Information on generative AI
Contact
Medical Didactics
Thomas Schmidt, Dipl.-Biologist
Fakultät 6 Geschäftsstelle (» Postal address)
Access to the LOOOP platform(looop-share.charite.de) for sharing links to existing digital resources is now available to all faculties free of charge. Links are sorted thematically in LOOOP and can be assigned to subjects and cross-sectional areas as well as mapped against the NKLM. All you have to do is register.
Information on generative AI
Generative Artificial Intelligence (GKI)
GPT-3 from OpenAI has been available free of charge since November 2022, and so-called generative AI is causing a stir in the media. These chatbots are able to use artificial intelligence to provide answers to questions (also known as prompts ) from a huge pool of data, some of which can no longer be distinguished from human statements. The speed and accuracy of the statements are often astounding, but the answers are not always error-free and up-to-date, as ChatGPT-3, for example, can only access data up to 2021.
When using it, it should also be noted that generative AIs are not search engines, but generate text based on probabilities from the available training data. As all data is considered to be of equal value (e.g. an anonymous forum post is equivalent to a scientifically sound publication), social prejudices can be reproduced or completely false statements can be issued.
Special care should also be taken when entering personal data and copyrighted materials. Therefore, please inform yourself before using generative AI on the university pages listed below.
Recommendations of the University of Oldenburg on the use of generative artificial intelligence for studying and teaching
The Vice President for Academic Affairs, Teaching and International Affairs has developed recommendations for dealing with generative artificial intelligence (GKI) in studying and teaching in coordination with the Department for Study Affairs and the extended group of deans of studies.
You can find the current version of this orientation framework here.
Chat-AI service of the GWDG via the Academic Cloud
The AcademicCloud is a service provided by the Gesellschaft für wissenschaftliche Datenverarbeitung mbH Göttingen (GWDG), which you can use via a federated login with your university account. The University of Oldenburg has concluded corresponding contracts with the GWDG, whose applicable data protection information can be found here. In addition to other digital applications, the GWDG offers its own ChatGPT service, which runs on its own hardware. On the website, all users with an AcademicCloud account can chat with a generative AI and receive support with their work.
Special emphasis was placed on data protection during the design phase. In order to protect users' enquiries, entries and responses are only saved locally by the user within a session. When the website is closed, all entries and responses are deleted. Only the number of enquiries per user is recorded in order to determine the demand for this new service. If you have any questions or re-registering students, please contact
Creation of multiple choice questions with the help of AI
The creation of multiple choicequestions(MCQs) is time-consuming and the choice of clearly incorrect distractors is often particularly difficult. Text-generating AIs can help with the design of questions and provide an initial starting point for iterative further development. When using it, please always bear in mind that AI can output plausible-sounding but technically completely incorrect texts, and the questions must be carefully checked and adapted before being passed on to students.
It is a good idea to use a template for the prompt to create the MCQ, which then only needs to be filled in with the corresponding clinical picture and further specifications regarding the level of difficulty and format. The following two publications provide examples of such sample prompts:
1) Kiyak YS. A ChatGPT prompt for writing case-based multiple-choice questions. RevEspEduMed 2023, 3: 98-103; doi: 10.6018/edumed.587451
2) Zuckerman M, Flood R, Tan RJB, Kelp N, Ecker DJ, Menke J, Lockspeiser T. ChatGPT for assessment writing. Med Teach. 2023 Nov;45(11):1224-1227. doi: 10.1080/0142159X.2023.2249239. Epub 2023 Oct 16. PMID: 37789636.
As a first step, you can simply copy the template and enter it into ChatAI in the Academic Cloud . You can then use the draft generated by the AI to continue working on your question and adapt the template for your purposes by making repeated adjustments in interaction with the large language model.
Alternatively, you can use the case-based MCQ generator based on ChatGPT from OpenAI, which directly generates cases based on the sample prompts cited above. Please note, however, that unlike the service in the Academic Cloud, this is an AI that uses the data you enter for training purposes.
If you have any questions, please contact us directly.
Translated sample prompt for MCQ from Kiyak et al., 2023:
You are developing a database of questions for medical exams with the focus topic [PLEASE INSERT TOPIC]. Please generate a high quality single best answer-multiple choice question. Follow the principles for creating multiple-choice items in medical education.
Create the question using the following scheme:
Case description
(Formulate as a coherent narrative paragraph without listing the parts individually).
Patient data (gender/age)
Leading symptom / current complaint
Relevant medical history
Clinical examination findings
Results of diagnostic tests (optional)
Question
[Insert relevant information from the case described above without anticipating the answer]
Note: The question should ask for the best answer option - not "true" or "false".
Answer options
[Insert plausible answer option]
[Insert plausible answer option]
[Insert plausible answer option]
[Insert plausible answer option]
[Insert plausible answer option]
Explanation
Identify and explain the correct answer.
Explain why this is the most appropriate answer according to evidence-based guidelines or expert consensus.
Briefly explain why the other answer options are less correct or incorrect.
Level of difficulty
[PLEASE INSERT LEVEL OF DIFFICULTY (E.G. "EASY", "DIFFICULT")]
Image-generating AI
In addition to various text-generating large language models, the Academic Cloud also offers a service for creating images, ImageAI. You can find ImageAI in the Academic Cloud under "Services" after you have logged in via the federated login with your university login. Images that you have created with the help of AI are not subject to copyright, so you can use them without restriction for your events and teaching materials. Please note, however, that image-generating AIs in particular often reproduce prejudices that are contained in the training data. For example, even with gender-neutral English input, male persons in management positions may be depicted more frequently or photos with several persons may depict less diverse groups. You can avoid this by giving instructions that are as precise as possible.
In addition, prompting, i.e. your input to the AI, works slightly differently when creating images than when creating texts.
Here are some basic tips:
1. you will get better results if you create your input in English.
2. do not use negations (e.g. image of a doctor without a stethoscope). These are often not related to the following word by imaging AIs and the unwanted element is then displayed in any case.
3. Formulate the prompt as briefly and precisely as possible. Imaging AIs process a significantly smaller amount of text than large language models and elements that occur later in the text may therefore not be mapped.
4. Use so-called tokens that help the AI to assign terms in your query and not map them as individual elements. For example, use the following formula and replace the terms in the square brackets with your details:
[Scene and/or Subject],
[Details],
in the style of [Style or Artist].
[Color] colour,
[Lighting] light.
[Image Type]
Helpful documents for teachers
If you would like to sensitise students to the sensible and safe use of generative AI in your courses, you will find sample handouts and legal bases under the following links:
1. KIKO the UOL's GKI handout configurator: https://uol.de/strategie-lehre/ki/kiko_intranet
2. legal aspects of the use of AI in studies, teaching and examinations by Dr Janine Horn, ELAN e.V.: https://www.hs-osnabrueck.de/fileadmin/HSOS/Homepages/eLCC/KI-Monat/AI_Act_Horn.pdf
3. Handout "Rules For Tools" for courses by Prof. Christian Spannagel: www.uni-giessen.de/de/fbz/zentren/hd/projekte/hessenhub/ki/ki_tabs/rulesfortools_prof-spannagel.pdf
4. Helpful prompts for teaching in the open prompt catalogue of the Hochschulforum Digitalisierung and the KI-Campus: https: //coda.io/@kic/prompt-katalog
5. Documentation of the lunchtime meetings of university didactics on the topic of AI: https://uol.de/lehre/hochschuldidaktik/digitale-lehre/mittagstreffen
Further information
The projects Souver@n and participate@UOL provide current information and short videos on the topic of generative AI in studying and teaching at the UOL. The content is summarised here: https://uol.de/strategie-lehre/ki
The Hochschulforum Digitalisierung provides an extensive curated collection of links on the topic of ChatGPT in the university context, which is constantly updated. Here you will find the most important publications on the topic as well as field reports and blog posts on the topic of AI in university teaching.