On the Current State of Generative Artificial Intelligence:

A Conceptual Model of Potentials and Challenges

Christian Daase

, Christian Haertel

, Abdulrahman Nahhas

, Alexander Zeier

Achim Ramesohl

and Klaus Turowski

Institute of Technical and Business Information Systems, Otto-von-Guericke University, Magdeburg, Germany

Google Cloud, Germany

Keywords: Generative AI, Artificial Intelligence, Conceptual Model, Systematic Literature Review.

Abstract: Generative artificial intelligence (GenAI) is one of the most promising recent advances in digital technology.

However, research often focuses on specific application scenarios, case studies and experiments. Overarching

and comprehensive studies that consider potentials and challenges for the entire field of GenAI across domains

are rather scarce. In this paper, the four domains of text, audio, image and code generation are examined by

means of a systematic literature review. Opportunities for industry and society are discussed, with the aim of

providing a conceptual model that enables a quick assessment of the current state-of-the-art and identifies

applications for GenAI that are either not yet sufficiently researched and therefore invite further exploratory

investigations, or are well researched and therefore represent recognized yet less experimental fields.

1 INTRODUCTION

The exponential rise in generated data, available

computational power, and more advanced statistical

techniques (Cobb 2023; Haertel et al. 2023;

Vartiainen and Tedre 2023) have significantly

increased the application of advanced machine

learning (ML) (e.g., deep neural networks (DNN)) in

various areas of the everyday life. Especially the

fields of image analytics, video analytics (Daase et al.

2023), and language models have gained attraction in

this regard (Liu et al. 2017). Recently, generative AI

(GenAI) has gained severe prominence. Generative

models pursue the objective of approximating

training data sets’ statistical distributions to generate

new data points (François-Lavet et al. 2018). In this

context, GenAI shifts the focus of AI from analytical

expert systems as decision-makers to creators of new

content that function similarly to creative human

collaborators (Turchi et al. 2023). From a conceptual

point of view, researchers see GenAI’s potential

between factual knowledge and creative thinking for

partially rule-based activities (Kanbach et al. 2023).

https://orcid.org/0000-0003-4662-7055

https://orcid.org/0009-0001-4904-5643

https://orcid.org/0000-0002-1019-3569

https://orcid.org/0000-0002-4388-8914

As integral technologies, GenAI includes inter alia

natural language processing (NLP), image

processing, and computer vision (Lv 2023), resulting

in the four main output categories examined in this

paper: text, image, audio, and programming code. To

grasp the future impact of GenAI on the business

sector, it is worth noting that Goldman Sachs

estimates that AI in general could have an impact on

up to 300 million jobs globally, while the major

consulting company Accenture estimates that 40

percent of human workload will be affected by large

language models (LLMs), one of the most prominent

examples being GPT-4 (i.e., Generative Pre-Trained

Transformer) and the chatbot ChatGPT that builds on

it (Kanbach et al. 2023).

The present paper examines explicit use cases for

GenAI from the four stated categories to provide a

comprehensive overview of potentials and challenges

in this area. Based on a systematic literature review

(SLR), applications from different subcategories are

evaluated and assessed, assuming that the future of

GenAI will focus on exploring new application

scenarios and integration with other technologies to

Daase, C., Haertel, C., Nahhas, A., Zeier, A., Ramesohl, A. and Turowski, K.

On the Current State of Generative Artiﬁcial Intelligence: A Conceptual Model of Potentials and Challenges.

DOI: 10.5220/0012707500003690

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 26th International Conference on Enterprise Information Systems (ICEIS 2024) - Volume 1, pages 845-856

ISBN: 978-989-758-692-7; ISSN: 2184-4992

845

increase its impact (Lv 2023). Furthermore,

anticipated issues of GenAI such as bias,

misinterpretation, societal impact, shifting

responsibilities and roles of people in the workplace

(Jiang et al. 2023; Kanbach et al. 2023; Vartiainen

and Tedre 2023) are considered. As a result, a

conceptual model of the fields of application of

GenAI is presented to conclude the current state of

research as it is reasonable based on the literature

reviewed. The contribution of this paper is intended

to help academics and practitioners to assess

opportunities and obstacles in the highly dynamic and

equally promising field of GenAI. In short, the

research question (RQ) posed in this paper is as

follows:

RQ: Which potentials and challenges of GenAI in

industry and society can be identified in recent

scientific studies and what is their expectable impact?

Following this introduction, the methodology of the

SLR is described in Section 2. In Section 3, the

identified fields of GenAI (i.e., text, audio, image,

and code) are presented, including their very specific

subcategories, industries, and use cases. Section 4

first assesses the overarching challenges and

classifies the GenAI categories in the conceptual

model as either well researched, moderately

researched, or currently under-researched or less

promising. The paper is concluded with an outlook on

further research and limitations in Section 5.

2 METHODOLOGY

A systematic literature review (SLR) was conducted

to answer the stated RQ. As the overall goal is to

provide a comprehensive overview of the potentials

and challenges of GenAI, the abstract and citation

database Scopus was chosen as the primary literature

source, since it refers to a large number of

interdisciplinary articles from a variety of full text

databases. For the automatic search for scientific

articles in the first phase of the review, the search

query was set to the inclusive phrase “generative AI”

AND <domain>. As described in the introduction, the

primary application domains of GenAI in the context

of this paper are text, audio, image and code. In

addition, the time frame was set to all years including

2023. A total of 228 articles were retrieved from the

individual domains, 63 of which turned out to be

duplicates. This left 165 articles after this phase.

In the second review stage, the abstracts were read

to identify the overarching areas of application of

GenAI in each article, whether the articles were

written in English and if the quality of the publication

could be considered sufficient (i.e. peer-reviewed in

a reputable research publication). In this phase, 108

articles were rejected, while 57 remained.

In the third and final review phase, the remaining

articles were read as a whole to extract the granular

information on specific GenAI use cases, potentials,

challenges, monetary value, and societal

implications. Therefore, publications that did not

provide sufficient detail on the implementation or

operation of such systems were rejected. The

industrial or societal value of GenAI's intervention

needed to be made clear without relying solely on

personal opinion or experience. Articles without

sufficient scientific backing were also rejected,

resulting in 12 articles being omitted and 45

remaining to form the final literature base. Figure 1

depicts the completed literature search process.

Figure 1: SLR search process.

ICEIS 2024 - 26th International Conference on Enterprise Information Systems

846

3 FIELDS OF GENERATIVE AI

GenAI systems are capable of generating novel and

creative content based on training data. The produced

outputs can be categorized as text, audio, or image

and video (Mannuru et al. 2023), each offering unique

potentials but also posing significant challenges.

Therefore, in the following, these aspects are

discussed based on the retrieved articles from the

SLR. The analysis is divided based on the above-

mentioned categorization on GenAI output types.

However, code is considered as a separate subsection

since its applications are fundamentally different

compared to the text category. Additionally, the use

cases of image and video are subsumed under the

same category.

3.1 Text

Almost half of the articles from the final literature

base discuss GenAI use cases for text output.

Accordingly, the opportunities are manifold. From

providing quick access to a bandwidth of information

for educational purposes to supporting and partially

replacing human involvement in various business

processes, the capabilities of GenAI seem sheer

overwhelming. However, on the same note, numerous

issues still have to be considered with GenAI usage,

leading to partially low trust levels in the technology.

Hence, complete detachment of human supervision is

not recommended for the foreseeable future.

Potentials and challenges of text-based application

scenarios in public services, industry, sustainability,

and art and media are outlined in detail below.

3.1.1 Public Services

The examined literature features numerous use cases

for text-based GenAI for the public service sector

(e.g., education, healthcare). K. Ali et al. (2023) and

Panthier and Gatinel (2023) show the ability of

ChatGPT to successfully pass exams in the healthcare

domain. While both studies point out the

predominantly high quality in the model’s answers,

ChatGPT was only able to process tasks based on text

and could not answer questions containing figurative

elements (K. Ali et al. 2023; Panthier and Gatinel

2023). Based on these findings, it can be concluded

that GenAI will have a significant impact on

assessment practices in higher education. In home

examinations, ChatGPT excelled in terms of language

and grammar, precision, completeness as well as

partially in creativity (Farazouli et al. 2023). Still, the

OpenAI chatbot still displays weaknesses in

argumentation strategy, use of references and

relevance to the topic of the course. Since these flaws

could also be found in the answers of humans,

teachers tend to more critical in grading texts of

students (Farazouli et al. 2023).

Apart from that, the application possibilities of

GenAI is assisting learners and teachers are vast.

Content generation is mentioned (Yan et al. 2023) as

a potential to create personified learning material

depending on the respective pupil’s skill level,

leading to improved learning outcomes (Jauhiainen

and Guerra 2023). GenAI can be utilized as additional

(virtual) teaching support (Ooi et al. 2023; Yan et al.

2023) to achieve learning objectives such as acquiring

proficiency in writing in a new language (Hwang and

Nurtantyana 2022). Furthermore, these models can

provide (personalized) feedback (Ooi et al. 2023; Yan

et al. 2023) to derive suggestions for enhancement.

For example, Siiman et al. (2023) use GenAI for

analysis and evaluation of conversations performed in

the context of problem-solving. Additionally, LLMs

have been utilized to support research and academic

writing (Ooi et al. 2023) by, for instance, producing

text transcriptions for video content (Amarasinghe et

al. 2023).

Despite the apparent massive potential that

GenAI offers for education, some drawbacks have to

be taken into consideration. Through the potential

processing of personal data, privacy and safety issues

arise (Ooi et al. 2023). In addition, there is concern

regarding the impacts of an increased psychological

distance between teachers and students. The

advancements in GenAI for education further spark

debate regarding the originality of submissions and

infusing the need to reevaluate forms of grading and

examination. Another obstacle constitutes the

potential bias in used data and algorithms since it

could influence an individual’s learning experience

and outcomes (Ooi et al. 2023), contradicting the

pursued aim of equality in education (Yan et al. 2023).

Moreover, several application possibilities of

GenAI for healthcare were retrieved from the

examined publications. Such models allow for the

rapid provision of answers to medical questions and

even the translation of such content (Ooi et al. 2023).

Patients can further use AI to better understand and

describe their experienced symptoms before seeing

an actual doctor which can lead to more accurate

diagnoses (Ooi et al. 2023). Furthermore, several

application possibilities of GenAI for healthcare were

retrieved from the examined publications. Such

models allow for the rapid provision of answers to

medical questions and even the translation of such

content (Ooi et al. 2023). Patients can further use AI

On the Current State of Generative Artiﬁcial Intelligence: A Conceptual Model of Potentials and Challenges

847

to better understand and describe their experienced

symptoms before seeing an actual doctor which can

lead to more accurate diagnoses (Ooi et al. 2023).

Similarly, Kanbach et al. (2023) describe the rise of

AI-powered chatbots to increase accessibility to

therapy, especially for mental health issues. However,

these sophisticated capabilities pose significant

requirements to the correctness and applicability of

the results (Ooi et al. 2023) and for revised regulatory

standards (responsibility of AI) (Mannuru et al. 2023).

Additionally, the adequate personalized medical

advice via GenAI necessitates multiple user

interactions and using the right text prompts, which

will pose a challenge for laypersons (Ooi et al. 2023).

Accordingly, literature suggests using GenAI only as

a supplement to traditional therapy (Kanbach et al.

2023).

Furthermore, the healthcare sector can benefit

from the predictive capabilities of GenAI models for

diagnosis and detection of diseases, for example,

from medical images and records (Mannuru et al.

2023; Ooi et al. 2023). As a consequence, progression,

monitoring, and potentially outcome of treatments

could be enhanced (Mannuru et al. 2023).

Nevertheless, ethical challenges like potential bias

and privacy concerns (Mannuru et al. 2023; Ooi et al.

2023) pose danger to patients’ acceptance and GenAI

effectiveness in healthcare. Besides, such algorithms

tend to overlook the social context in their analyses

(Mannuru et al. 2023). Finally, the study of Karinshak

et al. (2023) revealed that LLMs can also generate

more effective public health messages (e.g., pro-

vaccination messages) compared to human-initiated

communication. Despite the positive implications of

this discovery, a critical discourse with the potential

exploits such as spreading misinformation and

manipulating target groups is required (Karinshak et

al. 2023).

3.1.2 Industry

In the SLR, several use cases were found where

GenAI can offer support for various business process.

For example, based on sentiments scores calculated

by ChatGPT on corporate financial statements, the

risk management capability and stock return of a

company can be predicted (Chen et al. 2023). In

general, GenAI possesses lots of potential for the

finance sector such as the forecast of market trends,

detection of investment prospects, fraud detection,

and personalized financial advice (Kanbach et al.

2023; Ooi et al. 2023). With the LLMs’ ability to

process massive amount of demographic and browser

history data, better understanding of customer

behavior, leading to enhanced marketing strategies

and more customer touchpoints (e.g.,

hyperpersonalization), is enabled (Ooi et al. 2023).

Additionally, GenAI can be used in the product

ideation process and provide better product

descriptions. Therefore, this improved customer

experience opens up new revenue streams for firms

(Mannuru et al. 2023). Nevertheless, this new

development in marketing strategies reduces the

perceived empathy and emotion in service delivery

(“dehumanization”) (Ooi et al. 2023), which is

especially fostered with the tendency to a future

“overreliance on AI” (Mannuru et al. 2023).

Furthermore, Ooi et al. (2023) highlight

additional application scenarios in manufacturing.

GenAI is utilized for creating training programs of

employees, reducing the need for humans for this task.

Here, these models also gain attention for their

capability to enhance predictive maintenance through

sophisticated simulations and forecasts (anomaly

detection) (Ooi et al. 2023). For its effectiveness,

especially in manufacturing and finance, the

availability and quality of training data is important

but oftentimes constitutes an issue. Similarly,

infrastructure requirements, potentially impacted by

severe regulations, are significant and expensive.

Despite GenAI’s general ability to support decision-

making in various business processes, certain

considerations need to be made to ensure its actual

applicability in practice. These include ethical

concerns (e.g., bias), data privacy issues,

hallucinations, and currently missing robust

(international) regulations (Mannuru et al. 2023; Ooi

et al. 2023). Apart from that, seeing the huge

automation potential of GenAI, the fear of large-scale

losses comes as little surprise.

According to the literature, GenAI offers

capabilities to support tasks in the legal industry.

Ioannidis et al. (2023) developed an AI tool that

summarizes and categorizes regulatory information

for clients. Lam et al. (2023) assessed the usefulness

of LLMs for contract drafting to simplify the

workflow of legal professionals. While the basic

utility of the approaches could be confirmed, some

major obstacles still need to be considered. First of all,

significant testing before deployment is required

since GenAI models are not immune to creating

plausible sounding text that is wrong (Ioannidis et al.

2023). Because of the severe implications of potential

errors in the law domain where accurate information

is mandatory, this poses a massive risk. Thus, the

used prompt setting is important (Lam et al. 2023). As

a conclusion, also due to privacy and confidentiality

ICEIS 2024 - 26th International Conference on Enterprise Information Systems

848

doubts (Ioannidis et al. 2023), enhancements are still

needed in this area (Lam et al. 2023).

Next, human resources (HR) profit from GenAI

in various ways. Employee training initiatives can be

improved (Mannuru et al. 2023; Ooi et al. 2023) as

already seen in the educational sector. In general,

GenAI shows potential to significantly change the job

market. While new job types for working with GenAI

are created (e.g., prompt engineering), demand for

other positions involving automatable tasks (e.g.,

customer service) and even services with human

expertise (e.g., programming) is reduced. Self-

evidently, this sparks concerns regarding

employment opportunities as well (Mannuru et al.

2023). Nevertheless, GenAI allows to support

employees by delegating repetitive tasks. For

example, Lukauskas et al. (2023) describe how LLMs

can be utilized to extract skills demand from job

profiles to improve the recruitment process.

Furthermore, Ooi et al. (2023) outline several

opportunities for using GenAI to enhance team

collaboration such as establishing reminders for

deliverables and overcoming resource allocation

challenges. Despite the summarized advantages, the

actual applicability is still hampered by obstacles in

terms of significant data requirements to avoid bias,

lack of explainability leading to low trust, and

partially inaccurate as well as inappropriate responses

(Ooi et al. 2023).

3.1.3 Sustainability

The relationship of GenAI and sustainability

constitutes a double-edged sword. While certain

techniques supporting the reduction of carbon

emissions are facilitated, the needed compute

intensity of these models pose considerable energy

demands (Mannuru et al. 2023) which are not

necessarily compatible with sustainability

requirements. Accordingly, the environmental impact

of GenAI across its entire lifecycle needs to be

evaluated (Ooi et al. 2023). On this note, amongst,

others, regulatory frameworks and developing a

competent workforce in this domain are required for

AI to make a positive impact in sustainability

concerns.

The predictive capabilities of GenAI can be used

for energy savings for various sectors by forecasts

and resulting modifications to power management

and resource distribution (Mannuru et al. 2023; Ooi

et al. 2023). Moreover, AI can aid in the design

optimization of existing or new (IT) systems with

regards to efficiency, economic, and scalability

metrics (Mannuru et al. 2023; Ooi et al. 2023). This

can also include digital twin simulations that

incorporate environmental characteristic. In the next

step, necessary components can be compared

regarding their ecological impact (e.g., Echochain AI)

to achieve sustainable procurement (Ooi et al. 2023).

This is also applicable for the end of the product

lifecycle when the processing of data on (electronical)

waste and the resulting insights can lessen the

environmental impact. Finally, Villiers et al. (2023)

examine the potential of AI-powered sustainability

reporting due to high-speed analyzing abilities of

large quantities of corporate data. However, the

authors point out the risk of greenwashing, since

some information (e.g., financial) are “easier to

process” (Villiers et al. 2023).

3.1.4 Art and Media

Similar to the potential of GenAI in content

generation for educational purposes, ChatGPT can

assist journalists in writing as well, which is shown

by Pavlik (2023). As a consequence, in the long run,

this may constitute a threat to human professionals

that could fall victim to cost savings in this industry.

Despite the significant potential, reservations exist

because of possible misuse to produce information

with malicious intent or partial lack in range and

depth of knowledge in GenAI models (Pavlik 2023).

Additionally, the text generation ability of LLMs

finds use in producing new cooking recipes. In

conjunction with genetic algorithms, GenAI can be

used to create the recipe text from encoded recipe

solutions (Razzaq et al. 2023). However, the current

approach does not account for the nutritional values

of the food which is a notable limitation for being able

to compose a conscious diet.

3.2 Audio

Based on the conducted SLR, audio is the least

represented field of GenAI. Only three of the included

publications specifically discuss the application of

GenAI for audio-related use cases.

Lv (2023) discusses the role of GenAI in the

metaverse. In this context, it is mentioned that GenAI

can be utilized for speech synthesis and the

generation of music. The Wavenet model constitutes

one of the well-known examples and is allegedly able

to “produce convincing synthetic voice and music”

(Lv 2023).

Ocampo et al. (2023) also present an GenAI-

driven method for the domain of art. The authors

describe a GPT-3-powered data sonification

approach that uses the most-read Wikipedia articles

On the Current State of Generative Artiﬁcial Intelligence: A Conceptual Model of Potentials and Challenges

849

of a given day as data input. With this a basis, a set of

requests is given to the model to perform a semantic

mapping between the data interpretations and sound

descriptions (Ocampo et al. 2023). Therefore, the

authors show the general ability of GenAI to create

prompt-specific outputs that can be used for the music

creation process. Nevertheless, the experiments

discovered some limitations with the interpretation

capabilities of the used model. At times, the assigned

labels did not reflect the input data accurately

(Ocampo et al. 2023). The authors deduce that GPT-

3’s numerical reasoning abilities might negatively

influence the results in this regard and that the

prompting approach plays a significant factor in the

quality of the results.

Additionally, Mannuru et al. (2023) found that

accents in audio stream can be modified for better

understandability and high-quality speech for people

with disabilities related to communication skills can

be generated (Houde et al. 2020; Qadir 2022).

However, misusing GenAI for creating content with

malicious intents (e.g., deep fakes) constitutes a

specific danger in this regard (Tacheva and

Ramasubramanian 2023).

Although the literature features some great

potentials for audio content generation with GenAI,

the coverage remains sparse in comparison to the

other categories. Especially surprising is the absence

of concrete use cases related to the conservation of

voice of deceased (voice) actors in the examined

portion of the scientific body of knowledge.

3.3 Image

The creation of images with the help of GenAI

usually requires input data in the form of text or other

images, whereas the output quality largely depends

on the quality of that input (Fernberg et al. 2023).

While the capabilities to generate high-fidelity

images with appropriate descriptive prompts are

already quite sophisticated (Chang et al. 2023), the

underlying creativity cannot be expected to be taken

away from humans (Hoggenmueller et al. 2023;

Vartiainen and Tedre 2023), although the advantages

that humans have are becoming less significant with

the rapid improvements in the AI domain

(Seneviratne et al. 2022). However, the roles of

developers and prompt engineers might shift towards

reviewing, refining, and evaluating the results rather

than creating all output by themselves.

Throughout the literature reviewed, researchers

tend to view GenAI for images as a digitized

colleague in a human-machine partnership rather than

a mere tool. Especially when it comes to designing

digital content, products, sketches, fashion, or entire

story-boards (Chang et al. 2023; Hoggenmueller et al.

2023; Lv 2023), GenAI oftentimes exceeds the

expectations placed on a software tool. Consistent

with the vision of GenAI as a digital co-creator, the

technology also partially relieves humans of the

necessity to understand the underlying mechanisms

and instead lets them describe what they need before

the orders are executed (Vartiainen and Tedre 2023).

The most frequently mentioned models for generating

images include DALL-E, Stable Diffusion,

MidJourney, Parti and Imagen (Chang et al. 2023;

Hoggenmueller et al. 2023; Hong et al. 2023; Lv

2023; Turchi et al. 2023).

3.3.1 Product Design

The design and development of physical products is

one of the main application areas of AI-based image

generation. For example, fashion is a domain in

which a large number of preliminary conceptual

preparation might be necessary before a decision on

the final design of a collection is made. GenAI can

help to transfer a certain style from a manual sketch

to an aesthetic realistic clothing image in a fast

manner with a quality comparable to human designers

(Wu et al. 2023). However, as usual for sophisticated

ML models, this approach potentially requires a

substantial amount of training data, meaning

preparatory human labor. A similar use case for

GenAI in this domain is image inpainting. Existing

clothing images can be manipulated in an image-to-

image generation process, in which parts of clothes

are overwritten with new content (Muhammad et al.

2023). Newly designed outfits could be tried on

digitally as part of an existing catalog.

Although fashion companies in particular rely

heavily on creative minds to come up with

revolutionary designs while breaking away from

outdated ideas, this is not only posing a problem in

this sector. Robotics can be identified as an area in

which the issue of design fixation, meaning the

adherence to well-established conceptions, is widely

prevalent (Hoggenmueller et al. 2023). When

creating new robot designs, GenAI could help to

overcome simple notions of anthropomorphic or

zoomorphic depictions of robots by incorporating

other elements from the training phase into the

creation process.

While the visual design of physical systems might

be a secondary task in engineering, the actual

functioning of entire electrical assemblies can also be

supported by GenAI for images. In one of the most

topical branches of automotive engineering, e-

ICEIS 2024 - 26th International Conference on Enterprise Information Systems

850

mobility, a recent study investigated the capabilities

of generative adversarial networks (GANs) for the

preliminary design of dynamic inductive charging

stations (Curtis et al. 2023). With this technology,

cars do not need to stop for charging if the coil is

integrated into the road on which a car with

corresponding electrical systems is moving. This

approach shows that the inclusion of technical

requirements such as safety, energy consumption, and

flexibility is a possible option.

Lastly, hand crafting is identified as an area for

GenAI in the domain of product design. When

regarding crafting as a four-phase process consisting

of ideation, design, making, and reflective evaluation

(Vartiainen and Tedre 2023), artificially generated

images can especially benefit the first and partially

the second stage. As the intended design becomes

more and more complex in the course of its

finalization, the level of detail of ideas can

increasingly no longer be converted into adequate

prompts. Verheijden and Funk (2023) call this

circumstance the crooked bow tie effect, since one

side (i.e., the input) in its extent can only be

transformed into something smaller that does not

sufficiently represent the given input. However, in the

initial phases of craft product design, visualization is

seen as an important means of communication and

feedback and thus constitutes an additional use case

for GenAI for images.

3.3.2 Construction

Similar to engineering, the field of construction often

necessitates the integration of functional

requirements into design processes. For example,

urban design requires the consideration of complex

concepts such as health and safety (Seneviratne et al.

2022). When developing GANs with the goal of

generating images of meaningful urban and landscape

compositions from descriptive texts, engineers need

to be skilled in prompt engineering (Dortheimer et al.

2023), so that the model can be trained to visually link

specific terms with logical consequences. For

instance, in their study, Seneviratne et al. (2022)

associated terms like healthy with green

environments and safe with clean paint on completed

constructions. In contrast to the creation of still

images, another use case for GenAI is related to the

preparation of interactive 3D scenes in virtual reality.

Bussell et al. (2023) conducted a study in which

participants used GenAI to first create conceptual

designs of construction projects, which were then

modeled and integrated into an explorable virtual

world.

One step further, studies in architecture view

GenAI as part of a multistep workflow from early

inspirations to the enhancement of final renderings

(Dortheimer et al. 2023). When creating vivid

representations of buildings, landscapes or

community spaces, modern approaches often use 3D

models for further inspection. In addition to direct

image generation, this also offers potential for GenAI,

as the models must be provided with textures in the

3D software and the renderings need to be post-

processed, for example by adding context, light and

weather effects (Dortheimer et al. 2023; Turchi et al.

2023). In the ideation phase, GenAI for images can

also be used to overcome the aforementioned problem

of design fixation in the field of architecture

(Dortheimer et al. 2023; Hoggenmueller et al. 2023).

Concluding the discussion of the different levels

of detail in construction, the final area of application

identified here for GenAI revolves around interior

and asset design. In general, work in this sector

consists of three phases: ideation, schematic drafting

and layout planning (He et al. 2023), with the last

phase being similar to the arrangement of elements in

landscape and urban development at a reduced scale.

Interior designers need to build up asset libraries,

including materials, textures, or even representations

of people for more realistic visualizations (Fernberg

et al. 2023). While GenAI for images (e.g., through

GANs) can drastically increase productivity when

creating images of everyday things, limitations

become clear when the integration of implicit factors

is necessary, such as certain cultural styles and

aesthetics (He et al. 2023). Thus, human intervention

for the final composition of design alternatives is still

currently essential. For the future, not only in the field

of interior design, researchers see the expansion of

image generators to include linguistic models, more

sophisticated inpainting functions and the extension

to participatory process models as possible directions

(Fernberg et al. 2023).

3.3.3 Art and Media

As a basic example of what GenAI is capable of, pure

images resembling paintings or photographs can be

generated by models that mimic the artistic style of

different artists or have been trained to create novel

stylistic compositions themselves (Chang et al. 2023).

In the literature, the connection between art and AI

has recently led to intense discussions about whether

something that is considered to be art can be created

by AI at all and, if so, who owns the copyright to it.

Jiang et al. (2023) argue that the creation of art is an

endeavor unique to humans, as it is based on

On the Current State of Generative Artiﬁcial Intelligence: A Conceptual Model of Potentials and Challenges

851

aesthetical and cultural experience, while ML models,

once trained, are fixed on whatever was part of the

training data. In a similar sense, Chang et al. (2023)

discuss the question of whether GenAI can be

considered an artist or merely a tool for actual artists.

This in turn leads to the important question of who

should be credited for the images produced with AI

models: the one who developed the GenAI platform

or the one who created the prompt that leads to a

particular image. Furthermore, if the model imitates

an artistic style, the question arises as to whether there

is a copyright violation against the inventor of a style

(Vartiainen and Tedre 2023; Verheijden and Funk

2023), and how a style could be patented in such a

scenario in the first place.

Enjoyable art products for recreational purposes

include goods from the fields of entertainment and

games. Applications for GenAI from the construction

domain, such as assisting in the development of 3D

assets, can also accelerate the creation of consistent

scenes from descriptive prompts by providing

structures, materials, and textures for games or the

metaverse (Lv 2023). For simple games based on

images, such as spot-the-difference, GenAI can be

used to generate the essential game content directly

and with debatable quality (Hong et al. 2023). In

contrast, for more complex and extensive games,

GenAI can support subtasks of development, such as

providing maps, character designs, and equipment

based on images or, more broadly, dialogs, emotions

of non-playable characters, behaviors, and other

gameplay elements (Lv 2023).

3.3.4 Public Services

GenAI for images can also be employed for tasks in

domains that do not have a direct monetary value but

might benefit communities in other regards. Under

the umbrella term of public services, this includes

applications in education, science, and healthcare.

In education, two directions can be identified.

First, it is argued that students need to be educated on

how GenAI works, how it can be used, and how to

distinguish content that was created by GenAI from

human-made productions (S. Ali et al. 2021).

Especially the last aspect is of topical interest, as

researchers and society are observing an increase in

so-called deep fakes, misinformation, and the use of

GenAI for malicious purposes. The second avenue in

education is the use of GenAI itself during lessons

and training programs. A study by Vartiainen and

Tedre (2023) investigated how the ideation process in

crafting education could be enhanced by means of

text-to-image generative approaches. However, the

authors remark that the lack of materiality (i.e.,

haptics) cannot be compensated by GenAI.

In archaeology, a subcategory of science, Cobb

(2023) investigated whether images can be generated

which reconstruct archaeological elements. However,

the quality was not considered sufficient, as science

requires a high level of factual knowledge, experience

and understanding, for example when depicting

ancient writings or architectural plans. Nevertheless,

the author recognizes a potential for GenAI when it

comes to 3D reconstructions of ancient artifacts and

places. On the other hand, Cobb (2023) notes that

archaeology is not a very profitable field, which

makes it questionable whether GenAI is worth an

investment.

One particular use case found in healthcare for

GenAI for images does not rely on text-to-image or

image-to-image concepts as before, but on the

reconstruction of images from neural activity

(Ozcelik and VanRullen 2023). Although the research

is still at an early stage, the authors have obtained

overall satisfactory results when using functional

magnetic resonance imaging (fMRI) data as input for

image reconstruction and suggested that this idea

could be further developed to better understand, for

example, the sensory and semantic system or the

functioning of the visual cortex.

3.4 Code

GenAI for code offers multiple use cases that shift the

roles of human developers, from generating entirely

new code to translating from one language to another

and documenting its functionality in a meaningful

way (Kanbach et al. 2023). However, the extent to

which GenAI can actually improve certain KPIs such

as productivity, code quality, and program

correctness is disputed in the literature (Weisz et al.

2022). Thus, automation is still limited, and human

workers will most likely still be needed in

development processes in the near future.

While the generation of code can be driven by

developers' ideas and the general desire to create

something new, the reasons for translating existing

code can depend on external modernization pressure.

Use cases include, for example, preparing

applications for migration to cloud computing

environments or transitioning from legacy languages

to more current frameworks (Weisz et al. 2021).

Studies suggest that developers value code quality

and correctness higher than a boost in productivity,

whereby the generative capabilities of GenAI for

code are already quite advanced (Cámara et al. 2023;

Weisz et al. 2022). In contrast to natural language,

ICEIS 2024 - 26th International Conference on Enterprise Information Systems

852

code follows stricter rules, making errors more

critical, and only distinguishes the results into code

that can be compiled and code that causes errors

(Weisz et al. 2021). As a hybrid use case, GenAI can

be applied to document the semantic functionalities

of code, as was attempted by Muller et al. (2021).

In another stage of the software development

cycle, when the system is planned and designed,

GenAI can also be employed for software modeling.

Cámara et al. (2023) found that ChatGPT currently

has limited capabilities in this regard, as while the

syntax of software models is largely sound, the

semantics are sometimes not. The authors see

opportunities when interfaces for LLMs are provided

so that experts can customize the models.

Lastly, an early study by Nelson et al. (2023)

explored how ChatGPT could be used to generate 3D

models for computer-aided design (CAD) software,

bringing together the potential of GenAI for visual

creation and coding. The models are usually stored in

formats that include the positions of the vertices, the

edges or the sequential manufacturing processes. As

3D printers become more popular and 3D models in

specific formats are needed to produce results, this

could be a promising research direction.

4 CONCEPTUAL MODEL

While the presented application domains of GenAI

share general advantages, such as supporting human

decision making, requiring little technical knowledge

and being available at low cost (Kanbach et al. 2023),

serious challenges could be identified from the

literature that are prevalent in all areas. Since

generative models are dependent on the quality and

statistical distributions of the training data, the

learning process can be influenced by biases if the

data used suffers from corresponding structures. This

can be caused either by misinterpretation, when the

data unintentionally fuels stereotypes, or by under-

and over-representation, when certain phenomena are

not represented in the training data in adequate

numbers (Vartiainen and Tedre 2023).

Related to issues of ethics, challenges posed by

GenAI onto engineers and researchers reach from

trust, transparency, and explainability to fears of

misuse, deep fakes, and autonomously acting

malicious machines. In their study, S. Ali et al. (2021)

call for more education on such topics, as generative

models are becoming increasingly powerful and

produce hyper-realistic results that may become

indistinguishable to the average user. Particularly

affected domains are image, when untrue

circumstances are shown or an artist's style is imitated

for images with malicious messages (Vartiainen and

Tedre 2023), and audio, when the voices of

politicians, for example, are faked.

From a legal perspective, copyright issues are

frequently addressed in the literature, particularly in

art-related publications. One point of contention is the

question of who deserves the credit for the generated

works: the generative model and its developers or the

person who provided the prompt that leads to the

result (Chang et al. 2023). Furthermore, if GenAI is

imitating the style of an established artist, the

question arises as to whether there should be some

kind of financial compensation if it is decided that the

copying is legal at all (Jiang et al. 2023; Vartiainen

and Tedre 2023; Verheijden and Funk 2023).

Less financially attractive areas, on the other hand,

suffer from the challenge of whether an investment in

GenAI would be worthwhile. For example, a science

such as archaeology is heavily based on factual

knowledge and does not generate large profits (Cobb

2023). Similarly, in healthcare, efforts such as

reconstructing images from neural activity may not

directly lead to major financial gains, but are a hot

topic in medicine nonetheless (Ozcelik and

VanRullen 2023).

What all areas of GenAI utilization have in

common is the expectation that the roles of humans

in the workforce and society will shift, as the new

evolutionary stage of AI requires less expertise on the

fundamental usage level and still allows to easily

generate new content (Kanbach et al. 2023). The

conceptual model in Figure 2 illustrates all the use

cases discussed within the four areas of text, image,

audio, and code generation. However, it is not

claimed that the model covers all existing

applications of GenAI, as the area under investigation

is developing dynamically. The color coding is

divided into three groups of characteristics. Based on

the literature findings, green is associated with well-

researched areas for which studies with working use

cases, a comprehensible financial value and a low but

manageable level of challenges could be found. The

second variant, yellow, is intended for areas that are

considered more experimental, with working

examples and a justifiable financial impact. However,

this category includes areas where major technical

and legal challenges remain to be overcome, as

outlined in the corresponding section. Finally, areas

that have not been sufficiently researched but still

provide some conceptual ideas are highlighted in red.

In this category, the financial impact compared to the

necessary investment is lower than in the other areas,

while the challenges and requirements are higher.

On the Current State of Generative Artiﬁcial Intelligence: A Conceptual Model of Potentials and Challenges

853

Figure 2: Conceptual model for GenAI potentials and challenges.

5 CONCLUSION

GenAI applications have recently gained significant

attention because of the massive opportunities for

industry and society. Accordingly, it is expected that

GenAI will severely impact numerous areas of life in

the future. Nevertheless, the benefits of this new

phenomenon are accompanied by noteworthy

obstacles that should be taken into consideration (e.g.,

bias, accuracy of results, potential misuse). Hence, in

this work, the current state-of-the-art of GenAI in the

scientific literature is examined, focusing on

potentials and challenges. Based on the generated

types of content, the analysis is divided in the four

categories text, audio, image, and code. We derive a

conceptual model for GenAI application areas that

consolidates and assesses the findings from the

literature. One notable observation is that the view of

AI is shifting with the advance of GenAI, as AI is no

longer seen as a mere decision-making tool, but as a

kind of human-like collaborator in creative tasks

(Turchi et al. 2023). The results of the study could be

extended by considering further scientific literature

databases and including forward and backward

searches. Furthermore, specific emphasis could be

put on the potential of hybrid GenAI outputs (e.g.,

text combined with video, or generated audio in a

virtual environment created by GenAI) as additional

promising developments might become visible,

especially in the context of the metaverse (Lv 2023).

In conclusion, the full potential of GenAI for cross-

domain applications will still require considerable

research efforts in the foreseeable future.

REFERENCES

Ali, K., Barhom, N., Tamimi, F., & Duggal, M. (2023).

Chatgpt-A double-edged sword for healthcare

education? Implications for assessments of dental

students. In European Journal of Dental Education :

Official Journal of the Association for Dental

Education in Europe.

Ali, S., DiPaola, D., & Breazeal, C. (2021). What are

GANs? Introducing Generative Adversarial Networks

to Middle School Students. In Proceedings of the AAAI

Conference on Artificial Intelligence, 35(17), 15472–

15479.

ICEIS 2024 - 26th International Conference on Enterprise Information Systems

854

Amarasinghe, I., Marques, F., Ortiz-Beltrán, A., &

Hernández-Leo, D. (2023). Generative Pre-trained

Transformers for Coding Text Data? An Analysis with

Classroom Orchestration Data. In Lecture Notes in

Computer Science. Responsive and Sustainable

Educational Futures. Springer Nature Switzerland.

Bussell, C., Ehab, A., Hartle-Ryan, D., & Kapsalis, T.

(2023). Generative AI for Immersive Experiences:

Integrating Text-to-Image Models in VR-Mediated Co-

design Workflows. In Communications in Computer

and Information Science. HCI International 2023

Posters. Springer Nature Switzerland.

Cámara, J., Troya, J., Burgueño, L., & Vallecillo, A. (2023).

On the assessment of generative AI in modeling tasks:

an experience report with ChatGPT and UML. In

Software and Systems Modeling, 22(3), 781–793.

Chang, M., Druga, S., Fiannaca, A. J., Vergani, P., Kulkarni,

C., Cai, C. J., & Terry, M. (2023). The Prompt Artists.

In Creativity and Cognition. ACM.

Chen, B., Wu, Z., & Zhao, R. (2023). From fiction to fact:

the growing role of generative AI in business and

finance. In Journal of Chinese Economic and Business

Studies, 21(4), 471–496.

Cobb, P. J. (2023). Large Language Models and Generative

AI, Oh My! In Advances in Archaeological Practice,

11(3), 363–369.

Curtis, A., Shahid Chowdhury Oni, M. S., Inoue, S.,

Kamineni, A., Zane, R., & Flann, N. (2023). Generative

Neural Network Approach to Designing Dynamic

Inductive Power Transfer Systems. In 2023 IEEE

Wireless Power Technology Conference and Expo

(WPTCE). IEEE.

Daase, C., Volk, M., Staegemann, D., & Turowski, K.

(2023). The Future of Commerce: Linking Modern

Retailing Characteristics with Cloud Computing

Capabilities. In Proceedings of the 25th International

Conference on Enterprise Information Systems.

SCITEPRESS - Science and Technology Publications.

Dortheimer, J., Schubert, G., Dalach, A., Brenner, L., &

Martelaro, N. (2023). Think AI-side the Box! Exploring

the Usability of Text-to-Image Generators for

Architecture Students. In Proceedings of the

International Conference on Education and Research

in Computer Aided Architectural Design in Europe.

Farazouli, A., Cerratto-Pargman, T., Bolander-Laksov, K.,

& McGrath, C. (2023). Hello GPT! Goodbye home

examination? An exploratory study of AI chatbots

impact on university teachers’ assessment practices. In

Assessment & Evaluation in Higher Education, 1–13.

Fernberg, P., George, B. H., & Chamberlain, B. (2023).

Producing 2D Asset Libraries with AI-powered Image

Generators. In Journal of Digital Landscape

Architecture, 2023(8), 186–194.

François-Lavet, V., Henderson, P., Islam, R., Bellemare, M.

G., & Pineau, J. (2018). An Introduction to Deep

Reinforcement Learning. In Foundations and Trends®

in Machine Learning, 11(3-4), 219–354.

Haertel, C., Daase, C., Staegemann, D., Nahhas, A., Pohl,

M., & Turowski, K. (2023). Toward Standardization

and Automation of Data Science Projects: MLOps and

Cloud Computing as Facilitators. In Proceedings of the

15th International Joint Conference on Knowledge

Discovery, Knowledge Engineering and Knowledge

Management. SCITEPRESS - Science and Technology

Publications.

He, Z., Li, X [Xiaomei], Fan, L., & Wang, H. J. (2023).

Revamping Interior Design Workflow Through

Generative Artificial Intelligence. In Communications

in Computer and Information Science. HCI

International 2023 Posters. Springer Nature

Switzerland.

Hoggenmueller, M., Lupetti, M. L., van der Maden, W., &

Grace, K. (2023). Creative AI for HRI Design

Explorations. In Companion of the 2023 ACM/IEEE

International Conference on Human-Robot Interaction.

ACM.

Hong, Y.‑G., Jeong, J.‑Y., Jeong, J.‑W., & Lukianova, E.

(2023). Spot The Difference: AI, Please Make This for

Me! In Extended Abstracts of the 2023 CHI Conference

on Human Factors in Computing Systems. ACM.

Houde, S., Liao, V., Martino, J., Muller, M., Piorkowski, D.,

Richards, J., Weisz, J., & Zhang, Y. (2020, March 2).

Business (mis)Use Cases of Generative AI.

Hwang, W.‑Y., & Nurtantyana, R. (2022). The integration

of multiple recognition technologies and artificial

intelligence to facilitate EFL writing in authentic

contexts. In 2022 6th International Conference on

Information Technology (InCIT). IEEE.

Ioannidis, J., Harper, J., Sheng Quah, M., & Hunter, D.

(2023). Gracenote.ai: Legal Generative AI for

Regulatory Compliance. In Proceedings of the Third

International Workshop on Artificial Intelligence and

Intelligent Assistance for Legal Professionals in the

Digital Workplace (LegalAIIA 2023),

Jauhiainen, J. S., & Guerra, A. G. (2023). Generative AI

and ChatGPT in School Children’s Education:

Evidence from a School Lesson. In Sustainability,

15(18), 14025.

Jiang, H. H., Brown, L., Cheng, J., Khan, M., Gupta, A.,

Workman, D., Hanna, A., Flowers, J., & Gebru, T.

(2023). AI Art and its Impact on Artists. In Proceedings

of the 2023 AAAI/ACM Conference on AI, Ethics, and

Society. ACM.

Kanbach, D. K., Heiduk, L., Blueher, G., Schreiter, M., &

Lahmann, A. (2023). The GenAI is out of the bottle:

generative artificial intelligence from a business model

innovation perspective. In Review of Managerial

Science.

Karinshak, E., Liu, S. X., Park, J. S., & Hancock, J. T.

(2023). Working With AI to Persuade: Examining a

Large Language Model's Ability to Generate Pro-

Vaccination Messages. In Proceedings of the ACM on

Human-Computer Interaction, 7(CSCW1), 1–29.

Lam, K.‑Y., Cheng, V. C. W., & Yeong, Z. K. (2023).

Applying Large Language Models for Enhancing

Contract Drafting. In 3rd International Workshop on

Artificial Intelligence and Intelligent Assistance for

Legal Professionals in the Digital Workplace,

LegalAIIA 2023, 70–80.

Liu, W., Wang, Z., Liu, X [Xiaohui], Zeng, N., Liu, Y., &

Alsaadi, F. E. (2017). A survey of deep neural network

architectures and their applications. In Neurocomputing,

234, 11–26.

On the Current State of Generative Artiﬁcial Intelligence: A Conceptual Model of Potentials and Challenges

855

Lukauskas, M., Šarkauskaitė, V., Pilinkienė, V., Stundžienė,

A., Grybauskas, A., & Bruneckienė, J. (2023).

Enhancing Skills Demand Understanding through Job

Ad Segmentation Using NLP and Clustering

Techniques. In Applied Sciences, 13(10), 6119.

Lv, Z. (2023). Generative artificial intelligence in the

metaverse era. In Cognitive Robotics, 3, 208–217.

Mannuru, N. R., Shahriar, S., Teel, Z. A., Wang, T., Lund,

B. D., Tijani, S., Pohboon, C. O., Agbaji, D., Alhassan,

J., Galley, J., Kousari, R., Ogbadu-Oladapo, L., Saurav,

S. K., Srivastava, A., Tummuru, S. P., Uppala, S., &

Vaidya, P. (2023). Artificial intelligence in developing

countries: The impact of generative artificial intelligence

(AI) technologies for development. In Information

Development, Article 02666669231200628.

Muhammad, A., Lee, K., Lim, C., Hyeon, J., Salman, Z., &

Han, D. (2023). GAN vs Diffusion: Instance-Aware

Inpainting on Small Datasets. In 2023 International

Technical Conference on Circuits/Systems, Computers,

and Communications (ITC-CSCC). IEEE.

Muller, M., Wang, A. Y., Ross, S. I., Weisz, J. D., Agarwal,

M., Talamadupula, K., Houde, S., Martinez, F.,

Richards, J., Drozdal, J., Liu, X [Xuye], Piorkowski, D.,

& Wang, D. (Eds.) (2021). How Data Scientists

Improve Generated Code Documentation in Jupyter

Notebooks. : Vol. 2903.

Nelson, M. D., Goenner, B. L., & Gale, B. K. (2023).

Utilizing ChatGPT to assist CAD design for

microfluidic devices. In Lab on a Chip, 23(17), 3778–

3784.

Ocampo, R., Andres, J., Schmidt, A [Adrian], Pegram, C.,

Shave, J., Hill, C., Wright, B., & Bown, O. (2023).

Using GPT-3 to Achieve Semantically Relevant Data

Sonificiation for an Art Installation. In Lecture Notes in

Computer Science. Artificial Intelligence in Music,

Sound, Art and Design. Springer Nature Switzerland.

Ooi, K.‑B., Tan, G. W.‑H., Al-Emran, M., Al-Sharafi, M.

A., Capatina, A., Chakraborty, A., Dwivedi, Y. K.,

Huang, T.‑L., Kar, A. K., Lee, V.‑H., Loh, X.‑M., Micu,

A., Mikalef, P., Mogaji, E., Pandey, N., Raman, R.,

Rana, N. P., Sarker, P., Sharma, A., . . . Wong, L.‑W.

(2023). The Potential of Generative Artificial

Intelligence Across Disciplines: Perspectives and

Future Directions. In Journal of Computer Information

Systems, 1–32.

Ozcelik, F., & VanRullen, R. (2023). Natural scene

reconstruction from fMRI signals using generative

latent diffusion. In Scientific Reports, 13(1), 15666.

Panthier, C., & Gatinel, D. (2023). Success of ChatGPT, an

AI language model, in taking the French language

version of the European Board of Ophthalmology

examination: A novel approach to medical knowledge

assessment. In Journal Francais D'ophtalmologie,

46(7), 706–711.

Pavlik, J. V. (2023). Collaborating With ChatGPT:

Considering the Implications of Generative Artificial

Intelligence for Journalism and Media Education. In

Journalism & Mass Communication Educator, 78(1),

84–93.

Qadir, J. (2022). Engineering Education in the Era of

ChatGPT: Promise and Pitfalls of Generative AI for

Education. In TechRxiv.

Razzaq, M. S., Maqbool, F., Ilyas, M., & Jabeen, H. (2023).

EvoRecipes: A Generative Approach for Evolving

Context-Aware Recipes. In IEEE Access, 11, 74148–

74164.

Seneviratne, S., Senanayake, D., Rasnayaka, S.,

Vidanaarachchi, R., & Thompson, J. (2022). DALLE-

URBAN: Capturing the urban design expertise of large

text to image transformers.

Siiman, L. A., Rannastu-Avalos, M., Pöysä-Tarhonen, J.,

Häkkinen, P., & Pedaste, M. (2023). Opportunities and

Challenges for AI-Assisted Qualitative Data Analysis:

An Example from Collaborative Problem-Solving

Discourse Data. In Lecture Notes in Computer Science.

Innovative Technologies and Learning. Springer

Nature Switzerland.

Tacheva, Z., & Ramasubramanian, S. (2023). Challenging

AI Empire: Toward a Decolonial and Queer Framework

of Data Resurgence. In Advance.

Turchi, T., Carta, S., Ambrosini, L., & Malizia, A. (2023).

Human-AI Co-creation: Evaluating the Impact of

Large-Scale Text-to-Image Generative Models on the

Creative Process. In Lecture Notes in Computer Science.

End-User Development. Springer Nature Switzerland.

Vartiainen, H., & Tedre, M. (2023). Using artificial

intelligence in craft education: crafting with text-to-

image generative models. In Digital Creativity, 34(1),

1–21.

Verheijden, M. P., & Funk, M. (2023). Collaborative

Diffusion: Boosting Designerly Co-Creation with

Generative AI. In Extended Abstracts of the 2023 CHI

Conference on Human Factors in Computing Systems.

ACM.

Villiers, C. de, Dimes, R., & Molinari, M. (2023). How will

AI text generation and processing impact sustainability

reporting? Critical analysis, a conceptual framework

and avenues for future research. In Sustainability

Accounting, Management and Policy Journal.

Weisz, J. D., Muller, M., Houde, S., Richards, J., Ross, S.

I., Martinez, F., Agarwal, M., & Talamadupula, K.

(2021). Perfection Not Required? Human-AI

Partnerships in Code Translation. In 26th International

Conference on Intelligent User Interfaces. ACM.

Weisz, J. D., Muller, M., Ross, S. I., Martinez, F., Houde,

S., Agarwal, M., Talamadupula, K., & Richards, J. T.

(2022). Better Together? An Evaluation of AI-

Supported Code Translation. In 27th International

Conference on Intelligent User Interfaces. ACM.

Wu, D., Yu, Z., Ma, N., Jiang, J., Wang, Y., Zhou, G., Deng,

H., & Li, Y [Yi] (2023). StyleMe: Towards Intelligent

Fashion Generation with Designer Style. In

Proceedings of the 2023 CHI Conference on Human

Factors in Computing Systems. ACM.

Yan, L., Sha, L., Zhao, L., Li, Y [Yuheng], Martinez‐

Maldonado, R., Chen, G., Li, X [Xinyu], Jin, Y., &

Gašević, D. (2023). Practical and ethical challenges of

large language models in education: A systematic

scoping review. In British Journal of Educational

Technology, Article bjet.13370.

ICEIS 2024 - 26th International Conference on Enterprise Information Systems

856