By Gabriele Hooffacker
Abstract: There are already some indications of what generative language AI can and will be able to do. It will transform journalism, the »profession of the public sphere« (Pöttker 2010). What does journalism teaching look like under these conditions? Which competencies need to be taught? Which specific knowledge and skills? Instead of dealing with the topic in a theoretical way, this subjective debate piece attempts to approach it by exploring the topic together with students. It provides a workshop report, compiles possible learning objectives for both students and teaching staff, and inspires further thought about the competencies needed for the profession of the public sphere.
»I need to discuss this with my first-year students!« Having explored ChatGPT3 a little myself since the tool was released in late November 2022, I put my findings into practice straight away on January 2, 2023. As always, I start the lecture series »Content development I« with a question for the class. This time: »Who has heard of ChatGPT3?« Three quarters of the media technology students raise their hands. »And who has tried it out already?« The same three quarters raise their hands again. The session was to be educational for all of us – but more on that later.
Two points to note: 1. The students are at least at the same level as us teaching staff, if not ahead. 2. The level of awareness, familiar knowledge and thus teaching about generative language AI depend on the day – developments made the knowledge from January obsolete by April, and the content I write on the performance spectrum of ChatGPT3 today will be out of date by the time this piece is published.
Some things you might be expecting from this piece, but will not find:
- An overview of the services ChatGPT3 or 4 currently offer for journalistic texts
- A brilliantly written analysis of why life, media professions and journalists will never be the same again after the arrival of ChatGPT
- An absolutely logical argumentation of why, for mathematical reasons, language AI will never be able to replace journalism
The idea behind this piece is to show how language AI can currently be used in journalism teaching. There are already some indications of what generative language AI can and will be able to do. It will transform journalism, the »profession for the public sphere« (Pöttker 2010). What does journalism teaching look like under these conditions? Which competencies need to be taught? Which specific knowledge and skills? How can journalism be taught under these conditions?
This piece is designed as a workshop report. In terms of methods, I am guided in part by Bernhard Pörksen. In his book »Die Beobachtung des Beobachters« [Observing the observer], he calls on journalism teachers and students to see themselves as »participants in an expedition« (Pörksen 2015). I also pick up on the concept of »students as partners,« which I first became aware of in a training course on university didactics held by Anita Sekyra and Marie-Theres Lewe at HDS Sachsen (Sekyra/Lewe 2022). The concept is based on Donna Haraway’s »situated knowledge« approach (Haraway 1988).
I am therefore writing this workshop report in the radical first-person form and attempting to reflect on my own position in line with the »teaching through learning« model (Hooffacker 2009).
I have taken my working hypotheses from the future theses of university didactics and information technology expert Doris Weßels and applied them to journalism teaching:
»1. AI chatbot systems are becoming a personal learning companion and therefore an individualized learning bot for the learners. […] 4. We teaching staff are becoming architects and designers of the virtual (for example virtual spaces like the metaverse) and analog teaching space. The role of accompanying learning with a navigation function for the learners and designing the social space will become especially important here.« (Weßels 2022)
I also draw on the work of Doris Weßels for my definition. She speaks of »generative language AI« (Weßels 2022); most experiences are based on ChatGPT3 since this is currently freely available.
The competencies required for journalism teaching are derived from relevant publications on online journalism in journalism textbooks (Hooffacker/Lokk 2016; Hooffacker/Meier 2017; Hooffacker 2020).
I have structured my workshop report into learning objectives:
- Checking facts
- Knowing and applying journalistic rules of separation
- Transparency regarding invented persons and facts
- Prompting is everything!
- ChatGPT3 cannot write biographies
- Social media and press texts
- Reporting and reportage through prompting in dialog
Almost all of the examples come from lectures and seminars at Leipzig University of Applied Sciences (HTWK Leipzig), with one from an advanced training course for journalists at Stiftung Journalistenakademie in Munich. I would like to thank the students who shared these experiences with me.
Checking facts
A lecture and seminar on January 2, 2023 focus on the »news item« as a form of presentation. After introducing the structure and selection of news items, I present a »news item« that I have had written by ChatGPT3. The students are then asked to work out whether it is a journalistic news item. The news item is about the »Long night of computer games,« a popular event at our university.
The students are quick to find the first point of criticism: incorrect facts. This was easy to discover by researching online. Learning objective achieved: Because the data basis on which ChatGPT3 was trained is from late 2021, it is important to check whether more recent data is available.
Knowing and applying journalistic rules of separation
It takes them a little longer to uncover the second error in the news item, but a student does notice it after a while: ChatGPT3 ended the news item with the sentence: »The »Long night of computer games« at HTWK Leipzig offers a varied program for visitors.« »That does not belong in a news item,« says the student. »Information and opinion should be kept separate.« Learning objective achieved.
Transparency regarding invented persons and facts
In February, as part of an advanced journalistic seminar on the topic of »reportage,« I ask ChatGPT3 to write a reportage on a German carnival prince and princess, both the joys and the challenges of their role, and the importance of carnival in Bavaria. I do not explain the task in the seminar. The seminar participants note that the features of a reportage (people, quotes) have been met. They immediately check whether the location really exists – it does. »The text is oddly bloodless,« says one participant. »It could have been written without doing any research.« There is uproar when I tell them that ChatGPT3 is the author and that the prince and princess, Max and Lisa, are completely fictitious, including their quotes. Learning objective? Creating transparency in editorial offices about the use of generative language AI! See also the piece by Kim Björn Becker in this edition of Journalism Research (Becker 2023).
What I learned: My knowledge gave me a slightly unfair advantage over the participants. I therefore adjust my approach slightly in the key qualification seminar »cross-media press and public relations.« The rest of the examples in this piece are from this course, held in the summer semester of 2023.
This seminar is partly about journalistic writing, but also about text production for press releases or social media posts. One of the learning objectives I intend to achieve here in relation to the use of ChatGPT is the difference between »journalism« and »text generation for public relations« (Schrage 2023).
Prompting is everything!
In early April 2023, in the summer semester, the focus returns to writing news items. The students are given a content-based introduction to the structure and news factors, before being assigned a classic task common for trainee journalists: turning a press release into a news item. The students are free to choose whether to complete the task alone or with the help of ChatGPT.
When we discuss the results in the following seminar session, each student is asked not to say in advance who the author is. We guess together. And we are wrong together.
One outstanding journalistic news items comes from ChatGPT3. A less good one (sentences too long, news core not recognized) was written by students. Three students took the experiment further and used the same prompt, including the press release, in their three separate accounts. The result is three different, outstandingly written news items that use the correct form and structure. Only one of them has a small factual error.
I ask myself whether we still need to give trainees the classic task of writing a news item based on a press release. Is the discovery of where generative language AI can take on standard tasks for us not much more valuable than we currently see?
Amusing diversion: Further options for using language AI
I am working on this piece while sitting in a very full train. The young fellow passenger next to me is using Jenni.ai to write a cultural studies essay in English. We soon get talking. He is very happy with the tool, because it provides him with the right specialist terms and, above all, proper English. He did not know that the tools roll the dice differently each time; but he also uses it at work to write polite email responses, for example to his boss.
The lady across the aisle also gets involved: Does language AI also find literature? She has to write a lot of DFG research applications …
ChatGPT3 cannot write biographies
As part of the press and public relations seminar, the students are asked to write Instagram and Facebook posts. Actors need to be introduced. I stop short: This actor – was he really born in Munich? I don’t think so. Wikipedia confirms that he was born in Nuremberg. That is how we spot ChatGPT3: It cannot write biographies. It just makes them up based on probabilities. Fact checking – so important.
Social media and press texts
The students are making progress with prompting. They quickly learn to make precise entries in the text line – the prompt – and to enrich them with content. Soon, they are no longer satisfied with the mediocre responses, but demand improvements in dialog with the chatbot. ChatGPT3 can use pre-prepared information to generate beautiful dialog-like Instagram and Facebook posts, for example, complete with the common hash tags. What we all learn together: When the starting text entered as the prompt is good, the result is usually both correct and factually accurate.
ChatGPT does not perform so well when it comes to writing press releases. Again and again it mixes value judgements into the text, or the weighting and structure are wrong. One student notes, »I could have written the text faster myself.«
Reporting and reportage through prompting in the dialog
How to write lively reports is really a topic for advanced journalism students. The same goes for the use of ChatGPT. Again we used the method described above: Students were given the task of using pre-prepared material to write a lively report with quotes, a beginning, an ending and an arc of suspense, and were free to choose whether they wrote it themselves or using ChatGPT.
One less good report was written by the students themselves (who learned a lot in the subsequent discussion). One outstanding report was deemed by all of us to be written by a student, before the student described the approach she had taken using ChatGPT, divided into multiple steps: First, she entered the basic information and the task in the prompt. Then, she refined the process step by step, from »Choose a quote as the beginning,« to »give the beginning and ending the same theme,« »write an exciting title« etc. The description of the approach taught the students a lot. What I learned: The students could soon overtake us here.
Limitations and conclusions
The examples and considerations set out here by the author in May 2023 will probably already be obsolete when they reach their audience in summer 2023. The tools are being developed further and learning more all the time. The way they are used is moving forward. Something that is true today might look ten years old by tomorrow. New learning objectives will be added, while classic exercises might prove out of date. This workshop report can offer no more than a snapshot in time.
Journalistic writing and text production for public relations offer new opportunities for both tasks and methods in university teaching and journalism training. There is no need to define new professions at the moment – as was attempted at other moments of rapid change, such as upon the advent of online journalism (Altmeppen/Hömberg 2002). Instead, it is worth considering which competencies will be needed for the »profession of the public sphere.«
Trying out and assessing ChatGPT together has taught the students both rules for journalistic writing and knowledge of the possibilities and limits of generative language AI in a fast, compact and enjoyable way in comparably little time. Working hypotheses such as »language AI is well-suited to text generation in public relations, but only has limited uses for journalistic texts,« can be examined and scrutinized. A white paper by the University of Hohenheim, »Unlocking the Power of Generative AI Models and Systems like GPT-4 and ChatGPT for Higher Education – A Guide for Students and Lecturers,« which can be found at heise.de (Wiegand 2023), provides further options for use including advice for teaching staff on the use of generative language AI.
I showed the students my approach. What I learned: greater transparency about what we do together in the seminar, why we do it, and what we achieve and learn from it.
There appears to be no justification for the fear that students will use the new possibilities of language AI uncritically, as expressed in the white paper from the University of Hohenheim, namely »that students use the new possibilities only for passive absorption of information, instead of developing into critical minds« (quoted in Wiegand 2023). The investigations described invite the conclusion that students explore the possibilities for using generative language AI with curiosity and enjoyment, and gain from it. I can only concur with Doris Weßels (Hooffacker 2023): »Try it yourself.«
About the author
Gabriele Hooffacker, Prof. Dr. phil., (*1959), is co-editor of Journalism Research and teaches at HTWK in Leipzig in the field of »media-compatible content preparation.« Gabriele Hooffacker edits the textbook series »Journalistische Praxis,« founded by Walther von La Roche (1936-2010) and published by Springer VS, and the »Leipziger Beiträge zur Computerspielekultur« series. She is a jury member for the Alternative Media Prize.
Translation: Sophie Costella
References
Altmeppen, Klaus-Dieter; Hömberg, Walter (eds.) (2002): Journalistenausbildung für eine veränderte Medienwelt. Wiesbaden: Westdeutscher Verlag.
Becker, Kim Björn (2023): New game, new rules. An investigation into editorial guidelines for dealing with artificial intelligence in the newsroom. In: Journalistik/Journalism Research 6(2), pp. 133-153.
Haraway, Donna (1988): Situated knowledges: The science question in feminism and the privilege of partial perspective. In: Feminist Studies, 14(3), pp. 575 – 599.
Hooffacker, Gabriele (2009): Lernen durch Lehren. In: Hooffacker, Gabriele (ed.): Journalismus lehren. München: Dr. Gabriele Hooffacker.
Hooffacker, Gabriele (2020): Online-Journalismus. Ein Handbuch für Ausbildung und Praxis (5th ed.): Wiesbaden: Springer VS.
Hooffacker, Gabriele (2023): ChatGPT verändert Hochschullehre und Kommunikation. In: HTWK Leipzig, 25 January 2023. https://www.htwk-leipzig.de/studieren/newsdetail6/artikel/5596 (21 May 2023)
Hooffacker, Gabriele; Lokk, Peter (2016): Online-Journalisten – wer, wie, was, und wenn ja, wie viele? In: Hooffacker, Gabriele; Wolf, Cornelia (eds.): Technische Innovationen – Medieninnovationen? Wiesbaden: Springer VS, pp. 34 – 46.
Hooffacker, Gabriele; Meier, Klaus (2017): La Roches Einführung in den praktischen Journalismus (20th ed.). Wiesbaden: Springer VS.
Pörksen, Bernhard (2015): Die Beobachtung des Beobachters. Heidelberg: Carl Auer.
Pöttker, Horst (2010): Der Beruf zur Öffentlichkeit. Über Aufgabe, Grundsätze und Perspektiven des Journalismus in der Mediengesellschaft aus der Sicht praktischer Vernunft. In: Publizistik, 55(2), pp. 107 – 128.
Schrage, Klaus (2023): Was verändert die KI? In: dju Mittelfranken, 17 May 2023. https://dju-mittelfranken.verdi.de/themen/++co++ca5037c8-ef47-11ed-81ba-001a4a160110 (21 May 2023)
Sekyra, Anita; Lewe, Marie-Theres (2022): The potential of systematic reflection on one’s positioning: A feminist perspective on the Students-as-Partners approach. In: International Journal for Students as Partners, 6(3).
Weßels, Doris (2022): ChatGPT – ein Meilenstein der KI-Entwicklung. In: Forschung und Lehre, 20 December 2022. https://www.forschung-und-lehre.de/lehre/chatgpt-ein-meilenstein-der-ki-entwicklung-5271 (31 May 2023)
Wiegand, Dorothee (2023): ChatGPT im Hörsaal: Wie Hochschulen auf generative KI-Werkzeuge reagieren. In: heise online, 7 April 2023. https://www.heise.de/hintergrund/ChatGPT-im-Hoersaal-Wie-Hochschulen-auf-generative-KI-Werkzeuge-reagieren-8431537.html (31 May 2023)
About this article
Copyright
This article is distributed under Creative Commons Atrribution 4.0 International (CC BY 4.0). You are free to share and redistribute the material in any medium or format. The licensor cannot revoke these freedoms as long as you follow the license terms. You must however give appropriate credit, provide a link to the license, and indicate if changes were made. You may do so in any reasonable manner, but not in any way that suggests the licensor endorses you or your use. You may not apply legal terms or technological measures that legally restrict others from doing anything the license permits. More Information under https://creativecommons.org/licenses/by/4.0/deed.en.
Citation
Gabriele Hooffacker: How language AI could change journalism training. A workshop report by Gabriele Hooffacker. In: Journalism Research, Vol. 6 (2), 2023, pp. 189-196. DOI: 10.1453/2569-152X-22023-13410-en
ISSN
2569-152X
DOI
https://doi.org/10.1453/2569-152X-22023-13410-en
First published online
July 2023