Legal and ethical aspects
Together with considerations about the use of generative AI from pedagogical perspectives, it is crucial to address legal and ethical aspects.
Personal data, sensitive information, and confidentiality
When using services or tools that are not procured by the University of Gothenburg, it is necessary to ensure that they are not used in a way that risks violating any legislation, such as the General Data Protection Regulation or the Public Access to Information and Secrecy Act. There is a high probability that data is transferred to countries outside the EU, and it is then not permissible to input personal data into them. It is also important to avoid inputting sensitive or confidential information. Teachers must therefore be very careful with how they use these services and ensure that it is not done in a way that violates any legislation. A fundamental approach should be that only data that could be published openly should be input into the service. This currently also applies to Microsoft Bing Chat/Copilot.
Students’ use of generative AI: equal access to technologies and data sharing
If teachers plan to let students use generative AI in teaching activities or examinations, it is important to ensure equal access to the tools. In situations where the University of Gothenburg does not provide access to them, students cannot be expected to use them. There are also other ethical considerations to be made, especially whether it is permissible to require students to use services that involve sharing data or personal information. Based on this, it becomes clear that there are difficulties in developing setups that sustainably include students' use of generative AI.
Awareness of bias and limitations
The data used to train generative AI in turn affects the outcome when using services to generate material. Specifically looking at the language model GPT, previous models were primarily trained on English texts originating from Western cultures (the data used to train GPT4 has not been disclosed). Besides being trained to interpret and generate language, it has also become clear over time that models have been influenced by the underlying culture and thus reproduces it when generating text. When using generative AI, it is necessary to be aware that what is generated may be influenced by the training data used for the model. Questions about bias in training data, reproduction of culture and prejudice, etc., also need to be raised with students so they can develop a responsible and conscious approach to generative AI. Unfortunately, transparency around training data, algorithms, etc., is practically non-existent, so working with these issues can be challenging.
Economic aspects
In cases where the university has not procured a service or tool, it may be relevant to consider economic aspects before starting to use them. When using free services for generative AI, one might need to "pay" with the data input into the system and/or the results generated. These may then, for example, be used to develop existing or new services or for marketing purposes. Therefore, it may be relevant to consider whether this is acceptable.