strengthen its digital sovereignty and academic freedom. By choosing an in-house AI solution that is based on open technology and under our control, the UvA is reducing its dependency on external parties and commercial interests. This enables us to decide how the technology is used, which values are central and how privacy and data are handled. In this way, research and education at the UvA can take place without unwanted external influence. 

Energy consumption of UvA AI Chat 

Large AI models are known for their high energy consumption. The UvA actively takes this into account:  

• The underlying models used in UvA AI Chat were selected for their reliability and efficiency. 
• By choosing our own infrastructure, the UvA can actively apply sustainability criteria. For example, we can decide to implement energy-efficient developments in the model. With commercial alternatives, we have no control over sustainability criteria. 
• The chat environment has been designed so that users can easily select an energy-efficient model themselves. 
• We are working on ways to make energy usage transparent to users so they can decide how to use the tool more sustainably.