Generative Artificial Intelligence

What is GenAI?

GenAI models generate output based on a question or query (‘prompt’) from a user. This requires using huge amounts of data to train the algorithm to recognise patterns and features in the data. This data set is called a ‘training set’.

This group of models includes language models, such as ChatGPT (OpenAI), Bing (Microsoft), and Bard (Google), image models, such as DALL-E (OpenAI) and Midjourney (Midjourney, Inc.), and models that generate videos or music. There are also open source models, which are not owned by commercial organisations and whose underlying algorithms can be viewed by users.

Not your average chatbot

Language models like ChatGPT are capable of answering your question because they are trained in such a way that the output consists of the information that is statistically most likely to follow the data in the prompt. In addition, the model remembers your prompts and can respond to previously asked questions and given output. This makes it possible to ask a language model for clarification or to go deeper into a topic, for example. The applications of this technology are expanding all the time.

Development

In November 2022, OpenAI made the prototype of ChatGPT 3.5 publicly available. Since then, GenAI has received a lot of attention in the public debate, and new applications and uses are rapidly becoming available. The technology is now used by millions of people worldwide.

Impact on the university

The applications of GenAI are broad and touch on many Radboud University activities, in particular within teaching and research. Language models can help answer complex questions, brainstorm, create automatic summaries, write text and programming code, and correct language errors.

Assessment

Our focus lies primarily on the impact of GenAI on the exams and assignments completed by students out of sight of a lecturer or invigilator. We have for a long time been meaning to bring assessment more closely in line with student learning, but with the advent of GenAI, this has become a necessity.

Opportunities

Clearly, GenAI also offers opportunities: these applications can be used to enrich our teaching and reduce workload. Students can use this technology to assist their learning process, for example by requesting further explanation or using a language model as a brainstorming partner. GenAI can also benefit lecturers: a language model can help them formulate learning objectives, prepare a lesson plan, or give feedback. The time thus freed up can be spent on complex teaching content or more personal contact between students and lecturers, for example.

Recommendations and direction

In recent months, lecturers, study programmes, and Examining Boards have already gained a lot of experience with this new technology, and had the opportunity to reflect on how they wish to use it in their teaching and assessment. This page contains some focus areas and tips for lecturers. A vision document specifically devoted to GenAI in education will soon be available, describing how Radboud University relates to this technology and its application in education.

Fraud and policy

While GenAI offers opportunities for education, it also poses risks. In this context, the University must especially devote attention to plagiarism and fraud: The use of output (e.g. text or programming code) generated by GenAI models in an exam, assignment, or thesis falls under the definition of fraud. In addition, GenAI falls under the category of ‘unauthorised tools’, as referred to in the Integrity Code for Digital Assessment.

Personal responsibility

There are all kinds of developments underway that aim to combat plagiarism with GenAI. This may also require additional policy, for example regarding assessment methods. The reality is that as an educational institution, we must learn to deal with this new technology, with the fundamental question being “Can you use GenAI to enhance your own learning while maintaining responsibility for learning the relevant material or developing the relevant knowledge and skills?” In this context, we also expect students to take their responsibility.

Use of GenAI

If you decide to use GenAI as an employee or student, you should be aware that

• the output of GenAI models is not always correct. In addition, most models act as a ‘black box’ (e.g. training sets are often not made public) and it is often difficult as a user to trace where the output comes from.
• Training sets almost certainly contain copyrighted material, the use of which has not been authorised by the author. In addition, many GenAI models contain biases that affect the output.
• Developing and training GenAI models consumes massive amounts of energy, which has a significant impact on the environment and climate. In addition, developing this technology requires substantial financial investments that can almost only be generated by large commercial tech companies.
• In most free and paid GenAI models, all the input you provide becomes property of the developer. We therefore recommend that you do not share any sensitive information.