1、Executive Summary 1 pt Unlocking the Power of Generative AI Models and Systems such as GPT-4 and ChatGPT for Higher Education A Guide for Students and Lecturers The authors institutions Grateful acknowledgement of funding by 2 Unlocking the Power of Generative AI Models and Systems such as GPT-4 and

2、 ChatGPT for Higher Education A Guide for Students and Lecturers Henner Gimpel1,6 Kristina Hall2,6 Stefan Decker5,6 Torsten Eymann2,6 Luis Lmmermann3,6 Alexander Mdche4 Maximilian Rglinger2,6 Caroline Ruiner1 Manfred Schoch1,6 Mareike Schoop1 Nils Urbach3,6 Steffen Vandirk2 1 University of Hohenheim

3、 2 University of Bayreuth 3 Frankfurt University of Applied Sciences 4 Karlsruhe Institute of Technology 5 RWTH Aachen University 6 Fraunhofer Institute for Applied Information Technology FIT March 20,2023 Executive Summary 3 Executive Summary Generative AI technologies,such as large language models

4、,have the potential to revolutionize much of our higher education teaching and learning.ChatGPT is an impressive,easy-to-use,publicly ac-cessible system demonstrating the power of large language models such as GPT-4.Other compa-rable generative models are available for text processing,images,audio,v

5、ideo,and other outputs and we expect a massive further performance increase,integration in larger software systems,and diffusion in the coming years.This technological development triggers substantial uncertainty and change in university-level teach-ing and learning.Students ask questions like:How c

6、an ChatGPT or other artificial intelligence tools support me?Am I allowed to use ChatGPT for a seminar or final paper,or is that cheating?How exactly do I use ChatGPT best?Are there other ways to access models such as GPT-4?Given that such tools are here to stay,what skills should I acquire,and what

7、 is obsolete?Lecturers ask similar questions from a different perspective:What skills should I teach?How can I test students competencies rather than their ability to prompt generative AI models?How can I use ChatGPT and other systems based on generative AI to increase my efficiency or even improve

8、my students learning experience and outcomes?Even if the current discussion revolves around ChatGPT and GPT-4,these are only the forerunners of what we can expect from future generative AI-based models and tools.So even if you think ChatGPT is not yet technically mature,it is worth looking into its

9、impact on higher education.This is where this whitepaper comes in.It looks at ChatGPT as a contemporary example of a con-versational user interface that leverages large language models.The whitepaper looks at ChatGPT from the perspective of students and lecturers.It focuses on everyday areas of high

10、er education:teaching courses,learning for an exam,crafting seminar papers and theses,and assessing students learning outcomes and performance.For this purpose,we consider the chances and concrete appli-cation possibilities,the limits and risks of ChatGPT,and the underlying large language models.Thi

11、s serves two purposes:First,we aim to provide concrete examples and guidance for individual students and lecturersto find their way of dealing with ChatGPT and similar tools.Second,this whitepaper shall inform the more extensive organizational sensemaking pro-cesses on embracing and enclosing large

12、language models or related tools in higher educa-tion.We wrote this whitepaper based on our experience in information systems,computer science,man-agement,and sociology.We have hands-on experience in using generative AI tools.As professors,postdocs,doctoral candidates,and students,we constantly inno

13、vate our teaching and learning.Fully embracing the chances and challenges of generative AI requires adding further perspectives from scholars in various other disciplines(focusing on didactics of higher education and legal aspects),university administrations,and broader student groups.Overall,we hav

14、e a positive picture of generative AI models and tools such as GPT-4 and ChatGPT.As always,there is light and dark,and change is difficult.However,if we issue clear guidelines on the part of the universities,faculties,and individual lecturers,and if lecturers and students use such systems efficientl

15、y and responsibly,our higher education system may improve.We see a great chance for that if we embrace and manage the change appropriately.Table of Contents 4 Table of Contents Executive Summary.3 Table of Contents.4 About this Whitepaper.5 About the Projects ABBA,DeLLFi,and S3G.7 1.Introduction.9 2

16、.Overview on Generative AI,Large Language Models,and ChatGPT.13 3.Guidance for Students.18 4.Guidance for Lecturers.28 4.1 Teaching.28 4.2 Assessments(exams,seminar papers,theses,and the like).31 5.Outlook.39 References.43 About this Whitepaper 5 About this Whitepaper Acknowledgments We gratefully a

17、cknowledge funding by the German Federal Ministry of Education and Research(BMBF),the Ministry of Science,Research and the Arts(MWK)of Baden-Wrttemberg,the Bavarian State Ministry of Science and Art(STMWK),and the Hessian Ministry of Higher Education,Re-search,Science and the Arts(HMWK)for the ABBA

18、project(Grant Numbers 16DHBKI002,16DHBKI003,16DHBKI004,16DHBKI005).Further,we gratefully acknowledge funding by Stiftung Innovation in der Hochschullehre for the projects DeLLFi,and S3G(Grant Numbers FBM2020-EA-1670-01800,FRFMM-677/2022).The logos of our institutions on the cover page shall acknowle

19、dge the academic environment that shapes our thinking.In writing this piece,the authors used ChatGPT(Ver.Feb.13 and Mar.14,2023),Grammarly(Ver.6.8.261),DeepL(Ver.4.4.2.7961),and Microsoft Word to improve the linguistic presentation of their thoughts.Full responsibility for the contents lies with the

20、 authors.Further readings This whitepaper is not the first information about generative AI,large language models,and ChatGPT in higher education.We strongly recommend reading and listening more broadly.For ex-ample,engage with the educator considerations for ChatGPT provided by OpenAI at https:/plat

21、- such as Vogelgesang et al.(2023),news re-ports such as https:/ websites such as https:/www.lmu.de/en/newsroom/news-over-view/news/what-should-we-make-of-the-chatbot-chatgpt.html,blog posts such as on https:/hochschulforumdigitalisierung.de/en/,and videos such as https:/youtu.be/bbB9Ve4BzSY.In addi

22、tion,Dwivedi et al.(2023)provide a broad overview of ChatGPT in research and practice.Fur-ther,this whitepaper cannot and should not bypass the need for ministries,universities,and faculties to deeply think about supporting students in acquiring the right skills and leveraging tools,and how to set c

23、lear policies and guidelines.Instead,the whitepaper shall complement and inform such dis-cussions.In addition,it shall inform students in reflecting on their assessment and potential usage of ChatGPT and related systems.Further use This whitepaper is part of the Hohenheim Discussion Papers in Busine

24、ss,Economics and Social Sciences with ISSN 2364-2084.This material is licensed under the Creative Commons Attribution ShareAlike 4.0(CC BY-SA)license and may be edited,reproduced,and distributed in any format or medium for any purpose,including commercial,provided that the authors are credited and t

25、he material is republished under the same conditions.Legal code:https:/creativecommons.org/licenses/by-sa/4.0/en/legalcode Feedback We highly appreciate feedback.Please get in contact via digitaluni-hohenheim.de.About this Whitepaper 6 Recommended citation Gimpel,H.,Hall,K.,Decker,S.,Eymann,T.,Lmmer

26、mann,L.,Mdche,A.,Rglinger,R.,Ruiner,C.,Schoch,M.,Schoop,M.,Urbach,N.,Vandirk,S.(2023).Unlocking the Power of Generative AI Models and Systems such as GPT-4 and ChatGPT for Higher Education:A Guide for Students and Lecturers.University of Hohenheim,March 20,2023.Image sources Logos on the cover page

27、are property of the respective institutions.The copyright for images on the cover page and prior to the main sections is with https:/ https:/ University of Hohenheim,Deans Office of the Faculty of Business,Economics and Social Sciences Schloss Hohenheim 1B,70593 Stuttgart,Germany.The University of H

28、ohenheim is a public body legally represented by its President,Prof.Dr.sc.agr.Stephan Dabbert.Regarding sales tax,the VAT ID no is DE147794207 and the tax no is 99134/02103.The Deans Office of the Faculty of Business,Economics and Social Sciences is a legally non-independent institution of the Unive

29、rsity of Hohenheim.The employee responsible for this publication is Prof.Dr.Henner Gimpel.Please get in contact via mail or e-mail to digitaluni-hohenheim.de.About the Projects ABBA,DeLLFi,and S3G 7 About the Projects ABBA,DeLLFi,and S3G ABBA stands for“AI for Business|Business for AI.”It is a publi

30、cly funded joint project to establish a cross-university,modular offer for developing AI competencies of students in management,business administration,economics,information systems,and related fields.Using artificial intelligence(AI)technologies in business requires specific skills.In addition to t

31、ech-nical expertise,the business world requires knowledge to embed AI technologies in business pro-cesses,work systems,products and services,design and evaluate business cases,and manage AI-based information systems in the long term.This bridge-building role falls primarily to managers as the centra

32、l decision-makers in the company.Therefore,the joint projects target group includes busi-ness administration,information systems,and other courses of study relating to socio-economic and socio-technical topics.The goals of the joint project are developing and providing a modular teaching toolkit for

33、 AI,which teaches business students interdisciplinary AI competencies in a scientifically sound and practice-oriented manner.The modular toolkit supports teaching for bachelor,master,executive master,and doctoral students at research universities and universities of applied sciences.To achieve that

34、goal,the joint project unites twelve professorships from three research universities and one university of applied sciences,who share the focus on supporting the development of AI competencies:University of Hohenheim,University of Bayreuth,Frankfurt University of Applied Sci-ences,Karlsruhe Institut

35、e of Technology.Fraunhofer FIT is an associated partner.Teaching con-tents and formats are jointly developed,mutually used,and in part made publicly available.Com-pared to individual creation by each university and professorship,this substantially strengthens the breadth and depth of the offering an

36、d the efficiency and quality of teaching.The German Ministry of Education and Research and the State ministries for science in Baden-Wrttemberg,Bavaria,and Hesse support ABBA financially.DeLLFi is a project at the University of Hohenheim funded by the Foundation for Innovation in Uni-versity Teachin

37、g.It stands for integrating digitalization along teaching,learning,and research.The project integrates three departments of the Faculty of Business,Economics and Social Sciences,the CareerCenter Hohenheim,and,with the Office for University Didactics and Humboldt reloaded,two organizational units tha

38、t support teaching.S3G stands for Smart Sustainability Simulation Game and is a cooperative project between the Uni-versity of Hohenheim,the University of Bayreuth,and the Augsburg University of Applied Sciences.S3G develops a student-centered teaching program that focuses on techno-economic compete

39、ncies in the context of the central business challenges of our time:Digital transformation and sustainable transformation.The focus is on applying machine learning in different cases of an industrial com-pany.Teamwork,gamification,and competitive elements are incorporated in authentic challenge situ

40、ations.The Foundation for Innovation in University Teaching funds S3G.All three projects share one thing:they investigate and implement concrete AI-centered innovations in higher education.This whitepapers authors work for at least one of the partners in ABBA,DeLLFi,and S3G.About the Projects ABBA,D

41、eLLFi,and S3G 8 1.Introduction Introduction 9 1.Introduction“Would ChatGPT get a Wharton MBA?”This headline circulated in the media at the beginning of 2023,causing universities worldwide to sound the alarm,but also justified?On March 14,2023,OpenAI launched GPT-4 as a model that can power ChatGPT.T

42、his substantially improves perfor-mance over what has been discussed regarding ChatGPT over the last months.GPT-4 markedly outperforms the previous model GPT-3.5,for example,by scoring a higher approximate percentage among test takers on the uniform bar exam(OpenAI,2023b).This underscores the releva

43、nce of whether ChatGPT based on GPT-4 would get a Wharton MBA or other degrees.ChatGPT is an artificial intelligence(AI)-based conversational agent that can write college-level es-says.“Conversational agent”is an umbrella term for software that uses natural language to interact with its user,either

44、text-based(as chatbots)or speech-based(as virtual assistants)(Gnewuch,Mo-rana,&Mdche,2017;McTear,Callejas,&Griol,2016).Due to recent technological advances in AI,conversational agents are already used in some contexts(Feidakis et al.,2019).Even before the release of ChatGPT,higher education has been

45、 an emerging application area for the deployment of conversational agents(Al Muid et al.,2021).These tools can provide the advantages of being per-manently available,scalable,and location-independently accessible,leveraging the potential to ad-dress multiple learners concerns simultaneously while ad

46、apting to their individual needs(Elshan&Ebel,2020;Hobert,2019).Then came ChatGPT.OpenAI launched ChatGPT as a research preview on November 30,2022.ChatGPT has already reached a new level of conversation between humans and conversational agents since it was introduced with GPT-3.5 as the underlying m

47、odel.Now GPT-4 further expands this.Despite the general knowledge of the potentially disruptive power of conversational agents in higher education,ChatGPT brought the discussion to a new level for several reasons.The quality of results produced by the underlying large language models exceeds what mo

48、st people thought pos-sible.ChatGPT,as a user interface to these models,is very easy to use for most people.OpenAI offered access to ChatGPT for free.In effect,many people who had not interacted with conversa-tional agents before or perceived them as relatively unsophisticated machines could get fir

49、st-hand experience with an advanced conversational agent generating stunningly high-quality texts.This swirled up the field of higher education and many other domains within just a couple of weeks.Much of higher education is teaching and learning how to apply state-of-the-art knowledge and cre-ate m

50、eaningful,valuable new knowledge.Both application and creation of knowledge frequently manifest in text,images,or other representations that can be easily communicated,stored,pro-cessed,and the like.Generative AI systems can generate text,images,or other representations with relatively little human

51、input.Hence,it is not surprising that the advent of generative AI fundamentally challenges accepted knowledge,assumptions,and behaviors in higher education.Questions arise regarding the division of labor between humans and machines,learning goals,and forms of as-sessing student performance.Large lan

52、guage models are a specific machine learning approach with superior performance on NLP tasks(OpenAI,2023c;Vaswani et al.,2017).GPT-4 is an example of such large language model.Such models and the systems built on top of them,like ChatGPT,are impressive but disputed.After its release,ChatGPT became t

53、he fastest-growing consumer application in history,reaching over 100 million monthly users just two months after launch(Hu,2023).The media has dubbed ChatGPT as part of“the generative-AI eruption”(Benson,2023)that may revolutionize the way we work,think,and approach human creativity(Benson,2023).How

54、ever,opinions from the context of higher edu-cation demonstrate a sharp divide between those who are enthusiastic about the technology and Introduction 10 those who are heavily concerned.Professor Ute Schmid from the University of Bamberg pointed out that the GPT-3 model has a deeper problem:the ina

55、bility to trace the sources of each statement and the process used to create them(Schmidt,2023).Professor Jochen Schiller from the Free University of Berlin further highlighted that the program sometimes produces nonsensical output(Schiller,2023).Decker(2022)suggested:“If people using the generated

56、material are not careful and respon-sible(and let us be honest,not everyone is),we may be flooded with a barrage of half-truths,mis-representations,and simply falsehoods,all of which seem plausible and are written using perfect language.”Despite these concerns,there is a broad consensus that ChatGPT

57、 can potentially trans-form how we learn and teach,as it can be utilized for various applications,including writing,transla-tion,professional communication,and personalized learning(Atlas,2023).This whitepaper is relevant beyond ChatGPT.ChatGPT is a specific user interface for specific large languag

58、e models.We frequently refer specifically to ChatGPT,which is at the forefront of the current debate among students,lecturers,and other higher education stakeholders.Besides ChatGPT,other conversational agents and applications using large language models have similar or comple-mentary text processin

59、g and generative capabilities.Further,numerous other AI tools support other aspects of teaching and learning.An excellent overview of tools for text generation,text translation,audio-to-text transcription,image generation,image manipulation,slide generation,audio and music generation,audio and music

60、 processing,video generation and processing,programming,mathe-matics,and other elements of learning and teaching is provided by Professor Albrecht Schmidt from LMU Munich at https:/www.hcilab.org/ai-tools-directory/.The frequent and specific discussion of ChatGPT in this whitepaper should not obscur

61、e the consideration that many of the aspects and recommendations we discuss here are more broadly relevant for large language models and other generative AI systems in higher education.A brief look at the history of higher education and experience from other domains shows that ex-pected innovations

62、are often exaggerated in the short run(Rudolph,Tan,&Tan,2023).A recent example are MOOCs(Massive Open Online Courses).Although they have been called“the death of higher education”(Rudolph,2014),the present shows that university courses and certificates continue to be highly valued by students(Rudolp

63、h et al.,2023).So,what does the rise of AI tools,as another relevant digital innovation step,mean to higher education?First and foremost,we should not make either of two possible mistakes:For one,we should not panic in light of questions about whether such systems would pass university entrance test

64、s,whether they would get an MBA,or whether they annul the relevance of university lectures,exams,or master theses.Further,we should not neglect their importance for higher education and prohibit their use.It is good that students no longer write their theses using pen and paper or a typewriter.It is

65、 good that they do not calculate all statistics by hand or with a pocket calculator.Spell check and grammar support or auto-completion of words by Microsoft Words built-in editing functions are standard.Likewise,using advanced,machine learning(ML)-based translation programs such as DeepL Translator

66、or Google Translate is standard.Grammarly,DeepL Write,Hemingway,and other NLP-based services improve linguistic quality.Google Scholar algorithmically supports navigating the scholarly literature.Soft-ware such as R,Stata,and SmartPLS support data analytics.Software libraries,software develop-ment e

67、nvironments,and online discussion forums support students in developing software code.Both students and lecturers routinely use these tools,and despite some potential downsides the net effect of using such IT tools in higher education is overwhelmingly positive.Some stakeholders in higher education

68、discuss whether the use of ChatGPT by students should be prohibited.Others counter that it should not be prohibited because universities cannot monitor and enforce the ban.The latter is true,but it is not the primary reason why ChatGPT should not be Introduction 11 banned.The primary reason is that

69、ChatGPT and other generative AI tools can improve higher ed-ucation.They have become part of the everyday life for lecturers and students in their domains of life beyond higher level education.Therefore,we should all join forces to ensure these students are as prepared as possible for their future c

70、areers.At the university,students need to learn how to use ChatGPT and similar tools purposefully,productively,and responsibly.We should neither panic nor wait calmly and prevent innovation.Instead,we should focus on how we can innovate education to harvest the potential of AI in teaching and learni

71、ng.In this whitepaper,we reflect on the tremendous opportunities of generative AI tools in higher edu-cation and the potentially harmful effects of their increasing use.The whitepaper can guide students and lecturers to reflect on their learning and teaching to make sense of the generative AI disrup

72、tion in higher education.To this end,we focus on the teaching-learning trifecta between students,lec-turers,and technological tools.We focus on teaching courses,learning for exams,writing seminar papers and theses,and assessing student learning and performance.Further potential for using large langu

73、age models exists throughout the student lifecycle and in other areas of higher education administration and operations.Introduction 12 2.Overview on Generative AI,Large Language Models,and ChatGPT Overview on Generative AI,Large Language Models,and ChatGPT 13 2.Overview on Generative AI,Large Lan-g

74、uage Models,and ChatGPT One of the most relevant conversational agents right now is ChatGPT,a tool released by the US company OpenAI on Nov.30,2022.Before we go into more detail about why ChatGPT has gained so much attention in a relatively short time,we will first go into the technical basics of Ch

75、atGPT and clarify terms and key constructs accordingly.The relation of these constructs is sketched schemati-cally in Figure 1.Figure 1.Depicting key concepts as nested sets Artificial intelligence(AI)is a broad field encompassing various techniques and approaches to create intelligent machines that

76、 perceive their environment and take actions.Machine learning is a subfield of AI that allows computers to learn and improve their performance on a task without being explicitly programmed using algorithms that can identify patterns and make predictions based on data.Gen-erative AI refers to AI syst

77、ems that generate new data or outputs,such as images,music,or text,rather than classify or process existing data.Typically,generative AI uses machine learning.Large language models(LLMs)are a type of machine learning model that can process and generate nat-ural language text.LLMs are a type of gener

78、ative AI because they can produce novel text outputs based on patterns and learn from large amounts of input data.Conversational agents,also known as chatbots or virtual assistants,are AI systems designed to engage in natural language conversations with humans.Conversational agents can use LLMs as a

79、 component to generate text responses that mimic human-like language and style.Key concepts related to Generative AIGimpel,H.,Hall,K.,Decker,S.,Eymann,T.,Lmmermann,L.,Mdche,A.,Rglinger,R.,Ruiner,C.,Schoch,M.,Schoop,M.,Urbach,N.,Vandirk,S.(2023).Unlocking the Power of Generative AI Models and Systems

80、 such as GPT-4 and ChatGPT for Higher Education:A Guide for Students and Lecturers.University of Hohenheim.ARTIFICAL INTELLIGENCE MACHINE LEARNINGGENERATIVE AILARGE LANGUAGE MODELSCONVERSATIONAL AGENTSCHATGPTGPT-4Overview on Generative AI,Large Language Models,and ChatGPT 14 ChatGPT is a conversatio

81、nal agent that uses the LLM GPT-3.5 or GPT-4.GPT stands for Generative Pre-Trained Transformer.It is OpenAIs acronym used in the name of large language models trained on data available on the Internet.The earliest of such models,GPT-1,was published by OpenAI in 2018;the most recent one GPT-4 on Marc

82、h 14,2023.ChatGPT is a user interface to the models GPT-3.5 and GPT-4.The free ChatGPT research preview allows access to GPT-3.5.Since March 14,2023,the fee-based premium version ChatGPT Plus additionally provides access to GPT-4.The GPT models are also available via an application programming inter

83、face(API).In July 2022,OpenAI unveiled DALLE 2,a cutting-edge text-to-image model.Soon after,Stability.AI released a related open-source version:Stable Diffusion.Both these systems gained popularity and demonstrated impressive capabilities in terms of converting user-generated text prompts to images

84、.These are examples of generative AI beyond large language models.Machine learning technology is used in software systems to develop models which are embedded in other software systems,such as conversational agents or word-processing software.Innovation in higher education does not come from technol

85、ogy or models in general but from using software systems built on top of technologies and models.Figure 2 provides a rough sketch of this interrelation of technologies,models,and systems.The specific software systems,like ChatGPT,are essential for students and lectures.Hence,we frequently refer to C

86、hatGPT as a specific example of a gener-ative AI system.ChatGPT is impressive and topical.However,it is neither the only nor the last gen-erative AI system.Hence,university policies should not clarify the use of ChatGPT specifically but generative AI systems in general.Figure 2.From technology over

87、IT systems to IT use From technology over IT systems to IT useGimpel,H.,Hall,K.,Decker,S.,Eymann,T.,Lmmermann,L.,Mdche,A.,Rglinger,R.,Ruiner,C.,Schoch,M.,Schoop,M.,Urbach,N.,Vandirk,S.(2023).Unlocking the Power of Generative AI Models and Systems such as GPT-4 and ChatGPT for Higher Education:A Guid

88、e for Students and Lecturers.University of Hohenheim.RAW MODELDATAe.g.,books,online discussions,TRAINED MODELe.g.,GPT-3.5,GPT-4SOFTWARE SYSTEMe.g.,ChatGPTUSERe.g.,studentTASKe.g.,write thesisEngineering:creating IT systems based on technologyIT UseOverview on Generative AI,Large Language Models,and

89、ChatGPT 15 Now,why is ChatGPT becoming so prevalent in such a short period?Due to its high popularity,there is no surprise that the World Economic Forum in Davos,Switzerland,focused on the rise of AI,specifically the text-generator ChatGPT.In this context,Ina Fried,author of Axios Login states:“For-

90、get crypto and blockchain:The tech conversation at this years World Economic Forum in Davos is all about the rise of artificial intelligence,particularly the text-generator ChatGPT”(Fried,2023).One main advantage of large language models such as GPT-3.5 and GPT-4 underlying ChatGPT is their ability

91、to process and contextualize text information and generate appropriate responses(Susnjak,2022).GPT-3.5 is purely text-based:it accepts text input and produces text output.This text can be natural language or computer code.Beyond that,GPT-4 also accepts images as part of the input prompt.That could b

92、e documents with text and photographs,diagrams,or screenshots(OpenAI,2023a).Hence,GPT-4 includes elements of computer vision and is more than a pure large language model.The models performance and accuracy are significantly affected by the difference in the number of model parameters and the trainin

93、g data size.A larger training dataset leads to a more extensive and diverse language model,resulting in enhanced proficiency.According to OpenAI,GPT-3.5 is based on a deep neural network model with 175 billion parameters,which are adjusted through machine learning(Vogelgesang,Bleher,Krupitzer,Stein,

94、&Jung,2023).In con-trast to the previous models from OpenAI,GPT-1 and GPT-2,which aimed to generate human-like conversations,GPT-3 was already considered the most advanced and capable of the models due to its significantly larger size and the scaling up of data and parameters used in its training(Sh

95、reya,2023).GPT-3.5 is the successor of GPT-3.GPT-4 is the next evolutionary step.Both models work in English,German,and many other languages.According to OpenAI(2023a),they applied GPT-3.5 and GPT-4 to bar exams designed for human test takers.GPT-3.5s score was around the bottom 10%of humans typical

96、ly taking the exam.GPT-4s score is more around the top 10%of test takers.However,as any IT system(and any human),ChatGPT and the underlying models are not perfect.One fundamental limitation to keep in mind is that while the output might sound convincing,it is not necessarily factually correct.The GP

97、T models and,hence,ChatGPT sometimes“hallucinate.”Nat-ural text,academic references and the like might look perfect at first sight but lack truth or refer to non-existent objects or prior texts.For example,it may make up fake academic papers using new configurations of existing titles,outlets,and au

98、thors(Smerdon,2023).While this concern applies to all GPT models,our initial tests suggest that GPT-4 has substantially fewer hallucinations than GPT-3.5.One of the essential features of GPT-3.5 and GPT-4 is their capability to consider contextual infor-mation of a conversation when generating the r

99、esponse.This allows ChatGPT to maintain a conver-sational flow,yet its ability to maintain that flow continuously is limited.This is possible using a so-called“self-attention mechanism”(OpenAI,2023c;Vaswani et al.,2017),meaning that the models can weigh the importance of different words and phrases

100、in the input text based on their perceived relevance.The extent of the models context is given by the number of so-called input tokens.GPT-3.5 has a token limit of around 4,000 tokens(relating to about 3,000 words in the prompts and the models answers),GPT-4 an impressive 32,000 tokens.Hence,GPT-4 i

101、s substantially more robust in considering context affecting the length of text that can reasonably be generated.Based on GPT-3.5,ChatGPT can depending on the topic and the prompt reasonably produce a paragraph of text or a few paragraphs.Leveraging GPT-4,ChatGPT can potentially produce entire chapt

102、ers.It is easy to foresee that such models will eventually be able to create text in the length of a whole bach-elors or masters thesis.However,even if the text is linguistically perfect,its content does not nec-essarily show the same quality.Furthermore,ChatGPT cannot fully appreciate the nuances o

103、f sub-tleties of a conversation since it is still a machine learning model and is just trained on large amounts of input data(Azaria,2022).Overview on Generative AI,Large Language Models,and ChatGPT 16 A benefit of ChatGPT is its easy application,which is possible through a simple user interface tha

104、t does not require knowledge of using an API(Figure 3).ChatGPT is a ready-to-use service made available as a dialogue system with which one can easily interact.Immediately after prompting ChatGPT,the user is presented with relevant information about the topic of interest.Currently,OpenAI provisions

105、ChatGPT based on a freemium pricing model:ChatGPT usage is free of charge if the user accepts GPT-3.5 as the underlying model,somewhat tardy replies,and non-availability at times of high demand.A premium version with higher capacity is now available for a fee to handle a large volume of requests whi

106、le ensuring fast response times.Further,the premium version allows access to GPT-4 with its advanced ability to consider context and the option for image-based prompts.Microsoft and others use the API provided by OpenAI to integrate the large language mod-els in other software systems beyond ChatGPT

107、.Figure 3.ChatGPT user interface(screenshot from Feb.28,2023)The potential applications of ChatGPT are extensive.Particularly in higher education,its capability to comprehend and react naturally to language input renders it valuable for various tasks.ChatGPT can be helpful for students in various ta

108、sks,such as writing assignments,summarizing,paraphrasing text,making grammatical corrections on text,and translating text.For lecturers,ChatGPT can help collaborate on written work,engage in intellectually stimulating conversations,conduct research,and assist with administrative tasks such as report

109、 writing(Atlas,2023).As a student and lecturer,there are various factors to consider when incorporating ChatGPT into higher education,including productivity and efficiency benefits,learning objectives and didactics,as well as ethical considerations and concerns about the impact on human job roles an

110、d intellectual property.It is essential to approach the integration of ChatGPT with thoughtful consideration and a holistic perspective,considering all relevant factors.In this sense,it is crucial to consider the limita-tions and risks of using ChatGPT in higher education,such as data bias,plagiaris

111、m,or quality of the produced text.In the following sections,we will provide guidance for effectively using ChatGPT and identifying areas for innovation in education.Overview on Generative AI,Large Language Models,and ChatGPT 17 3.Guidance for Students Guidance for Students 18 3.Guidance for Students

112、 Conversational agents are valuable tools for university students,aiding with academic work,saving time,offering accessibility,improving critical thinking skills,and improving language skills(Atlas,2023).In this section,we critically reflect on how students can(and should)use ChatGPT,an exam-ple of

113、a tool based on large language models,to maximize its potential benefits and avoid risks.This includes us also highlighting the limitations and dangers of ChatGPT.In summary,we have nine recommendations for students,as summarized in Figure 4.In the following,we detail these recom-mendations.Figure 4

114、.Summary of recommendations for students Recommendation 1:Respect the law and examination regulations Before discussing the possibilities of working efficiently with ChatGPT or other AI-based tools,we would like to highlight the need to respect the relevant national and regional legislation and the

115、indi-vidual examination regulations of each university,school,and course of study.Students must comply with these and observe any instructions in the examination regulations on using AI-based tools.Furthermore,students must follow the rules regarding quotations and good scientific practice.Even-tual

116、ly,they need to indicate whether a text is created by AI or even which information was given to the AI.01Respect the law and examination regulations02Reflect on your learning goals 03Use ChatGPT as a writing partner04Use ChatGPT as a learning partner05Iterate and converse with ChatGPT06Summarize lea

117、rning material with ChatGPT07Boost coding with ChatGPT08Beware of risks when using ChatGPT09Read the checklist at the end of this section before using ChatGPTRecommendations for StudentsGimpel,H.,Hall,K.,Decker,S.,Eymann,T.,Lmmermann,L.,Mdche,A.,Rglinger,R.,Ruiner,C.,Schoch,M.,Schoop,M.,Urbach,N.,Va

118、ndirk,S.(2023).Unlocking the Power of Generative AI Models and Systems such as GPT-4 and ChatGPT for Higher Education:A Guide for Students and Lecturers.University of Hohenheim.Guidance for Students 19 Recommendation 2:Reflect on your learning goals Higher education is more than acquiring domain-spe

119、cific skills.It also includes cultivating and en-hancing essential skills such as critical and structured thinking.Generative AI increases the need for these skills and affects the chances of developing them during your studies.Critical thinking refers to the ability to evaluate information,ideas,an

120、d arguments systematically and rationally.It involves questioning assumptions,analyzing evidence,and considering multiple perspectives to arrive at well-reasoned conclusions.Given that generative AI systems like ChatGPT produce large volumes of linguistically polished text,which may need to be revis

121、ed in content,critically engaging with texts is more important than ever.Using ChatGPT as a writing partner(see recommendation 3 below)requires critically inspecting your partners contributions before further processing them.Structured thinking refers to the process of organizing and categorizing in

122、formation logically and coherently.It involves breaking down complex concepts into smaller components,identifying pat-terns and relationships,and using this framework to generate new insights and solve problems.ChatGPT is good at elaborating short text of a few sentences or a paragraph on a given to

123、pic.ChatGPT is not(yet)good at developing a long text that builds up an argument step by step or analyzes an issue from multiple perspectives in detail.Here,structured thinking is required for de-veloping the scaffolding(potentially with support by generative AI),which can be filled step by step wit

124、h ChatGPT.Since ChatGPT makes it easier to put clear thoughts into beautiful text and increases the linguistic quality,students can focus more on the thoughts and structure and less on formulating prose.In addition,the skillful use of tools such as ChatGPT requires that you know which prompts will g

125、et you valuable results.Skillfully interviewing people so that you get the information you are looking for is an art.Optimizing individual search terms and entire search strategies for an Internet search is an art.Likewise,it is crucial to understand how to properly prompt ChatGPT and other generati

126、ve AI tools so that they effectively assist you in each task.On that account,prompt engineering emerges as a new,valuable skill.In short,with the advent of generative AI,critical thinking,structured thinking,the evaluation of text and other media,and prompt engineering gain in importance,the initial

127、 creation of text and other media becomes less important.By reflecting on the desired learning outcomes,students can,in the next step,determine how to integrate ChatGPT into their education best and use it to supplement and enhance their learning experience.Further,their learning goals and the offer

128、ings of different courses should guide their selection of specific courses.Recommendation 3:Use ChatGPT as a writing partner When promoting ChatGPT and asking if one can use it for writing assignments for students,the AI noted that it could not replace critical thinking and creativity,which are esse

129、ntial components of writing assignments(Figure 5).Therefore,students in higher education should only use ChatGPT as a supplementary tool,such as Wikipedia,Google,or translation programs such as DeepL(Span-nagel,2023).Therefore,ChatGPT will never be the author of a work,and the responsibility for the

130、 written content will always lie with the human author,in this case,the student(arXiv,2023).Since ChatGPT is not an author,ChatGPT cannot be cited as a source.Furthermore,because ChatGPT can always generate new data,the answers it provides are not verifiable.Thus,students should adhere to the princi

131、ples of good scientific practice issued by each university.We will elaborate more on this issue in the assessment section(see Guidance for Lectures,Assessments,Recommenda-tion 6).Guidance for Students 20 Figure 5.Answer from ChatGPT on writing assignments(screenshot from 08.03.2023)Weels(2022)emphas

132、izes that ChatGPT might be(only)a good writing partner.This means that although ChatGPT provides some topics and perspectives on a given topic,the user is still required to add further knowledge and creativity,which is a large part of creating essays or theses.Further,despite the daily increase in t

133、echnological advancements in large language models,there are also limitations to the technology.The underlying models of ChatGPT were primarily trained on data up to September 2021.A typical claim is that ChatGPT is not obtaining new training data from the Internet,which restricts it to the data set

134、 and information it was trained on and what users feed into it.That is not always comprehensive,as can be seen by,for example,prompting ChatGPT:“Please tell me a few sentences about the earthquakes in Turkey and Syria in 2023”.ChatGPT is not able to answer with any specifics about the event.Also of

135、concern is that while a text generated by ChatGPT may sound plausible,it may also contain hallucinations,false information,or meaningless information.In addition,ChatGPT can generate sources that seem very realistic in appearance and syntax;however,being entirely fabricated.This can be dangerous for

136、 students who rely on it for literature research.Although ChatGPT,with the underlying model GPT-4,is now capable of generating sources with an attached Digital Object Iden-tifier(DOI),there is still plenty of room for erroneous outputs.Partially,the DOIs are also completely fabricated;partially,they

137、 exist but refer to a different paper than ChatGPTs text output.Nonetheless,it may only be a matter of time before ChatGPT is connected to Internet search engines(such as Bing,Google),and services providing bibliographic information(such as Google Scholar,Scopus,Web of Science).In fact,Microsoft alr

138、eady combines ChatGPT with Bing and the ChatGPT for Google browser extension for Chrome and Firefox shows ChatGPT answers alongside search re-sults from Google,Baidu,DuckDuckGo and others.Other AI-based language models(e.g.,Perplex-ity)can even aid in literature research,as they link citations to th

139、eir sources.Consequently,students and lecturers must verify any information they receive from the AI-generated content.Otherwise,there might be the risk that students use incorrect information or false citations to create user-generated content.Further,the AI model could be trained on false informat

140、ion and thus recreate a spiral of misinformation on these platforms.To effectively use AI-based writing tools for scientific topics,students must have a basic understand-ing of the topic and the limitations of the technology.Once this understanding is achieved,tools such as ChatGPT can assist in pro

141、viding ideas and a basic text structure(Friedrich,2023).For this,ChatGPT can be accessed with OpenAIs ChatGPT website or via integration into Google Docs or Microsoft Word(e.g.,with docGPT).Table 1 presents exemplary use cases,following Atlas(2023),where ChatGPT can act as a“co-partner”for formulati

142、ng text on a high linguistic level.Guidance for Students 21 Table 1.Exemplary prompts on writing text Activity to be supported Exemplary prompts Generate a headline for an essay“Generate five headlines for an essay about paste your text”Summarizing“Summarize the following text in two sen-tences:past

143、e your text”Paraphrasing“Paraphrase this sentence:paste your text”Correction“Act as a professional spelling and grammar corrector and improver”Wait for response Paste your text Proofreading and editing“Please revise the following sentence to make it clearer and more concise:Paste your text”Recommend

144、ation 4:Use ChatGPT as a learning partner Another useful application of ChatGPT is as a learning partner for university students.Since promi-nent conversational agents such as ChatGPT are accessible 24/7,they provide students with nu-merous opportunities to help them acquire new knowledge or test ex

145、isting knowledge.Since many best practices are already circulating on the internet,we compiled and tested what we consider to be the most helpful ones in a short table(Table 2).A good example of how the integration of AI models such as GPT-4 into existing learning apps can succeed has been shown by

146、the language learning app Duolingo.The app uses the new GPT-4 model to provide the user with new features such as“Roleplay”or“Explain my Answer.“With Roleplay,users can practice real-world conversation skills with fictitious characters.“Explain my an-swers”offers learners the opportunity to receive

147、an explanation of why their answer was right or wrong.Further,they can ask for further clarification or examples(Duolingo Team,2023).These feedback mechanisms might also be possible with ChatGPT if students choose the right prompts to convert with ChatGPT(notice the prompt on“Self-testing of specifi

148、c knowledge”).Guidance for Students 22 Table 2.Exemplary prompts on learning with ChatGPT Activity to be supported Exemplary prompts Generation of pattern solutions from old exam tasks“Please generate a pattern solution for the following task:past your task”Create a mind map to gain a quick overview

149、 of a new topic paste your topic“Create a mind map on the topic above,list the central idea,main branches,and sub-branches”Explanation of concepts (e.g.,mathematical equations)“I want you to act as a math teacher.I will provide some mathematical equations or con-cepts,and it will be your job to expl

150、ain them in easy-to understand terms.My first question is:I need help to understand how paste your concept works”Vocabulary acquisition “Could you please provide me with terms related to paste your text”Create flashcards “Topic:paste your topic Please help me create a two-column spreadsheet with que

151、stions and corresponding answers on the topic above”Self-testing of specific knowledge“Topic:paste your topic Please ask me five questions on the topic above.I will then respond to it.After my response,you will tell me if my answer was right or wrong and provide an explanation”Recommendation 5:Itera

152、te and converse with ChatGPT When using ChatGPT,it is sometimes unclear how to use it effectively.Providing ChatGPT with accurate and specific information is crucial to obtain the desired text and information.Since ChatGPT relies solely on prompts and words,providing additional information and conte

153、xt,including the in-tended purpose and information on the target audience,unique position,and intended tone,is es-sential.If the text or result is unsatisfactory,users can request more information and provide detailed feedback to improve the models response and generate a better match for their requ

154、irements(Atlas,2023).Possible prompts from Atlas(2023)that are consistent with these principles include:“Summarize this text and highlight why X has a relevant role”“Write a text on the topic A from the perspective B with the target audience C in a D tone”“Rewrite the text to make it sound more like

155、 A and highlight the bene-fits of B more”Guidance for Students 23 Further,ChatGPT offers a“Regenerate Response”option.If one is unsatisfied with the result,this option produces a new response for the same prompt.In addition,one can manually adapt the prompt or continue conversations with prompts lik

156、e:“Please shorten the summary to 150 words”“Please elaborate the second point”“Tell me more about the last argument in your previous answer”Recommendation 6:Summarize learning material with ChatGPT You may use ChatGPTs functionality with videos and texts,like lengthy notes from a lecture or a diffic

157、ult-to-read,lengthy paper.This may be a helpful way to skim learning materials and identify important aspects of the material.In doing so,one should be aware that a summary always misses details and that these details might be necessary.Therefore,this functionality should be used with caution.A nice

158、 gimmick is automatically summarizing the automatically generated transcript from a(YouTube)video.Several third-party browser plug-ins(e.g.,YouTube Summary with ChatGPT)pro-vide students with an opportunity to generate summaries of YouTube videos based on ChatGPT.The plug-in accesses and exports the

159、 standard YouTube transcript and transmits it to ChatGPT,requesting a summary which is then presented to the user via the browser plug-in.The browser plug-ins are available for Chrome,Safari,and Firefox.After installation,the plug-in enables students to access the transcript of the video and utilize

160、 ChatGPT to create a summary.The extension also facilitates easy navigation to the current time frame of the video and permits the copying of tran-scripts.The plug-ins support multiple languages and allow for the customization of prompts to im-prove the accuracy of the summary,as shown in Figure 6.F

161、igure 6.Customize prompt for YouTube summary(screenshot from 08.03.2023)Guidance for Students 24 Recommendation 7:Boost coding with ChatGPT ChatGPT can be used not only as a learning or writing partner but also as a partner to generate or correct codes.With the right prompt,ChatGPT can give students

162、 suggestions on coding.In Figure 7,we depict a potential prompt with code as the response,which we retrieved and modified from an expert talk at HCUM Munich(Hauck-Thum et al.,2023).These code snippets can then be used as a starting point for a project or task.ChatGPT can be an excellent tool to assi

163、st in debugging code.Students can share their code with ChatGPT,and the model can help identify errors and suggest possible solutions.Furthermore,ChatGPT can assist in optimizing code to make it more efficient by suggesting better data structures and algorithms.Figure 7.Exemplary prompt for code gen

164、eration with ChatGPT(screenshot from 08.03.2023)Table 3 provides examples of prompts that students can use to improve their code.Additionally,OpenAI has developed a more advanced tool called GitHub Copilot,which can help write code faster and with less effort than ChatGPT.GitHub Copilot is an AI pai

165、r programmer that utilizes a generative pre-trained language model created by OpenAI.By analyzing the comments and code,it instantly suggests individual lines and entire functions to improve the code.Guidance for Students 25 Table 3.Exemplary prompts on coding with ChatGPT Activity to be supported E

166、xemplary prompts Help with syntax“What is the syntax paste your text in Py-thon?”Coding examples“Please give me an example of how to use the paste your function in JavaScript?”Programming tips“What are some tips for paste your text?”Debugging help“How can I fix the TypeError:int object is not subscr

167、iptable error in my Python code”Explaining codes Can you explain the code for me step by step:paste your code?”Recommendation 8:Beware of risks when using ChatGPT In addition to the advantages of utilizing ChatGPT for creating scientific text,users should be aware of potential sources of error and m

168、isconduct.Thus,although we recommend using AI-based tools,such as ChatGPT,for creating scientific text(if there is no explicit prohibition by the examination regulations or the university),we encourage students to reflect on each AI-generated outcome.This applies not only to text but also to code ge

169、nerated by ChatGPT.Rademacher(2023)and others discovered issues with accurate citation of ChatGPT-generated sources.In some instances,ChatGPT created non-existent sources,posing a risk of propagating false information.Thus,students must verify each statement made by ChatGPT,which is a consid-erable

170、workload.Further,there are substantial risks relating to accidental plagiarism and copyright infringements.Please see our discussion relating to recommendation 6 for lectures in section 4 below for details.One is responsible for writing if one submits a work under ones name(e.g.,a seminar paper or a

171、 thesis).One is still responsible for the work if one uses content generated by ChatGPT or similar tools and includes them in ones work.One accepts the praise if the work gets praise(e.g.,a good grade).One must accept the blame and other negative consequences if the work triggers blame e.g.,for plag

172、iarism,copyright infringement,unethical discrimination,unlawful content,or errors.For good or for bad,the author is responsible for her or his use of tools.Guidance for Students 26 Recommendation 9:Read this checklist before using ChatGPT ChatGPT and other tools based on generative AI will change ho

173、w students learn,write exams,and study for tests.In the following,we summarize example recommendations that should be considered while working with ChatGPT:Review the universitys rules and regulations regarding generative AI,Large Language Mod-els,and ChatGPT(considering policies for usage,acknowled

174、gments,citations,etc.).Understand the capabilities and limitations of ChatGPT Check if the use of ChatGPT is clever or if the task requires the learning of basic knowledge Verify that the results given by ChatGPT are trustworthy and accurate and reflect the findings(Atlas,2023)Consider which topics

175、could be cleverly linked to produce novel insights.The variety of smart use cases of ChatGPT is enormous,and with time new ways of using it are sure to come.Students should empower themselves to use ChatGPT responsibly so that the first reflex of the(public)perception is not the possibility of cheat

176、ing but the possibility of learning new things that will lead to mature students being prepared for digital work.Guidance for Students 27 4.Guidance for Lecturers Guidance for Lecturers 28 4.Guidance for Lecturers The emergence of new educational technologies often arouses strong emotions,ranging fr

177、om doomsday predictions to endless euphoria(Rudolph et al.,2023).In the case of ChatGPT,experts also speak of an“educators dilemma”between banning these technologies or promoting their use(Teubner,Flath,Weinhardt,van der Aalst,&Hinz,2023).Although the use of conversational agents and ChatGPT,particu

178、larly in higher education,is diverse,we aim to examine two main areas of application for teachers and lecturers in more detail:(1)the teaching process and the(2)the assess-ment process.4.1 Teaching There are numerous potentials for the teaching side across all stages of teaching-related activities,f

179、rom planning and implementation to evaluation.In the following,we present different recommenda-tions where generative AI systems,and especially ChatGPT,can support lecturers in their teaching activities.We illustrate this with five recommendations for lecturers regarding teaching,as summa-rized in F

180、igure 8.In the following,we detail these recommendations.Figure 8.Summary of recommendations for lecturers regarding teaching Recommendation 1:Reflect on which learning objective you are pursuing in your teaching Before considering how to use ChatGPT to support your teaching,it is essential to deter

181、mine the actual learning objectives of your course.As mentioned earlier in the student section,learning ob-jectives in higher education may vary depending on the field of study and subject matter.While critical thinking and structured thinking are considered essential goals in higher education,you c

182、an utilize ChatGPT to develop these skills in your students by utilizing the limitations of generative AI,such as the potential for false information or the potential to receive text of low quality.By reflecting on the output generated by AI tools and providing ChatGPT with purposeful prompts and su

183、fficient information,students can practice structuring their arguments logically,which is an essential com-ponent of structured thinking.This interaction with generative AI can effectively develop students Recommendations for Lecturers-Teaching01Reflect on which learning objective you are pursuing i

184、n your teaching02Create learning materials with ChatGPT03Support students with quizzes 04Boost learning with ChatGPT05Encourage students to use ChatGPTGimpel,H.,Hall,K.,Decker,S.,Eymann,T.,Lmmermann,L.,Mdche,A.,Rglinger,R.,Ruiner,C.,Schoch,M.,Schoop,M.,Urbach,N.,Vandirk,S.(2023).Unlocking the Power

185、of Generative AI Models and Systems such as GPT-4 and ChatGPT for Higher Education:A Guide for Students and Lecturers.University of Hohenheim.Guidance for Lecturers 29 skills in critical reflection and structured thinking in an iterative manner.Today,generating high-quality output with ChatGPT depen

186、ds on creating the correct prompts.The ability to create and refine prompts that are tailored to specific tasks or goals can be crucial in achieving the desired outcome.This makes domain-specific prompt engineering expertise a valuable skill for the future.Recommendation 2:Create learning materials

187、with ChatGPT ChatGPT can be a valuable tool for personalized learning in higher education.In addition to support-ing lecturers with various classroom tasks,ChatGPT can create custom exercises and quizzes,offer feedback,and generate tailored educational materials that align with a students learning s

188、tyle and progress.Moreover,ChatGPT can assist in developing lecture ideas,drafting seminar plans and module descriptions,and crafting announcement texts.Another possible application of ChatGPT is to assess students prior knowledge using AI.In Figure 9,we illustrate a ChatGPT-prompt that can be used

189、to create a lesson plan on a specific topic.Furthermore,as mentioned in the students section,lecturers can utilize the“YouTube Summary with ChatGPT”feature to summarize video transcripts and design classroom elements.YouTube automatically generates transcripts of videos(be it learning videos produce

190、d by the lecturer or third-party videos).A click on the plug-in button sends the transcript to ChatGPT with a prompt to provide a summary.Figure 9.Prompt to ChatGPT:Create a lesson plan(screenshot from 28.02.2023)Recommendation 3:Support students with quizzes Quizzes on a courses contents can suppor

191、t students and assess their level of learning.ChatGPT prompts can help lecturers in creating quizzes.Figure 10 represents an example of a suitable prompt with the ChatGPT answer.Figure 10.Prompt to ChatGPT:Quiz(screenshot from 08.03.2023)Guidance for Lecturers 30 To stay close to the lectures conten

192、t,some lecturers first provide transcripts of their lectures to ChatGPT(e.g.,via“YouTube Summary with ChatGPT”)and then use prompts such as“Please gen-erate a single choice exam question about this content with four potential answers,exactly one of them being correct.”Such prompts can(and likely wil

193、l)also be used to generate exam questions.While this might be an exciting feature to spur ideas regarding exam questions,lecturers cannot rely on the questions and suggested answers being correct.Therefore,they should do more than accept and use them.This is true for all uses of ChatGPT,but it is es

194、sential when it comes to exams.Recommendation 4:Boost learning with ChatGPT ChatGPT cannot just help lecturers in designing course materials but also boost learning by over-coming three challenges that are typically hard to address in classrooms:helping students apply their knowledge to new situatio

195、ns,showing them that they may not know as much as they think they do;and teaching them how to think critically about information(Mollick&Mollick,2022).One way to overcome the challenges in teaching is to incorporate ChatGPT as a learning tool.This approach allows for utilizing AIs strengths and weak

196、nesses to enhance the learning experience.In Table 4,and according to Mollick and Mollick(2022),we present three ways to integrate ChatGPT into the curriculum.Table 4.How to boost learning with ChatGPT Learning Objectives ChatGPT Task Students Task Train transfer with ChatGPT ChatGPT can be used to

197、demonstrate and clarify con-cepts and to transfer them to a different context.However,AI is limited in under-standing complex relationships and combining information from different sources.The task of the students is to evaluate the AIs response as it transfers a concept to a different context.Furth

198、er,students should apply strategies to im-prove AIs output.Train evaluation with ChatGPT ChatGPT can be used to write essays about a particular topic.Although AI is good at simplify-ing text,it has weaknesses in providing insightful analysis.The task of the students is to continuously improve the AI

199、-cre-ated essay by providing further prompts to the AI,adding new information,or clarifying points.Help students to identify and close gaps in their knowledge with ChatGPT ChatGPT can be utilized to out-line the steps involved in a spe-cific process.However,the AI-generated pro-cess might not be com

200、plete.The task of the students is to evaluate and improve the AIs output by adding information from different sources Guidance for Lecturers 31 Recommendation 5:Encourage students to use ChatGPT The way teaching is done in higher education should adapt to the technological developments and the vario

201、us possibilities they offer.In this sense,ChatGPT is seen as a support rather than a threat.Lecturers should encourage students to use ChatGPT creatively and critically to improve,expand or vary their own texts,but not to replace or plagiarize them.This whitepaper might help inform students about th

202、e possibilities and risks of using ChatGPT and similar tools.By teaching students how to effectively use these tools,educators can equip them with important skills for their future careers,while also emphasizing the importance of academic integrity and originality.4.2 Assessments(exams,seminar paper

203、s,theses,and the like)One of the most common concerns with using ChatGPT is the fear that essays will become increas-ingly obsolete as an assessment method in higher education.Some lecturers are concerned that written assignments are being outsourced to ChatGPT without being detected by a plagiarism

204、 de-tector.Further,lecturers see the challenge that generated text reads naturally.While traditional pla-giarism detection tools can identify copied and pasted text from scientific sources,they are not equipped to identify text generated by AI models such as ChatGPT(Khalil&Er,2023).Tools such as the

205、 OpenAI Text Classifier,which have been developed to address the issue of false claims that a human wrote AI-generated text,are currently improving.However,the current success rate of the tool in recognizing texts generated from large language models is only around 26%.While this is still a remarkab

206、le achievement,it highlights the limitations of existing plagiarism detection software in its current state(Wiggers,2023).Traditionally,some lecturers might have taken the linguistic quality of a text as an indicator of its content quality.However questionable such a shortcut might always have been

207、with increasingly advanced language models,the potential link between linguistic quality and content quality is no longer there.These concerns only arise when lecturers do not engage in changing assessment for-mats and criteria.However,there is already a debate about the suitability of written essay

208、s to test students knowledge.The second fear is related to the inability of ChatGPT to truly understand the context and interpret text since ChatGPT is just predicting the probability of the next word in a sequence that has already been monitored(Arif,Munaf,&Ul-Haque,2023).As a result,institutions c

209、oncerned about AIs output could take this as legitimacy for implementing policies that prohibit the use of AI for quality reasons.Given the rapid technological development of AI applications and their prospects,such as the inte-gration of ChatGPT into widely used Microsoft products,it quickly become

210、s apparent that there is no alternative to using AI applications in the university context(Salz,2023).Universities and lecturers should therefore focus on ensuring responsible use by addressing potential challenges related to ChatGPT rather than implementing policies that restrict use(Brown et al.,2

211、020;Vogelgesang et al.,2023).Below,we present several ways teachers can turn the challenges associated with ChatGPT and assessment formats into opportunities to adapt to emerging changes in higher education.Ex-ams,seminar papers,and bachelor or master theses are the assessment formats mainly concern

212、ed about.In summary,we have eight recommendations for lecturers regarding assessments,as summarized in Figure 11.In the following,we detail these recommendations.Guidance for Lecturers 32 Figure 11.Summary of recommendations for lecturers regarding assessments Recommendation 1:Design exams carefully

213、 with respect to new technological possibilities Closed book exams,where the students write by hand,using only pen and paper,or using a com-puter in a controlled environment and a kiosk mode without Internet access,might be the easiest way to test the students knowledge in the future.ChatGPT is no m

214、ore of a threat to the assessment process in such a setting than traditional paper-based cheat sheets.However,such an assessment approach might be antiquated since students acquire much knowledge shortly before the exam that is subsequently forgotten.Thus,we notice a trend in higher education,not le

215、ast driven by the Covid-19 pandemic,from“pen-and-paper-only”exams to“open exams”or“take at-home exams”.When creating open exams where technical aids are also allowed,care should be taken to test the exam questions in advance with ChatGPT.In case of excellent results,the questions should be discarded

216、;in case of not good results,“regenerate”if there is still room for improvement,keep the question.The following are additional items to consider when administering exams that can help prevent student misconduct(Koenders&Prins,2023).Ask for personal reflection in your exam.Focus your exam questions o

217、n very recent events.ChatGPT,in its current version,was trained on data up to September 2021.Thus,it is not informative on more recent events.State on the exam which tools are allowed(e.g.,Stata)and which are not(e.g.,ChatGPT).All these suggestions have limitations.Assessing the correctness or quali

218、ty of personal reflections takes much work and is only possible to a certain degree.It is easy to foresee that their limits not 01Design exams care-fully with respect to new technological possibilities02Require students to declare how ChatGPT and other advanced tools were used03Innovate your assessm

219、ent formats04Rethink the supervision process for assignments05Rethink the evaluation criteria for assignments06Implement guidelines for avoiding plagia-rism and copyright infringements07Teach students how to use ChatGPT properly08Implement“Rules for Tools”Recommendations for Lectures AssessmentGimpe

220、l,H.,Hall,K.,Decker,S.,Eymann,T.,Lmmermann,L.,Mdche,A.,Rglinger,R.,Ruiner,C.,Schoch,M.,Schoop,M.,Urbach,N.,Vandirk,S.(2023).Unlocking the Power of Generative AI Models and Systems such as GPT-4 and ChatGPT for Higher Education:A Guide for Students and Lecturers.University of Hohenheim.Guidance for L

221、ecturers 33 being up-to-date will be annulled in the future with continuous updates of the models or integration of Internet search and language models.Despite these limitations,the general recommendation to critically rethink exams in light of generative AI remains.Recommendation 2:Require students

222、 to declare how ChatGPT and other ad-vanced tools were used Monitoring which tools are used in a take-home exam is impossible,and not all students might be truthful.However,explicit declarations about whether special tools were used or not increase the binding nature and the consequences in case of

223、misconduct,as is also standard practice in decla-rations of independence in,for example,theses.Cambridge University Press and other publishers have already implemented guidelines that require authors to disclose their use of AI-based tools,such as ChatGPT,when writing articles.The pub-lisher has ann

224、ounced that these guidelines protect authors who wish to utilize ChatGPT and similar AI-based programs,encouraging them to do so(Forschung&Lehre,2023).Hence,we suggest that universities develop such declarations of independence that explicitly address the usage of genera-tive AI tools.At best,such d

225、eclarations are not binary regarding using tools such as ChatGPT(e.g.,“I used ChatGPT”)but are differentiated like author contribution statements in some academic jour-nals.The statements should highlight which steps in the research and writing process ChatGPT and other tools were used for(e.g.,deve

226、loping an outline or proofreading).Further,such declarations should include a statement of student responsibility regarding potential errors,copyright violations,or plagiarism that technical tools inserted in their work.Further,lectures might require students to provide a list of prompts used.Even f

227、urther,they might require students to provide a full transcript of the conversations with ChatGPT that informed the students work.Browser plug-ins exist that allow to export and share a conversation with ChatGPT.Recommendation 3:Innovate your assessment formats ChatGPT has raised many concerns about

228、 its potential to undermine the effectiveness of assign-ments as an assessment method.This concern has been present since the early days of its imple-mentation.ChatGPT has the potential to be a significant innovation in higher education by enabling lecturers to use assessment as a tool for their tea

229、ching.However,few lecturers currently possess the necessary skills,making AI a potential educational milestone to take the assessment as a learn-ing(Earl,2012).One possibility to innovate assessment formats arises from the fact that ChatGPT has a restricted capacity for creativity.ChatGPT is trained

230、 on specific patterns and existing text,limiting its ability to generate original content or ideas.As a result,ChatGPT can only replicate what it has been trained on and what already exists and needs to be more capable of thinking creatively(Susnjak,2022).Consequently,there is potential for lecturer

231、s to explore innovative assessment for-mats that demand thinking beyond traditional boundaries.These assessment formats could lead to new outcomes through students creative linking of topics.Therefore,our recommendation is to steer clear of standardized assessments that could quickly be completed by

232、 a computer,as suggested by Herman(2022),and instead design assessments that promote students abilities to think creatively and critically,as advocated by Brookfield et al.(2019).Here are some examples to innovate your assessment formats:Administer specific assessments only during class(Rudolph et a

233、l.,2023).Encourage oral presentations to assess students public speaking skills,as well as their un-derstanding of the material(McCormack,2023).Guidance for Lecturers 34 Encourage collaborative group projects where students work in small teams to complete a specific task or project(McCormack,2023).P

234、romote critical thinking by requiring students to reflect on their learning through written or oral reflection.Let students prepare other forms of materials,for example,webpages,videos,and anima-tions that express critical thinking(McCormack,2023).None of these formats is a silver bullet.In one form

235、 or the other,they favor students being able to participate in person in the classroom,deal well with high time pressure,or have good oral commu-nication skills.Assessing individual contributions in group work is challenging,as is evaluating indi-vidual reflections.Further,ChatGPT and other AI tools

236、 can also support creating webpages,videos,and animations.Nevertheless,lectures should consider innovating their assessment formats in light of students ability to use generative AI tools and in light of the potentially changes learning goals to account for the diffusion of generative AI tools.In ch

237、anging the assessment formats,lecturers must consider the time required for the assessments.Potentially,the staffing needs to be increased to allow for meaningful assessments.Recommendation 4:Rethink the supervision process for assignments Since ChatGPT is particularly good at formulating text,it is

238、 almost impossible to detect whether AI or students wrote specific text passages of assignments.Thus,we encourage teachers to empha-size the supervision process more than the written assignment.Thus,teachers can better evaluate how the assignment was created and whether students acquire specific exp

239、ertise during the creation process.We thus align with the suggestions of Frlich-Steffen(2023)and propose to consider the following steps:No assignments without supervision process.Require information on work steps during the process.Place significant emphasis on study design and the careful executio

240、n of the study in empirical works.Require the use of ChatGPT as a work step including a clear indication of what ChatGPT was used for.If possible,add a presentation and oral defense of the results,as this shows how far the student has mastered the contents.As with recommendation 3 above,improving th

241、e supervision process might require additional time and,thus,staffing.Recommendation 5:Rethink the evaluation criteria for assignments Given AI tools exceptional ability to generate and compose a text,higher education must establish novel assessment criteria that surpass the mere formulation of text

242、 for evaluating assignments.Alt-hough ChatGPT occasionally makes mistakes in its content,its writing is often convincing.Therefore,students can assess the content of the text while placing less emphasis on the structure and writing style,which are areas where language models tend to excel.Hence,the

243、following evaluation criteria should be considered more seriously in cases where ChatGPT is not explicitly forbidden(Frlich-Steffen,2023).Quality and individuality of the research question as well as fit to the assigned topic Quality of the theoretical background,including proper references Coherenc

244、e of the presentation Guidance for Lecturers 35 Alignment of the research question,theories used,methods used,and results Unique(theoretical,empirical,or technical)contributions beyond summarizing literature Inclusion of personal reflections,such as a learning log or personal statement Recommendatio

245、n 6:Implement guidelines for avoiding plagiarism and copy-right infringements One of the most significant challenges in higher education will be detecting plagiarism.In 2000,The Guardian warned of a plagiarism epidemic and raised concerns that Google,Wikipedia,and Co.will make students stupid.Over 2

246、0 years later,we all know that none of these technology-driven developments made students stupid but brought innovations to higher educations teaching and learning process.However,with easy access to electronic texts at a massive scale,plagiarism be-came easier(e.g.,copy&paste).Many universities use

247、 plagiarism detection software that matches a newly submitted text against previously known texts to fight that.This is good for detecting blunt,direct plagiarism.Recent advancements in automatic translations and large language models ease the task of wrong-doers and complicate plagiarism detection.

248、Automated translations(e.g.,using tools such as DeepL or Google Translate)of preexisting texts into the language of the assignment evades direct text comparison.For some years,some students have used translation chains to paraphrase text.An example:You take an English text from literature,and automa

249、tically translate it to German,then to Spanish,then to Russian,and back to English.Even within a single service,the resulting text para-phrases the original.The discrepancy between original and paraphrased text likely increases when switching between translation services.This task could be more conv

250、enient for students.Large lan-guage models such as ChatGPT unfortunately ease this task.Students may,for example,prompt ChatGPT,“Please paraphrase the following text:“Even if not direct but paraphrased:If the source is not credited,this is plagiarism.However,it is more difficult to detect.A more fun

251、damental concern regarding plagiarism is whether ChatGPT might provide it without dis-closing it or the user noticing it.ChatGPT was trained on pre-existing texts and learned typical flows of text.It can quickly happen that text generated by ChatGPT is in parts identical to pre-existing text.With hu

252、man writers,it can happen by chance that one writer produces a short piece of text fully independent of the other,just by chance.However,with language models trained on the text and then reproducing some of that text,this is not by chance a type of plagiarism.Likely,this will not recreate large port

253、ions of a single text but may lead to what is known as patchwork plagiarism or mosaic plagiarism,where multiple different sources are interwoven without adequately referencing them.Further,ChatGPT is accused of copyright infringements.Thus,copying text generated by ChatGPT into a document one draft

254、puts one at risk of plagiarism and copyright infringement.Re-sponsibility for such violations of legal regulations or good scientific practice lies with the human author(s)of a work(arXiv,2023).Even if one might argue that this is accidental plagiarism from the users perspective,users of tools such

255、as ChatGPT should be aware of the risks and not be negli-gent.A new legal framework called“learningright”is already being discussed as an alternative to“copyright”(Malone,2023).This legal concept aims to balance the need to protect creators intellec-tual property rights while still allowing AI syste

256、ms to learn from and build upon existing knowledge.This responsibility of the human user relates not only to plagiarism and copyright infringement but also to“biased content,errors,mistakes,incorrect references,or misleading content”(arXiv,2023).Lecturers should inform their students about the risks

257、 and their responsibility for texts they submit under their names.Further,to prevent plagiarism,lecturers can implement guidelines to encourage students to deal responsibly with literature sources,for example,by forcing them to provide detailed notes or screenshots/photos of the literature base(Frli

258、ch-Steffen,2023).In summary,the handling of plagiarism should be consistent with the regulations of the university policy and their guidelines.Guidance for Lecturers 36 Finally,some consider the use of text generated by ChatGPT as plagiarism,as not the student de-veloped the text,but someone else,na

259、mely ChatGPT.Instead,we take the perspective that using ChatGPT and other tools is not a problem per se.However,using advanced tools such as ChatGPT requires transparency on tool use.Various“ChatGPT Content Dectors”or“AI Content Detectors”are already available to distinguish between human-written an

260、d machine-written text.Like plagiarism detection software,they analyze text,highlight dubious sections(in this case,likely computer gen-erated),and typically provide a percent value of how much of the text is computer generated or how likely it is computer generated.Currently,these detectors are far

261、 from perfect.They will improve in the future just like other systems try to avoid correct classifications.These imperfect content detector tools may be used to check the plausibility of the students disclosure.Recommendation 7:Teach students how to use ChatGPT properly Incorporating AI tools such a

262、s ChatGPT in higher education is not merely a threat to conventional assessment methods but can also aid in enhancing students academic performance.As not all learners possess strong writing skills or might encounter linguistic obstacles,utilizing large language models such as ChatGPT can promote eq

263、uity and fairness in the educational setting.By providing learners with a tool to generate well-composed texts,students can demonstrate their knowledge and comprehension of a subject matter rather than being handicapped by writing deficiencies or lan-guage barriers.Therefore,AI technology can be piv

264、otal in promoting inclusivity and excellence in higher education.Although AI can assist in creating substantial and insightful content,it is not a simple task and de-mands proficiency in both technical skills and knowledge of ethical considerations.Students need to be made aware of,for instance,ster

265、eotypical ChatGPT answers.As ChatGPT is trained on a large dataset of text,it is very likely that the data used to train ChatGPT may contain societal biases(Dahmen et al.,2023).Consequently,the model will reflect these biases in the output data and,thus,reinforce existing societal issues and discrim

266、ination(Atlas,2023).Therefore,when using ChatGPT in higher education,students should be aware of this potential bias and critically reflect on each statement created by ChatGPT.Further issues of copyright and intellectual property need to be discussed with students.The media recently reported that l

267、eading AI developers such as Microsoft,OpenAI,StabilityAI,and Midjourney are increasingly facing lawsuits over alleged copyright infringement in their programs outputs and the data they are trained on(Wiggers,2023).Students and lecturers who reuse AI-generated texts or other content risk infringing

268、the original authors copyright if the AI was trained on data and infor-mation subject to copyright,leading to incalculable legal consequences.To help students master the use of AI,lecturers can redefine their curriculum and,thus,explicitly teach the use of AI.This may include:Include subject-specifi

269、c reflection on the impact of AI into the curriculum.Develop study programs focused on AI in science,ethical implications of AI use,and knowledge creation through AI systems.Redefine core competencies in classes and reflect on what should be tested within each subject.Guidance for Lecturers 37 Recom

270、mendation 8:Implement“Rules for Tools”AI tools such as ChatGPT have immense potential for uncovering novel approaches to pedagogy.Nonetheless,while proficiently generating plausible information,AI risks disseminating false data,fabricated quotes,inaccurate information,plagiarism,and the like.Learner

271、s must possess adequate knowledge of the subject under scrutiny to achieve satisfactory outcomes.Establishing a transparent policy governing AI implementation in higher education represents a crucial stride toward fostering a learning setting wherein AI is embraced with accountability and candor.The

272、reby,“Rules for Tools”can help build an AI policy(Spannagel,2023).Based on Spannagel(2023)and in line with what we describe in other parts of this whitepaper,rules for tools may include:In general,students may use all types of media and tools,with the use of said tools subject to the course requirem

273、ents.Students are accountable for their achievements,as AI tools,such as ChatGPT,while capa-ble of generating well-composed texts,can still contain mistakes and violate regulations or norms.It is mandatory for students to report the aids used during a course,for example,listing the tools,the fields

274、of application of these tools,and recording,for example,the prompts when using AI tools such as ChatGPT.Exceptions can be made to the rules outlined,such as prohibiting tools in specific learning or assessment situations,which will be communicated to the students in advance.Guidance for Lecturers 38

275、 5.Outlook Outlook 39 5.Outlook As AI becomes more prevalent in daily life,it will be impossible to ignore that students will use AI-based tools to succeed in higher education(Jacobsen,2023).Furthermore,they should use AI tools to be productive and acquire important digital skills.Although higher ed

276、ucation is not the fastest-moving field,it is forced to innovate its inherent educational structures as technological improve-ments are rapid and vast.Consequently,there is no doubt that teaching and learning will change drastically.The media coverage has mainly focused on“cheating”and how universit

277、ies can put pol-icies and procedures in place to manage“the AI problem.”We argue for a more positive view on technological advancements such as generative AI.The debate and the innovation should focus on the potential benefits of generative AI,such as improved learning,teaching,and the creation of e

278、qual opportunities for different groups of students.On the side of technologies and generative AI systems,it is straightforward to assume further rapid developments.Ever more potent GPT-x models by OpenAI,comparable models by other vendors,more multimodel input and output to generative AI models,dif

279、ferent interfaces to such models be-yond conversational agents,and integration with other classes of IT systems are clear paths ahead.OpenAI,Microsoft,and many others in academia and industry are working in these directions.Hence,what we see with the current version of ChatGPT is likely only the fir

280、st small step on a big road towards increasingly powerful generative AI tools in higher education and beyond.Conversational agents are an essential resource that lecturers and students can and should use in teaching and learning already today.However,innovating teaching,assessing,and learning is onl

281、y one field in a more complex higher educational landscape.As the technology develops,the potential of generative AI goes beyond the“teaching-learning trifecta”between students,lecturers,and tech-nical tools.It will transform the entire student lifecycle,including admissions,enrollment,career ser-vi

282、ces,and further areas of higher education management,as exemplified in Figure 12.The easy access and fast dissemination of ChatGPT,along with the associated challenges in learn-ing,assessing,and teaching,have shown how quickly traditional patterns can be disrupted by tech-nology.As we deliberately f

283、ocus on the impact of ChatGPT and related tools on teaching and learn-ing in this whitepaper,we consciously exclude other parts of higher education.However,we believe that we should seize the opportunity presented by technological developments in AI to rethink the world of higher education as a whol

284、e.In this context,we should be aware that ChatGPT may change the expectations of future AI technologies,especially conversational agents,whether in terms of interaction or information quality.On a macro level,universities must watch out for potential social inequality when tools such as ChatGPT are

285、only available for a service fee.On the other hand,we should not expect that every IT-based service provided via the Internet is for free.From a societal perspective,it is crucial to ensure that all students can access the same tools and resources to complete their education suc-cessfully.However,it

286、 is uncertain how long a free version of ChatGPT will be available.In addition to the free version of ChatGPT,OpenAI released the premium version ChatGPT Plus for 20 USD per month(OpenAI,2023c).It promises improved availability,unrestricted usage,and access to GPT-4.As this new AI model boasts advan

287、tages over its predecessor,GPT3.5,we conclude that users who can afford the premium version will enjoy advantages in utilizing ChatGPT.Thus,it is questionable whether this premium versions availability is already compromising the principle of equal opportunities for all students.Outlook 40 Figure 12

288、.Potential application fields of generative AI in higher education To ensure equal access to education,universities might consider providing fee-based tools,such as the premium version of ChatGPT,free of charge or at a significantly reduced rate for students.This may require substantial financial fu

289、nds for the campus license.With ChatGPT Plus at 20 USD per month before academic discounts,the monthly fees for a university of 10.000 students might add up to 200.000 USD,summing up yearly to over 2,4 million USD.With a growing number of necessary cloud-based tools requiring license fees rather tha

290、n open-source software installed on-premise in the Universitys data centers,ChatGPT adds to an ongoing discussion about funding digital trans-formation in higher education.Another key question beyond the individual course is which competencies students should develop in higher education.Domain-speci

291、fic skills always have been important and remain important.From our point of view,transferable skills such as cognitive,metacognitive,and socio-emotional skills as well as digital skills are becoming increasingly important.These skills include logical reasoning,structured and critical thinking,probl

292、em-solving,collaboration,and emotion regulation.Not least,students should have the ability to develop and use AI-based tools responsibly.Such skills will help students succeed in an interconnected and fast-paced digital world.Students will succeed in an interconnected and fast-paced digital world.Ao

293、un(2018)describes the necessary mindset as“robot-proof,“focusing on unique cognitive human capabilities like critical thinking,systems thinking,entre-preneurship,and cultural agility.As generative AI continues to advance,it is crucial to explore how it impacts the development of these skills in high

294、er education.With AIs ability to generate and provide information,there is a risk that students may become passive recipients of information rather than active thinkers.For instance,to ensure that critical thinking remains an essential education component,higher education must actively develop strat

295、egies to foster critical thinking in their classes.Thus,expanding the current curriculum by including lectures on how AI can be used to support critical thinking is crucial.Higher Potential application fields of generative AI in higher educationGimpel,H.,Hall,K.,Decker,S.,Eymann,T.,Lmmermann,L.,Mdch

296、e,A.,Rglinger,R.,Ruiner,C.,Schoch,M.,Schoop,M.,Urbach,N.,Vandirk,S.(2023).Unlocking the Power of Generative AI Models and Systems such as GPT-4 and ChatGPT for Higher Education:A Guide for Students and Lecturers.University of Hohenheim.AlumnirelationsSetting policies and regulationsInfrastructure pr

297、ovisionPortfolio ManagementOperationsManagementQuality assuranceLecturerStudentsTechnical tools and media-Supporting the lecturer-Supporting the students-Mediating the interactionGenerative AIManagement and operation of highereducation(select processes)Teaching and Learning TrifectaStudent lifecycle

298、Sequence and portfolio of courses,theses,and learning achievements throughout a course of studiesApplication&admissionCertificate awardingCareer supportEffectiveness,efficiency,and quality of educationFocus of this whitepaperOutlook 41 education must incorporate tasks and activities that promote cri

299、tical thinking and develop assess-ment methods that measure the students development of critical thinking skills in the context of generative AI.Ultimately,the challenge is to ensure that students continue to develop the skills they need to make informed decisions and solve complex problems in a wor

300、ld where AI is increasingly prevalent,as these skills are highly appreciated in the business context and highly needed in society.The focus of this whitepaper is on students and lecturers.Nevertheless,we see a clear need for action beyond the level of individual students and lecturers.Universities s

301、hould encourage broad,multi-perspective dialogue among many stakeholders in higher education.They should include all faculties and disciplines since different fields have different traditions,requirements,and opportunities,which should be reflected in any universitys approach to generative AI.They s

302、hould involve their own experts from information systems,computer science,data science,and related disciplines,who have been researching IT-based innovations and digital transformation for years and decades and,in many cases,also research generative AI.They can contribute to the knowledge of the tec

303、hnologies and the transformation process and have first-hand experience in teaching at their university.They should involve their career centers and representatives from industry and society to inform the dialogue with perspectives on the required educational profiles and skills.They should involve

304、students who contribute their perspectives on learning objectives,for-mats,and study conditions.They should involve experts on university didactics,who bring important perspectives on learning objectives,teaching-learning formats,assessments,and the like.They should involve legal experts to examine

305、the legal possibilities offered by the current legislation and university regulations and the changes required to make the desired use of generative AI tools possible and legally sound.They should involve the universitys divisions that administrate study and teaching.These are important to the proce

306、sses that should deliver fair,efficient,and high-quality teaching.They should involve the universitys IT department,which can consult on access,infrastruc-ture,licenses,IT security,and the like.Together with all these stakeholders,universities should engage in a dialogue on how to promote and levera

307、ge ChatGPT in the short term and other generative AI tools in the medium term.The results of the dialogue should lead to multi-perspective insights that result in regulations,guidelines,handouts,tutorials,and implementations.If appropriate,it may be helpful to talk to external experts,exchange exper

308、iences with other universities,speak to the responsible supervisory authority and politics,and demand the necessary resources for excellent university education.To summarize,integrating generative AI tools such as ChatGPT in higher education requires a sig-nificant educational transformation that ca

309、nnot be achieved overnight.While there are plenty of ideas and discussions on managing and leveraging such tools,lecturers should first learn how ChatGPT and comparable tools operate and modify their teaching methods,contents,and processes accord-ingly.Additionally,changes to examination formats can

310、not happen immediately but rather require careful development and adaptation following examination regulations.As such,integrating ChatGPT into higher education will require patience and careful planning to ensure its successful implementation.Students should not wait for the university-level discou

311、rse to end and for the lectur-ers to have adapted.Instead,we suggest that students actively engage with generative AI.If they did not use ChatGPT yet,they should get a free account and gain first-hand experience with the possibilities and limits.They should reflect on their learning goals,methods,an

312、d processes and Outlook 42 engage with other stakeholders in higher education to shape the dialogue on AI-powered higher education.As we call lecturers and students to action regarding ChatGPT and provide recommendations for them,it is important to stress that their use of ChatGPT should comply with

313、 legislation,university regulations,good scientific practices,and OpenAIs terms and conditions.If this is the case,this whitepaper hopefully provides food for thought regarding using generative AI,large language models such as GPT-4,and tools such as ChatGPT in higher education.References 43 Referen

314、ces Al Muid,M.A.,Reza,M.M.,Kalim,R.B.,Ahmed,N.,Habib,M.T.,&Rahman,M.S.(2021).EduBot:An unsupervised domain-specific chatbot for educational institutions.In T.Masrour,I.El Hassani,&A.Cherrafi(Eds.),Lecture Notes in Networks and Systems.Artificial Intelligence and Industrial Applications(pp.166174).Ch

315、am:Springer International Publishing.Aoun,J.(2018).Robot-proof:Higher education in the age of artificial intelligence(1st paperback ed.).Cambridge,MA,London:The MIT Press.Arif,T.B.,Munaf,U.,&Ul-Haque,I.(2023).The future of medical education and research:Is ChatGPT a blessing or blight in disguise?Me

316、dical education online,28(1).arXiv.(2023).Policy for authors use of generative AI language tools.Retrieved March 12,2023,from https:/info.arxiv.org/help/moderation/Atlas,S.(2023).ChatGPT for higher education and professional development:A guide to conversa-tional AI.Retrieved March 12,2023,from http

317、s:/digitalcommons.uri.edu/cba_facpubs/548/Azaria,A.(2022).ChatGPT usage and limitations.Retrieved March 15,2023,from https:/hal.sci-ence/hal-03913837 Benson,R.(2023).The generative-AI eruption.Retrieved March 12,2023,from https:/rodben- Models are Few-Shot Learners.Retrieved March 15,2023,from https

318、:/arxiv.org/abs/2005.14165 Dahmen,J.,Kayaalp,M.E.,Ollivier,M.,Pareek,A.,Hirschmann,M.T.,Karlsson,J.,&Winkler,P.W.(2023).Artificial intelligence bot ChatGPT in medical research:the potential game changeras a double-edged sword.Knee surgery,sports traumatology,arthroscopy:official journal of theESSKA,

319、Decker,S.(2022).ChatGPTan arms race between large language models and knowledge graphs?Retrieved March 12,2023,from https:/ Team.(2023).Introducing Duolingo Max,a learning experience powered by GPT-4.Re-trieved March 15,2023,from https:/ what if ChatGPT wrote it?”Multidisciplinary perspectives on op

320、portunities,challenges and implications of generative conversational AI for research,practice and policy.International Jour-nal of Information Management,71,102642.Earl,L.M.(2012).Earl,Lorna M.Assessment as learning:Using classroom assessment to maximize student learning.:Corwin Press.Elshan,E.,&Ebe

321、l,P.(2020).Lets team up:Designing conversational agents as teammates.ICIS 2020 Proceedings.2.Retrieved March 15,2023,from https:/aisel.aisnet.org/icis2020/digi-tal_learning_env/digital_learning_env/2 Feidakis,M.,Kasnesis,P.,Giatraki,E.,Giannousis,C.,Patrikakis,C.,&Monachelis,P.(2019).Building pedago

322、gical conversational agents,affectively correct.In Proceedings of the 11th Inter-national Conference on Computer Supported Education(pp.100107).SCITEPRESS-Science References 44 and Technology Publications.Forschung&Lehre.(2023).Renommierter britischer Verlag regelt Umgang mit ChatGPT.Retrieved March

323、 16,2023,from https:/www.forschung-und-lehre.de/zeitfragen/renommierter-britischer-verlag-regelt-umgang-mit-chatgpt-5473?Fried,I.(2023).ChatGPT is the talk of Davos.Retrieved March 12,2023,from https:/www.ax- Frlich-Steffen,S.(2023,February).In Zeiten von ChatGPT“Haus-und Abschlussarbeiten als Pr-fu

324、ngsformat einsetzen?Zentrum fr Hochschullehre(ZHL)der Universitt Bayreuth.Zentrum fr Hochschullehre(ZHL)der Universitt Bayreuth,Gnewuch,U.,Morana,S.,&Mdche,A.(2017).Towards designing cooperative and social conver-sational agents for customer service.ICIS 2017 Proceedings.1.Retrieved from https:/aise

325、l.aisnet.org/icis2017/HCI/Presentations/1 Hauck-Thum,U.,Kreuter,F.,Kuhm,J.,Weller,J.,Schtze,H.,&Schmidt,A.(2023).Was verndert sich fr uns durch ChatGPT?:Wie werden KI-Sprachmodelle Lehren und Lernen verndern.Re-trieved March 12,2023,from https:/ Herman,D.(2022).The End of High-School English.Retriev

326、ed March 12,2023,from https:/ are you,chatbot?Evaluating chatbots in educational settings Results of a Literature Review.In Gesellschaft fr Informatik e.V.(Ed.),DELFI 2019(pp.259270).Bonn.Hu,K.(2023).ChatGPT sets record for fastest-growing user base-analyst note.Retrieved March 12,2023,from https:/

327、ChatGPT will change higher ed for the bett.Retrieved March 12,2023,from https:/fee.org/articles/chatgpt-will-change-higher-ed-for-the-better/Khalil,M.&Er,E.(2023).Will ChatGPT get you caught?Rethinking of plagiarism detection.Re-trieved March 15,2023,from https:/arxiv.org/abs/2302.04335 Koenders,L.&

328、Prins,F.(2023).ChatGPT in education:can you still use take-home exams and essays?Retrieved March 12,2023,from https:/www.uu.nl/en/education/educational-develop-ment-training/knowledge-dossier/the-influence-of-chatgpt-on-assessment-can-you-still-use-take-home-exams-and-essays Malone,T.(2023).MIT GenA

329、I Summit.Retrieved March 14,2023,from https:/web.mit.edu/webcast/mitgenaisummit/s23/McCormack,G.(2023).Chat GPT Is here!5 alternative ways to assess your class!Retrieved March 12,2023,from https:/.au/chat-gpt-is-here-5-alternative-ways-to-as-sess-your-class/McTear,M.,Callejas,Z.,&Griol,D.(2016).The

330、Conversational Interface:Talking to Smart Devices(1st ed.2016).Cham:Springer International Publishing.Mollick,E.R.,&Mollick,L.(2022).New Modes of Learning Enabled by AI Chatbots:Three Methods and Assignments.SSRN Electronic Journal,OpenAI.(2023a).GPT-4.Retrieved from https:/ References 45 OpenAI.(20

331、23b).GPT-4 is OpenAIs most advanced system,producing safer and more useful re-sponses.Retrieved from https:/ OpenAI.(2023c).Introducing ChatGPT.Retrieved March 12,2023,from https:/ Rademacher,M.(2023).Warum ChatGPT nicht das Ende des akademischen Schreibens bedeutet.Retrieved March 12,2023,from http

332、s:/digiethics.org/2023/01/03/warum-chatgpt-nicht-das-ende-des-akademischen-schreibens-bedeutet/Rudolph,J.(2014).Massive Open Online Courses(MOOCs)as a disruptive innovation in higher education?Retrieved from https:/www.academia.edu/15137335 Rudolph,J.,Tan,S.,&Tan,S.(2023).ChatGPT:Bullshit spewer or

333、the end of traditional assess-ments in higher education?Journal of Applied Learning&Teaching,6(1).Retrieved March 12,2022,from https:/journals.sfu.ca/jalt/index.php/jalt/article/view/689/Salz,M.(2023).Microsoft-Plne fr ChatGPT:KI soll auch zu Word,Excel,PowerPoint und Co.kommen.Retrieved March 12,2023,from https:/www.chip.de/news/Microsoft-bringt-ChatGPT-auch-zu-Word-Excel-und-Co._184651857 Schill