Generative AI
Open accessThe DOI matches its Crossref record; 145 of the 146 references it cites were found in a registry or an index; 20 of the 45 sources its claims cite were checked for support.
Abstract
Tom Freston is credited with saying “Innovation is taking two things that exist and putting them together in a new way”. For a long time in history, it has been the prevailing assumption that artistic, creative tasks such as writing poems, creating software, designing fashion, and composing songs could only be performed by humans. This assumption has changed drastically with recent advances in artificial intelligence (AI) that can generate new content in ways that cannot be distinguished anymore from human craftsmanship.
Overview
CoChat Summary
TL;DR
This Catchword article conceptualizes generative AI as a model-, system-, and application-level entity within socio-technical systems, reviews its limitations, and proposes a BISE research agenda. It is a conceptual/agenda-setting piece rather than an empirical study.
Key findings
- Generative AI is defined as computational techniques that generate seemingly new, meaningful content such as text, images, or audio from training data; the paper distinguishes generative from discriminative modeling and organizes generative AI into model-, system-, and application-level views.
- The paper cites industry estimates that generative AI could raise global GDP by 7% and replace 300 million knowledge-worker jobs (Goldman Sachs 2023), and notes that training GPT-3 was estimated to produce 552 t CO2.
- It identifies four salient limitations of current generative AI: incorrect outputs/hallucination, bias and fairness, copyright violation, and environmental concerns.
- It proposes BISE research directions across departments including business process management, decision analytics and data science, digital business management and leadership, economics of information systems, enterprise modeling and engineering, human–computer interaction and social computing, and information systems engineering and technology.
Methods
- Conceptual Catchword article; no empirical data, sample, or formal analysis are reported.
- The paper synthesizes prior literature and examples to develop a conceptualization of generative AI, discuss limitations, and outline a research agenda for the BISE community.
- It organizes implications by BISE journal departments and presents exemplary research questions in Table 2.
Limitations
- The paper states four limitations of current generative AI: incorrect outputs (including hallucination), bias and fairness, copyright violation, and environmental concerns; it notes these are technical/model-level limitations likely to persist with system- and application-level implications.
- The paper also notes that outputs are typically not easily verifiable, that state-of-the-art models are black-box, and that closed-source commercial systems prohibit further tuning and re-training.
- As a conceptual Catchword article, it does not provide empirical validation, a systematic review protocol, or tested design principles; the research agenda is proposed rather than evaluated (evident from the text).
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
15 Valid · 1 Not valid · 19 Not checked · 31 Own results and interpretations
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Deep neural networks are particularly well suited for the purpose of data generation, especially as deep neural networks can be designed using different architectures to model different data types (Janiesch et al. 2021; Kraus et al. 2020), for example, sequential data such as human language or spatial data such as images.”
What the sources say
- Partly supportedcites Janiesch et al. 2021 sourceExperimental — accuracy not yet measured
Model building (p. 5) says different deep-learning architectures are better suited to particular data types, including time series and images. Model assessment (p. 6) describes GANs generating new data samples and content. The paper does not establish the broader claim that deep neural networks are particularly well suited to data generation because of their architectural variety. (LLM, conf=0.82)
- Partly supportedcites Kraus et al. 2020 sourceExperimental — accuracy not yet measured
Section 2.4 (p. 10) says different network architectures target different data structures, including recurrent networks for sequential data and convolutional networks for vision tasks. Section 5.3 (p. 25) describes a generative network that produces sample inputs, but the paper does not say neural networks are particularly well suited to data generation because of their ability to model language and images; it discusses those architectures chiefly in predictive applications. (LLM, conf=0.85)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“For both open-source and closed-source models, unexpected deterioration of model performance over time highlights the need for continuous model monitoring (Chen et al. 2023).”
What the sources say
- Partly supportedcites Chen et al. 2023 sourceExperimental — accuracy not yet measured
The Conclusions and Future Work section reports that GPT-3.5 and GPT-4 worsened on some tasks and recommends continuous monitoring. The paper studies these LLM services, but does not establish deterioration over time for open-source models. Page numbers are unavailable in our copy. (LLM, conf=0.89)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“or other personality traits as personalized marketing content is more effective than a one-content-fits-all approach (Matz et al. 2023).”
What the sources say
- Supportedcites Matz et al. 2023 source
The Abstract states that matching persuasive content to a recipient’s psychological profile enhances its effectiveness compared with non-personalized communication. The Results report that participants significantly preferred personalized over generic ads for a weekend getaway, though the difference for sneakers was not statistically significant. (LLM, conf=0.90)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“Recent studies expose the harmful biases embedded in multimodal generative AI models such as CLIP (contrastive language-image pretraining; Wolfe et al. 2022) and the CLIP-filtered LAION dataset (Birhane et al. 2021), which are core components of generative AI models (e.g., Dall-E 2 or Stable Diffusion).”
What the sources say
- Supportedcites Wolfe et al. 2022 source
The Introduction reports biased associations in CLIP, and the Discussion describes the potential for biases learned by CLIP to spread into other models. The Discussion also says a larger version of the CLIP architecture was used to train DALL-E 2. The separate LAION claim is attached to Birhane et al., not Wolfe et al. (LLM, conf=0.90)
- Supportedcites Birhane et al. 2021 source
The Abstract (p. 1) identifies LAION-400M as a CLIP-filtered dataset and reports that it contains malign stereotypes, racist slurs, and other problematic content. Section 2 (p. 3) also describes offensive material found in the dataset. This supports the claim attributed to Birhane et al. about harmful content in the CLIP-filtered LAION dataset; the separate citation addresses CLIP itself. (LLM, conf=0.92)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“Yet, due to the black-box nature of state-of-the-art AI models (Rai 2020), the usage of such systems critically hinges on users' trust in reliable outputs.”
What the sources say
- Supportedcites Rai 2020 source
The section on inherently interpretable versus black-box deep-learning models (p. 2) says deep-learning models are inherently uninterpretable to human users. The Introduction (p. 1) says users need to trust AI systems and that information about prediction accuracy can help establish that trust; the discussion of XAI (p. 2) links trust in predictions to whether users act on them. (LLM, conf=0.92)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
22 more claims · 2 with their cited sources checked
Create a free account to see all 66 claims and every source verdict.
- Probably not a substantial claim. 25
- It may state two separate claims as one. 22
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The cited source doesn’t support the claim as written.
In the citing paper
“These quotes indicate that as teachers and students widely use GenAI, their thought processes may gradually be influenced by it, weakening critical thinking and, thus, the emergence of homogeneous behaviors amid the fact that GenAI is not creating any new content [44].”
From Understanding the Practices, Perceptions, and (Dis)Trust of Generative AI among Instructors: A Mixed-methods Study in the U.S. Higher Education (2025) · cites this paper as [44]
Checked by a reasoning model — may be wrong.
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“Known as 'hallucinations' , generative AI's ability to produce seemingly logical but nonetheless incorrect statements is already well-documented (Feuerriegel et al. 2024).”
From Summarise, elaborate, try again: exploring the effect of feedback literacy on AI-enhanced essay writing (2026) · cites this paper as Feuerriegel et al. 2024
Citations
Cites
Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, L., & Polosukhin, I. (2017) · arXiv · doi:10.48550/arxiv.1706.03762 · Cited by 120,404
Exploiting Generative AI to Scale up Intelligent Tutoring Systems
Jakubův, J., Chvalovský, K., Goertzel, Z., Kaliszyk, C., Olšák, M., Piotrowski, B., Schulz, S., Suda, M., & Urban, J. (2023) · DROPS (Schloss Dagstuhl – Leibniz Center for Informatics) · doi:10.4230/lipics.itp.2023.19 · Cited by 76,658
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2018) · arXiv · doi:10.48550/arxiv.1810.04805 · Cited by 73,173
Language Models are Few-Shot Learners
Brown, T. B., Mann, B., Ryder, N., Subbiah, M., Kaplan, J., Dhariwal, P., Neelakantan, A., Shyam, P., Sastry, G., Askell, A., et al. (2020) · arXiv · doi:10.48550/arxiv.2005.14165 · Cited by 23,372 · cited 3×
“Prominent examples of massive LLMs are BERT (Devlin et al. 2018) and GPT-3 (Brown et al. 2020) with (cid”
Book Review: Pattern Recognition and Machine Learning
Nasrabadi, N. M. (2007) · Journal of Electronic Imaging · doi:10.1117/1.2819119 · Cited by 22,066
High-Resolution Image Synthesis with Latent Diffusion Models
Rombach, R., Blattmann, A., Lorenz, D., Esser, P., & Ommer, B. (2022) · 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr52688.2022.01042 · Cited by 20,302 · cited 7×
“Recent studies expose the harmful biases embedded in multimodal generative AI models such as CLIP (contrastive language-image pre-training; Wolfe et al. 2022) and the CLIP-filtered LAION dataset (Birhane et al. 2021), which are core components of generative AI models (e.g., Dall-E 2 or Stable Diffusion).”
InfluentialAuto-Encoding Variational Bayes
Kingma, D. P., & Welling, M. (2013) · arXiv · doi:10.48550/arxiv.1312.6114 · Cited by 20,067 · cited 2×
“VAEs differ from traditional autoencoders by using a probabilistic approach to the encoding and decoding process, which enables them to capture the underlying structure and variation in the data and generate new data samples from the learned latent space (Kingma and Welling 2013).”
Labaca-Castro, R. (2023) · Machine Learning under Malware Attack · doi:10.1007/978-3-658-40442-0_9 · Cited by 19,272
Sequence to Sequence Learning with Neural Networks
Sutskever, I., Vinyals, O., & Le, Q. V. (2014) · arXiv · doi:10.48550/arxiv.1409.3215 · Cited by 17,766 · cited 1×
“…allow for more complex data-driven prompt engineering, such as tuning prompts via reinforcement learning (Liu et al. 2023). seq2seq The term sequence-to-sequence (seq2seq) refers to machine learning approaches where an input sequence is mapped onto an output sequence (Sutskever et al. 2014).”
Kaugeranna, R., Kaugeranna, E., & Aion, (. S. 4. 6. (2023) · DROPS (Schloss Dagstuhl – Leibniz Center for Informatics) · doi:10.4230/lipics.giscience.2023.43 · Cited by 13,690
Cited by
Agentic AI: Autonomous intelligence for complex goals–a comprehensive survey
Acharya, D. B., Kuppan, K., & Divya, B. (2025) · IEEE Access · doi:10.1109/access.2025.3532853 · Cited by 717
“The first is how Generative AI [8] works.”
Bengesi, S., El-Sayed, H., Sarker, M. K., Houkpati, Y., Irungu, J., & Oladunni, T. (2024) · IEEE Access · doi:10.1109/access.2024.3397775 · Cited by 362
“Despite numerous recent studies to address the surge of GAI [31], [31], [34], [34], [35], [35], [36], [36], [37], [38], [39], [40], [41], [44], [45], [46], discussing challenges, tasks, and models, there still needs to be more thorough exploration into the theoretical and mathematical foundations…”
BackgroundEthical challenges and solutions of generative AI: an interdisciplinary perspective
Al-kfairy, M., Mustafa, D., Kshetri, N., Insiew, M., & Alfandi, O. (2024) · Informatics · doi:10.3390/informatics11030058 · Cited by 328
“…from deep learning models like generative adversarial networks (GANs) to recent breakthroughs in language models and image generators, has demonstrated unprecedented capabilities in creating text, images, music, and even synthetic data that closely mimic human-like creativity and understanding [2].”
BackgroundAl Naqbi, H., Bahroun, Z., & Ahmed, V. (2024) · Sustainability · doi:10.3390/su16031166 · Cited by 302
“According to Feuerriegel et al. [7], the term “generative” in the context of AI refers to an AI system’s capacity to generate new material autonomously, without human involvement.”
BackgroundStrzelecki, A., & ElArabawy, S. (2024) · British Journal of Educational Technology · doi:10.1111/bjet.13425 · Cited by 260
Metadata
- Authors
- Feuerriegel, S., Hartmann, J., Janiesch, C., & Zschech, P.
- Affiliations
- Munich Center for Machine Learning (DE)
- Technical University of Munich (DE)
- TU Dortmund University (DE)
- Friedrich-Alexander-Universität Erlangen-Nürnberg (DE)
- Corresponding author
- Feuerriegel, Stefan
- Venue
- Business & Information Systems Engineering
- Published
- 2023-09-12
- Volume
- 66
- Issue
- 1
- Pages
- 111-126
- ISSN
- 1867-0202 (ISSN-L)
- 2363-7005
- OpenAlex
- W4386693657
- Cited by
- 2,267
- References
- 123
- Citations by year
- 2026: 908 · 2025: 963 · 2024: 380 · 2023: 15
- FWCI
- 161.28
- Type
- journal article
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- cc-by
- creativecommons.org/licenses/by/4.0 (tdm)
- creativecommons.org/licenses/by/4.0 (vor)
- Locations
- Business & Information Systems Engineering (publishedVersion, cc-by)
- SSRN Electronic Journal (submittedVersion)
- arXiv (Cornell University) (submittedVersion)
- Econstor (Econstor) (submittedVersion, cc-by)
- Open access LMU (Ludwid Maxmilian's Universitat Munchen) (submittedVersion, cc-by)
- RePEc: Research Papers in Economics (submittedVersion)
- Journal of the Association for Information Systems (submittedVersion)
- mediaTUM – the media and publications repository of the Technical University Munich (Technical University Munich) (submittedVersion)
- mediaTUM (Technical University of Munich) (submittedVersion, other-oa)
- arXiv (Cornell University)
- Funders
- Ludwig-Maximilians-Universität München
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Artificial Intelligence Applications
- Social Sciences › Business, Management and Accounting › Management Information Systems › Information Systems Theories and Implementation
- Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
- Keywords
- generative AI, socio-technical systems, AI-generated content, DALL-E 2, GPT-4, Copilot, research agenda
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-01, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-01 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Generative AI. https://app.cochat.ai/papers/10.1007/s12599-023-00834-7
- The paper
- Feuerriegel, S., Hartmann, J., Janiesch, C., & Zschech, P. (2024). Generative AI. Business & Information Systems Engineering. https://doi.org/10.1007/s12599-023-00834-7
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.