In publica commoda

DEALING WITH KI MODELS LIKE CHATGPT

Recommendation for Teachers

General assessment

ChatGPT is a particularly popular version of an AI-based chatbot. The service is available at https://chat.openai.com/chat . For the university context, it is particularly significant that the service can generate texts in question-answer mode in a high-quality manner. This also applies, for example, to the generation of source texts of programmes by AI-based chatbots. The recommendations presented in the following therefore explicitly apply to both classic texts and programme text. We use ChatGPT here as an example for the rapidly developing group of (text and image) AI models that will be the basis of many different programmes and services in the near future. For example, you can ask ChatGPT to introduce itself as follows:

What is ChatGPT, explained in about 100 words?

ChatGPT is a large language model developed by OpenAI, based on the GPT-3.5 architecture. It has been trained on a vast corpus of text data and can understand and generate human-like language responses to a wide range of queries and topics. ChatGPT can be used for various natural language processing tasks, such as text generation, summarization, translation, and conversation generation. It has the ability to provide insightful and informative answers to complex questions, making it a powerful tool for education, research, and communication.

ChatGPT's capabilities are so powerful that, as of the end of March 2023, it can already "answer" many questions from typical bachelor's exams at least at the performance level of average students.

The texts generated by ChatGPT are so similar to those written by humans that their use can currently hardly be detected by automated identification, e.g. using plagiarism software. Even for readers, differentiation is already difficult and will probably be almost impossible in the foreseeable future.

ChatGPT will therefore change learning and teaching. ChatGPT also has an impact on which competences are needed in later professional life.
However, ChatGPT does not guarantee that the texts generated by the system are correct. On the contrary, the system also learns from false information entered and disseminates it further. In this way, even completely false and freely invented information is presented convincingly:

Why is the building "Blaue Turm" in Göttingen actually called that?

The building "Blauer Turm" in Göttingen, Germany, is named after its distinctive blue tower. It was originally built in the 14th century as part of the city's fortifications and was known as the "Hexenturm," or "Witches' Tower." The tower was later incorporated into a larger building, which was used as a prison until the 19th century. In the early 20th century, the building was renovated and turned into a museum. Today, it is home to the Municipal Museum of Göttingen, which features exhibits on the city's history, art, and culture. The blue tower remains a prominent landmark in the city and is a popular tourist attraction.

In the academic context, it is additionally relevant that ChatGPT not only "invents" facts but also sources and presents them in the same form as existing sources. Thus, references listed by ChatGPT with all source information such as authors' initials, journal, year, page numbers, etc. may be fictitious. It is therefore the particular responsibility of the user how the texts generated by ChatGPT are used..

The power of GPT 4, Bard, Claude and similar language models is rapidly increasing and the extensive feedback on these AI models will further support this process. At the moment it is not clear how far the capabilities of ChatGPT and similar systems will extend. A strategy for dealing with AI-supported systems must therefore not build on existing gaps in the current system, but must take a fundamental position on this. The development observed in the course of dealing with Wikipedia has also shown that the use of such systems can neither be prevented nor prohibited. On the contrary, the University of Göttingen therefore understands ChatGPT as a tool that is used and for whose intelligent use students and teachers must acquire competences. The prerequisite for this is a transparent approach: all those involved must disclose the extent to which ChatGPT has been used.

Recommendations for users

Due to the rapid pace of further development, the following recommendations for users can only be of a provisional nature. ChatGPT is a synonym for comparable systems that will be available in the near future. Sooner or later, speech AI models will be of great influence in many areas of society, which is why it is indispensable for us as a University to repeatedly deal with opportunities and risks.


Exams

In examinations that are taken without supervision (e.g. assignments, essays), students have access to ChatGPT or similar AI text models. Banning ChatGPT is not only inadvisable because of the lack of control. It would also not be purposeful, since a sensible use of such tools will be important for later professional life and professional handling should therefore also be part of the examination. At the same time, it must be ensured that the examination performance is independent. In principle, a critical analysis of the forms of examination should be carried out for each course or module.
Against this background, the following recommendations are made:

  • In supervised examinations (e.g. written examinations and oral examinations), the use of tools such as ChatGPT is not permitted unless it is explicitly allowed. Use during toilet breaks or similar would be an attempt to cheat. In some cases, instead of texts written at home, it may make sense to use the e-examination room for the examinations, for example, where a text must be delivered with a certain time limit.
  • In examinations without supervision the following should apply:
    1. ChatGPT should in principle be an admissible aid in examinations without supervision, also because it is not possible to check whether non-admission would be adhered to..
    2. The admissibility of ChatGPT as an aid ends where parts of the submitted performance originate from ChatGPT and are not made transparent as such. This is equivalent to an attempt to cheat..
    3. The use of ChatGPT must therefore be completely transparent. Explicit labelling of passages created with ChatGPT is required. Students should also explain in which way ChatGPT was used for the creation of the examination performance. An example template can be found in the appendix.
    4. Eine solche vollständig transparente Nutzung von ChatGPT soll weder positive noch negative Auswirkungen auf die Bewertung der Prüfungsleistung haben.
    5. Examiners should adapt questions in such a way that the independence of an examination performance can also be understood. Possible forms of questions can be, for example: text comparisons, empirical studies,... (see, for example, . hier https://www.wiso-elearning.uni-hamburg.de/blog/erste-praktische-ideen-zum-umgang-mit-chatgpt-in-lehre-und-pruefung/ under "further tips and possibilities”).
    6. The way of independent performance changes when using ChatGPT. For example, text passages in question-answer dialogue can be developed, evaluated and linked to sources with ChatGPT on the basis of the competences acquired in the course. It does not per se speak against an independent performance if text modules developed in this way are taken over into the result delivered as an examination performance after linking with the previously identified sources.

At the beginning of the summer semester 2023, all study commissions / deans of studies, together with the teachers, should deal with the recommendations outlined here regarding ChatGPT and the effects on teaching and examinations. Particular attention should also be paid to the examination formats and the design of examinations.

In the course of the summer semester, regulations for the General Examination Regulations (APO) will be developed, which will clarify the following questions, among others:

  • the possibility of adapting the forms of examination, e.g. combining homework with oral examinations,
  • how to proceed in case of suspicion of cheating.
If necessary, this shall be supplemented by the study or examination commissions in the area of their responsibilities with subject-specific regulations.

Annex: Declaration on the use of ChatGPT and comparable tools in the context of examinations
In this paper, I have used ChatGPT or another AI as follows.:
[ ] not at all
[ ] during brainstorming
[ ] when creating the outline
[ ] to write individual passages, altogether to the extent of ...% of the entire text
[ ] for the development of software source texts
[ ] for optimising or restructuring software source texts
[ ] for proofreading or optimising
[ ] further, namely: …
I hereby declare that I have stated all uses completely. Missing or incorrect information will be considered as an attempt to cheat.  


Teaching

The intelligent use of AI systems offers opportunities, but also some risks for university teaching. The main goal when dealing with ChatGPT and other AI systems should be the acquisition of competences by the users. These include, for example:

  • the recognition of the connections between the formulation of a question and the answer of ChatGPT and, derived from this, the competence to optimise a question with regard to the desired level of detail of an answer,
  • the competence to recognise the limits, relevance and implications (values) of the answers that AI systems give to questions,
  • an understanding of how AI systems in general (ChatGPT in particular) work, which social biases are/can be contained in the training data and reproduced by AI, and how this can be addressed in research and education.

Building competence for the critical and responsible use of AI systems must become an integral part of teaching at our University. Competence orientation has long been an important aspect for university teaching, which requires special attention when dealing with AI systems.

The following recommendations for dealing with ChatGPT in teaching serve this goal:

  1. Consistent culture of transparency: the use of ChatGPT should be openly communicated by teachers and students.
  2. Integration of ChatGPT into teaching: ChatGPT can be integrated into the teaching and learning process in very different ways depending on the subject. Different approaches to this can be exchanged via good practice examples across departmental boundaries. Examples for students (from p. 18) and teachers (from p. 28) on this can be found at: Unlocking the Power of Generative AI Models and Systems such as GPT-4 and ChatGPT for Higher Education: A Guide for Students and Lecturers (uni-hohenheim.de)
  3. In subjects in which the production of own texts, pictures or music is an essential goal, the AI systems should be actively integrated as a future-oriented tool in the creation process and accompanied by dialogue. A move away from written work is not recommended.
  4. Integration of ChatGPT into teaching: ChatGPT can be integrated into the teaching and learning
  5. The existing AI systems have also noticeably increased learners' self-responsibility for their own success. This can be supported, for example, by teachers creating opportunities for regular voluntary learning assessment (without AI).

In order to support the implementation of these recommendations, it will be examined to what extent suitable material for further education and counselling can be developed and made available, possibly within the framework of Hochschule Digital Niedersachsen.


Curricular development

In addition to integrating AI systems such as ChatGPT into the individual courses, it makes sense to take up these developments in the curriculum. In this context, an increase in the following competences is considered important

  • Media and IT skills
  • Language and text skills
  • Assessment, evaluation and reflection skills
  • Methodological competence (scientific work) and practices of good scientific practice

Further procedure

These recommendations are the result of a workshop of the ThinkTank Digitisation in Teaching and the subsequent discussion in the zKLS. They are intended to serve as a basis for further discussions in the central bodies, but above all also at the level of the faculties.

The ThinkTank will develop concrete templates on open points (further technical development, law, plagiarism, incorporation into examination regulations, etc.). In addition, questions of barrier-free and equal opportunity access to the AI systems, which are usually commercial, must be addressed and discussed, as well as questions of data protection.

In the summer semester of 2023, teachers and students will gather experience with the use of ChatGPT as a tool for teaching and examinations without supervision.

As an update to the recommendations presented here, concrete recommendations for teachers on "ChatGPT in teaching" and "ChatGPT in examinations" as well as recommendations for students will be sought by the end of the summer semester at the latest. However, these too can only describe an interim status and will retain their provisional character.

Overall, the recommendations aim to find ways to integrate ChatGPT in academic contexts. At the same time, the aim is to ensure that AI systems are used responsibly and with a view to the potential inherent in these systems.


Linkliste