Generative AI Risk Management in Digital Economy

Victor Chang, Leigh Draper and

Simin Yu

Department of Operations and Information Management, Aston Business School, Aston University, Birmingham, U.K.

Keywords: Generative AI, Risk Management, Mitigation Strategies.

Abstract: In Healthcare Procurement, This Study Delves into the Integration of Generative AI, Focusing on Its

Application within HealthTrust Europe's marketing and communication frameworks. By analyzing the

interplay between innovative AI-driven content personalization and the associated ethical, security, and

operational risks, the research offers a nuanced perspective on leveraging technology for enhanced efficiency

and engagement. The study employs qualitative research methods to assess risks and propose mitigation

strategies, advocating for best practices in AI governance and risk management. It emphasizes the importance

of maintaining network security, data integrity, and ethical standards in deploying AI solutions.

1 INTRODUCTION

Generative artificial intelligence (AI) has sparked

considerable interest at both the human and corporate

levels. This is because generative AI has many

potential applications where it can be used to create

personalized and engaging marketing and

communications materials for healthcare

procurement companies (Ooi et al., 2023). These

technologies, such as Chat GPT and Copilot, can

generate new content and have many possible

applications (Preiksaitis & Rose, 2023). However,

generative AI poses significant challenges that may

be addressed effectively, ethically, and equitably

(Preiksaitis & Rose, 2023). However, more

organizations may still need to fully explore the long-

term implications of AI integration on outcomes and

services, so the gap points towards longitudinal

studies assessing the enduring effects of AI-driven

innovations. Thus, this study aims to identify

generative AI risks and mitigations to launch a

protected generative AI to create marketing and

communications content that considers and mitigates

potential pitfalls and threats.

The significance of this study is to identify the

risks of novel technology and risk management to

consider the best approaches for organizations. This

study will provide overviews and recommendations

for HealthTrust Europe and healthcare procurement

companies on development.

This study aims to identify novel generative AI

risks and mitigations and investigate the risks and

benefits of using best practice approaches to maintain

network security.

This study's objectives were stated as follows:

• exploring the risks, feasibility, and benefits of

developing a pilot scheme

• identifying and analyzing novel technology

risks and mitigations

• considering best practice approaches to

maintain network security

This study is categorized into several sections to

deal with the research question. The introduction

demonstrated the generative background and found

research significance. The literature review

introduced the risk of novel technologies and

generative AI models. Meanwhile, this study's

research method adopts qualitative research using

interpretivism theory. Result and recommendation

conducted mitigations. Finally, the last section

presented the conclusion and recommendation.

2 LITERATURE REVIEW

2.1 Healthtrust Europe Scenario

Health Trust Europe, which is the leading provider of

healthcare products and services to NHS Trusts,

private healthcare providers, and public sector

organizations, has a large and diverse customer base

and a network of suppliers, including market leaders

and small and medium-sized enterprises (SMEs) the

120

Chang, V., Draper, L. and Yu, S.

Generative AI Risk Management in Digital Economy.

DOI: 10.5220/0012729800003717

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

In Proceedings of the 6th International Conference on Finance, Economics, Management and IT Business (FEMIB 2024), pages 120-127

ISBN: 978-989-758-695-8; ISSN: 2184-5891

organization aims to deliver high-quality, cost-

effective, innovative solutions that meet its

customers' and suppliers' needs and expectations.

Health Trust intends to use generative AI solutions to

improve its marketing and communication strategy,

resulting in personalized and engaging content for its

target audiences. According to the Gartner

Organisation, expects that by 2026, more than 80% of

enterprises will have used or deployed generative AI-

enabled app. In contrast, Harvard Business Review

recently shared that a recent McKinsey survey

revealed that only 10% to 14% of companies

consistently deploy generative AI in their marketing

and sales initiatives (Lamarre et al., 2023). Thus,

there are risks to enterprises' declining to deploy

generative AI, which states that generative AI is

unreliable for customers.

2.2 Implementation of Health Trust

Europe

HealthTrust Europe aspires to deploy a secure

generative AI solution that will facilitate attaining

several strategic objectives. These objectives

encompass augmenting customer loyalty and

retention by delivering pertinent and timely

information, promotions, and suggestions

(Khennouche et al., 2024). Additionally, the

organization aims to attract new clientele and enlarge

its market presence by effectively communicating its

unique value proposition, domain expertise, and

distinctive competencies through generative AI.

Lastly, HealthTrust Europe seeks generative AI to

bolster its reputation and credibility by exemplifying

a steadfast commitment to quality, innovation, and

environmental sustainability (Archer, 2021).

2.3 Novel Technologies

2.3.1 Office 365 Co-Pilot

Given the existing utilization of Office 365 within

HealthTrust Europe's technological framework, the

integration of Office 365 Co-Pilot emerges as the

favored alternative. Office 365 Co-Pilot, a novel

enhancement, leverages Natural Language

Processing (NLP) and transformers (Hadi et al.,

2023b). The core of NLP and transformers lies in their

capacity to enable computational systems to

comprehend and produce human language,

encompassing both text and speech. Co-Pilot employs

these technologies to scrutinize the user's input and

contextual cues, generating pertinent and constructive

recommendations. These include, but are not limited

to, sentence completion, summary creation,

information retrieval, and query resolution.

Furthermore, Co-Pilot is designed to evolve by

assimilating user feedback and preferences, tailoring

its functionality to align with individual user styles

and requirements. A comprehensive evaluation is

imperative to cement this choice (Balk et al., 2021).

Therefore, this study focuses on harnessing

generative AI technologies to optimize healthcare

procurement processes.

2.3.2 Novel Generative AI Technologies

The generative AI sector is rapidly evolving, driven

by significant advancements and investments from

leading tech companies such as Google, Microsoft,

and Amazon. Notably, Large Language Models

(LLMs) like Microsoft 365 Copilot, Google Bard, and

OpenAI's GPT-4 are at the forefront, offering

capabilities that range from code generation to

multilingual customer support and marketing content

creation (Hadi et al., 2023a). Furthermore, tools such

as Amazon Bedrock and Google's Vertex AI Platform

democratize AI access and enhance project

management for professionals. Meanwhile, the trend

towards using LLMs trained on an organization's data

aims to produce content that reflects its unique brand

identity and meets the nuanced demands of its

audience (Budhwar et al., 2023). Consequently, this

dynamic landscape is shaped by tech giants and

invigorated by innovative AI startups like Anthropic,

signaling a vibrant and expanding field (Pasquero &

Poletto, 2023). Thus, the generative AI industry is

innovative and emerging in the world

(Badrinarayanan et al., 2017).

2.4 Generative AI Model

2.4.1 Encoder-Decoder Architecture

The encoder is a neural network that encodes the

input prompt into a vector representation in the latent

space. The latent space is a high-dimensional space

that captures the semantic and syntactic features of

the input prompt and the data (Zermatten et al., 2023).

The decoder is another neural network that decodes

the vector representation into the output text. The

output text is the generated content that matches the

input prompt and the data. The diagram below shows

the basic structure of the generative AI model, which

is based on an encoder-decoder architecture.

Figure 1: Basic structure of the generative AI model.

Input Encoder State Decoder Output

Generative AI Risk Management in Digital Economy

121

2.4.2 Generative AI: End-to-End Process

The data acquisition methodology involves gathering

diverse sources such as procurement contracts,

business and market intelligence, and other relevant

data. This data is then pre-processed and cleansed to

meet the generative AI model's requirements,

ensuring its compatibility and effectiveness

(Mourtzis, 2021).

Central to the solution is the generative AI model,

which is capable of producing textual content based

on the processed data and input prompts. This content

undergoes thorough evaluation and refinement by

human editors or automated systems to maintain

quality and relevance (Bandi et al., 2023). The

dissemination phase involves distributing the

polished content to targeted audiences via various

channels, including email, social media, and

websites, to maximize reach and engagement. This

process encapsulates the core components and steps

of the comprehensive solution (Nagy et al., 2023).

Figure 2: End-to-End Process AI model.

This paper showed that generative AI has many

applications; Health Trust Europe organizations

adopted generative AI to create customized and

engaging marketing and communications materials

for healthcare procurement organizations. In addition,

this study also summarized novel technology and

generative AI models, which help researchers explore

risks and mitigations.

3 METHODS

Our research methodology adopted the qualitative

research method; we observed that Health Trust

Europe acknowledges that the generative AI solution

presents significant dangers to digital and cyber

privacy and security and designed a full Risk

Assessment matrix for analysis and result mitigation.

Moreover, this study's philosophy adopts social

action theory. Social action theory focuses on how

people take actions that hold personal significance

and how these interactions impact societal norms

(Coleman, 1986). This study also identifies and

analyses novel technology risks and mitigations to

consider best practice approaches to maintain

network security. Additionally, interpretivism theory,

as the research approach in this study, interpretive

approach delves into understanding events in society

based on the specific value system of the society

(Ryan, 2018). This study interpreted generative AI

risk and mitigations to contribute to a risk-based

approach. The study applied an inductive approach

across the whole study; we analyzed generative AI

risk and management using variables to make risk

assessments for analysis and get results.

3.1 Definition of Variables

HealthTrust Europe is aware of the significant risks

of generative AI, including digital and cyber privacy,

security, reputational, financial, regulatory, ethical,

and legal challenges. In utilizing generative AI for

external communications and marketing, the

organization may address various risk categories as

variables:

Table 1: Full Risk Assessment matrix.

Potential Consequences

L6 L5 L4 L3 L2

Not

nificant

Minor Moderate Major Severe

Likelihood

Expected to occur

regularly under normal

Circumstances

Almost

Certain

5 Moderate 10 Major 15 Major 20 Severe

25 Very

Severe

Expected to occur at

some time

Likely

4 Moderate 8 Moderate 12 Major 16 Major 20 Severe

It may occur at some

time.

Possible

3 Minor 6 Moderate 9 Moderate 12 Major 15 Major

It is not likely to occur in

normal circumstances.

Unlikely

2 Minor 4 Moderate 6 Moderate 8 Moderate 10 Major

Could happen but

robabl

never will

Rare 1 Minor 2 Minor 3 Minor 4 Moderate 5 Moderate

Define

Objecti

Gather and

Pre Process

Data

Choose

appropri

ate

model

Train

the

Model

Evaluat

e and

Refine

Test and

Validate

Deploy

and

Iterate

FEMIB 2024 - 6th International Conference on Finance, Economics, Management and IT Business

122

Table 2: Organization's risk assessment result.

Step Risk Likelihood Impact Risk Level

Generic - Conduct a

thorough risk assessment

and impact analysis

Incomplete or inaccurate identification and

assessment of risks and impacts

Possible (3) Medium (3) Moderate (9)

Generic - Establish clear

policies, guidelines, and

est practice

Non-compliance or inconsistency with

relevant laws, regulations, standards, and

guidelines

Possible (3) Severe (5) Major(15)

Generic - Manage the

project budget, timeline,

and scope

The project exceeds the budget due to

unexpected costs of data acquisition, model

development, or content generation

Possible( 3) Moderate (3) Moderate (9)

Generic - Manage the

project budget, timeline,

and sco

The project is delayed due to technical

issues, stakeholder feedback, or content

roval

Likely (4) Major (4) Major (16)

Generic - Engage with

customers, suppliers, and

other stakeholders.

Lack of trust, confidence, or consent from

the customers, suppliers, or other

stakeholders due to insufficient or

ineffective communication or engagement

Likely (4) Severe (5) Severe (20)

Generic - Manage the

project budget, timeline,

and sco

The project scope changes due to new

requirements, features, or content types

Possible (3) Medium (3) Moderate (9)

Technology - Incorporate

human oversight and

intervention

Human bias, error, or manipulation

affecting the content quality, accuracy,

authenticit

, and accountabilit

Possible (3) Medium (3) Moderate (9)

Technology- Ensure the

project quality

The project quality is compromised due to

poor data, model, or output quality and

reliabilit

Likely (4) Severe (5) Severe (20)

Technology- Ensure the

data quality and reliability

The data is incomplete, inaccurate,

outdated, biased, or irrelevant, affecting the

quality and reliability of the generated

content

Likely (4) Severe (5) Severe (20)

Technology - Ensure the

model's complexity and

explainability

The model is complex and difficult to

understand, explain, or debug, posing

challenges for the development,

deployment, and maintenance of the

enerative AI technolo

Possible (3) Severe (5)

Major

(15)

Technology - Ensure the

model scalability and

performance

The model requires high computational

resources and costs, hindering the

scalability and performance of the

enerative AI technolo

Possible (3) Severe (5)

Major

(15)

Technology - Ensure the

output ethics and social

impact

The output is misused, manipulated, or

stolen, causing legal and security problems,

such as plagiarism, fraud, identity theft,

etc.

Unlikely (2) Severe (5) Major (10)

Cyber - Implement robust

technical and

organisational measures

Data breach, loss, or corruption due to

cyberattacks, human errors, or system

failures technical and organisational

measures

Possible (3) Severe (5)

Major

(15)

Cyber - Ensure data

protection and privacy

The data is breached, leaked, or accessed

by unauthorised parties, violating data

rotection and privacy laws and regulations

Unlikely (2) Severe (5) Major (10)

Cyber - Ensure the model's

integrity and functionality

The model is hacked, corrupted, or

tampered with, compromising the integrity

and functionalit

of the

enerative AI

Unlikely (2) Severe (5) Major (10)

Cyber - Ensure the output

authenticity and

accountabilit

The output is spoofed, impersonated, or

falsified, deceiving or harming the target

audience or stakeholders

Possible (3) Severe (5)

Major

(

)

Generative AI Risk Management in Digital Economy

123

Generic Risks: Common to all projects,

including budget, schedule, scope, and quality

concerns.

Technology-Specific Risks: Unique to

generative AI, such as data integrity, model

reliability, and output accuracy.

Cyber Risks: Related to the security and privacy

of AI technology, including threats like hacking,

phishing, and spoofing.

3.2 Risk Assessment Matrix

A full Risk Assessment matrix has been attached in

Table 1. This study explores the risks, feasibility, and

benefits of developing a pilot scheme in likelihood

form. These investigations have potential

consequences from L2 to L6. According to the

organization, various risk categories, Generic,

Technology-specific, and cyber risks, analyze

likelihood, impact, and risk level to get results.

4 RESULT

According to the organization's various risk variables:

Generic risks, Technology-specific risks and cyber

risks, this study created several criteria like

likelihood, impact and risk level to get results.

Through the whole risk assessment matrix to get

results. Pertinent risks associated with content

generation and security through generative AI

encompass:

Data Privacy and Security: The necessity for

generative AI solutions to access sensitive data

mandates stringent protective measures, including

encryption and access controls, to mitigate data

breaches and misuse risks in adherence to regulations

like GDPR.

Content Quality and Accuracy: There exists a

potential for generative AI to yield content that may

be inaccurate or misleading. It is imperative for

organizations to implement rigorous verification and

editorial processes, supplemented by human

oversight, to ensure content integrity and address any

resultant feedback or grievances.

Content Authenticity and Accountability: The

indistinguishability of AI-generated content from

human-produced content necessitates clear labeling

and attribution to the originating organization,

alongside verification methods such as digital

signatures, to uphold content authenticity.

Content Ethics and Legality: Generating

content that could be deemed unethical or illegal

poses significant concerns. Organizations may ensure

that AI-generated content is congruent with their

ethical standards and complies with applicable legal

frameworks and industry

5 DISCUSSION

In light of the risks identified, adopting a

comprehensive and forward-looking strategy for the

ethical and responsible deployment of generative AI,

as proposed in the detailed Risk Mitigation Report

found in Appendix 1, emerges as a prudent approach.

This strategy advocates for a thorough risk

assessment, engaging various stakeholders to explore

potential risks and their implications rigorously. It

echoes the holistic perspectives emphasized in the

academic literature reviewed. The formulation of

Policy and Governance Frameworks, rooted in the

principles of transparency, accountability, fairness,

and safety, is crucial, reflecting the scholarly

consensus on the ethical governance of AI

technologies (Birkstedt et al., 2023). Adopting

stringent Data Management and Security measures,

including data cleansing, anonymization, and

encryption, alongside robust authentication and

auditing protocols, aligns with the best practices

identified in the existing literature (Ooi et al., 2023),

underscoring the importance of safeguarding data

integrity.

The integration of Human Oversight through

approaches like human-on-the-loop or human-in-

command is vital for ensuring accountability and

maintaining content integrity, resonating with the

recommendations by Birkstedt et al., (2023) and

paralleling discussions in the literature on human-AI

collaboration. Utilizing both qualitative metrics,

continuous monitoring, and evaluation is imperative

for continually refining the AI system's effectiveness

and societal impact, mirroring the iterative evaluation

processes highlighted in the academic discourse.

Engaging Stakeholders in ongoing dialogues about

the AI solution's benefits, potential risks, and

limitations is essential for establishing a foundation

of transparency and trust, reflecting the existing

literature's emphasis on stakeholder inclusivity in AI

deployments (Khennouche et al., 2024). This

multifaceted strategy aligns with the academic

literature. It encourages a more measured and

nuanced interpretation of the findings, avoiding

definitive judgments and fostering a balanced

understanding of AI's potential and challenges in

organizational contexts.

FEMIB 2024 - 6th International Conference on Finance, Economics, Management and IT Business

124

6 RECOMMENDATION

Based on the existing literature and frameworks, I

have reviewed best practice approaches/models for

generative AI risk analysis and selected two for this

project:

6.1 The NIST AI Risk Management

Framework (AI RMF)

This is a voluntary set of standards, guidelines, and

best practices to help organizations manage

cybersecurity risks. The AI RMF can help

HealthTrust Europe design, develop, use, and

evaluate trustworthy and responsible AI solutions by

following the four core functions: Govern, map,

measure, and manage. It can help identify, protect,

detect, respond to, and recover from cyber threats that

may affect the generative AI solution, such as data

breaches, adversarial attacks, or system failures

(Tabassi, 2023).

6.2 AI Trust Risk and Management

Framework (AI TRiSM)

Gartner defines AI TRiSM as “a framework that

supports AI model governance, trustworthiness,

fairness, reliability, robustness, efficacy, and data

protection.” (Gartner, 2016)

The AI TRiSM Index covers six dimensions of

trust: transparency, reliability, usability, security,

ethics, and robustness. The model can help

HealthTrust Europe to identify and address the key

factors that influence the trustworthiness of its

generative AI solution, such as the AI system itself,

the data, the environment, the stakeholders, and the

governance. These factors are interrelated and affect

each other so that they may be considered holistically

and dynamically.

HealthTrust Europe can map out the solution's

architecture, design, and functionality, such as the AI

model, the data pipeline, the user interface, and the

output generation. It can also identify the data

sources, data types, data quality, data processing, and

data storage for its generative AI solution, as well as

the use cases, scenarios, and user groups for its

generative AI solution, such as marketing,

communication, customer service, and supplier

management.

Utilizing both NIST and AI TRiSM for generative

AI risk analysis offers distinct advantages:

Complementary Frameworks: NIST's

comprehensive risk management process synergizes

with AI TRiSM's succinct trustworthiness

framework, offering varied perspectives and

granularity.

Holistic Analysis: Together, they foster a

balanced evaluation of generative AI, weighing

potential impacts on diverse stakeholders and societal

implications.

Standardized Communication: These

frameworks facilitate transparent, consistent

reporting on AI risks and trustworthiness, employing

a unified language and structure.

Stakeholder Engagement: NIST and AI TRiSM

enhance the clarity and accessibility of generative AI

risk analyses for a broad audience, including

developers, users, regulators, and the general public.

6.3 Best Practice Approaches

The following scenarios are examples of how both

Best Practice approaches can be used to support the

analysis and mitigation of identified risks associated

with the proposed solution:

In the first scenario, the National Institute of

Standards and Technology (NIST) framework is

employed to meticulously guide the identification,

assessment, and management of potential technical,

ethical, legal, and social risks associated with the AI

solution, including challenges related to data quality,

privacy, model explainability, transparency,

cybersecurity, adversarial attacks, and issues

stemming from human bias, error, or manipulation.

Concurrently, the AI Trust, Risk, and Security

Management (TRiSM) Index is instrumental in

measuring and enhancing the solution's

trustworthiness by utilizing indicators and metrics

focused on transparency, reliability, usability,

security, ethics, and robustness.

In the second scenario, the creation of external

articles and social media posts is envisioned as a

means to communicate and report the performance

and impact of the AI solution to a diverse array of

stakeholders, including regulators, investors,

partners, and the general public. Utilizing the NIST

framework, the generative AI risk analysis's rationale,

methodologies, and outcomes are meticulously

documented and disclosed, ensuring that the

addressed potential risks and impacts are

transparently conveyed. Furthermore, the AI TRiSM

framework is leveraged to provide clear, accessible

information regarding the source and nature of the

generative AI content, alongside evidence and

verification mechanisms. Employing both

frameworks consistently and transparently to

communicate and report on the generative AI risk

Generative AI Risk Management in Digital Economy

125

analysis significantly contributes to building trust,

confidence, and a positive reputation for the

organization and its solution.

In the third scenario, the evaluation of generative

AI solutions for crafting presentations and emails

involves a comparative analysis to select the most

superior options based on quality, reliability, and

trustworthiness. The NIST framework is utilized to

conduct a thorough and systematic risk assessment

for each potential solution, evaluating its efficacy in

addressing anticipated risks and impacts.

Additionally, the AI TRiSM framework is applied to

compare and rank the alternatives based on their

trustworthiness scores, enabling the identification of

each option's strengths and weaknesses, thereby

facilitating a more informed decision-making

process.

7 CONCLUSION

This research elucidates the substantial opportunities

and advantages of generative AI in marketing and

community engagement for HealthTrust Europe,

alongside a set of strategic recommendations to

navigate and alleviate the inherent risks. The study

outlines a comprehensive AI governance strategy by

adopting frameworks like the NIST AI RMF and AI

TRiSM model, emphasizing collaboration with legal

and data protection entities to define clear roles,

responsibilities, and risk boundaries. This includes a

detailed examination of the AI system's structure, risk

evaluation, and performance measurement against

predefined metrics, coupled with implementing risk

mitigation tactics to ensure data integrity and uphold

the principles of transparency, fairness, and

accountability.

However, the study acknowledges certain

limitations, such as the potential for evolving

technological landscapes to outpace current

governance frameworks, and the challenge of fully

anticipating the social implications of generative AI.

Future research directions should focus on dynamic

governance models that can adapt to technological

advancements, and deeper inquiries into the long-

term societal impacts of AI integration, ensuring that

the organization's commitment to social

responsibility remains at the forefront of its

technological adoption strategy.

ACKNOWLEDGMENT

This research is partly supported by VC Research

(VCR 0000230) for Prof Chang.

REFERENCES

Archer, M. S. (2021). The mess we are in: How the

Morphogenetic Approach helps to explain it: IACR

2020 Warsaw. Journal of Critical Realism, 20(4), 330–

348. https://doi.org/10.1080/14767430.2022.1984698

Badrinarayanan, V., Kendall, A., & Cipolla, R. (2017).

Segnet: A deep convolutional encoder-decoder

architecture for image segmentation. IEEE

Transactions on Pattern Analysis and Machine

Intelligence, 39(12), 2481–2495.

Balk, A. D., Hernandez, T. R., Lennan, K. J., Martinez, C.

E., Olbricht, N. M., & Wimberly, B. D. (2021).

Taxonomy of Situational Awareness Information for the

Future Long-Range Assault Aircraft (FLRAA) Medical

Evacuation (MEDEVAC) Co-Pilot. https://apps.dtic.

mil/sti/citations/trecms/AD1164209

Bandi, A., Adapa, P. V. S. R., & Kuchi, Y. E. V. P. K.

(2023). The power of generative ai: A review of

requirements, models, input–output formats, evaluation

metrics, and challenges. Future Internet, 15(8), 260.

Birkstedt, T., Minkkinen, M., Tandon, A., & Mäntymäki,

M. (2023). AI governance: Themes, knowledge gaps

and future agendas. Internet Research, 33(7), 133–167.

Budhwar, P., Chowdhury, S., Wood, G., Aguinis, H.,

Bamber, G. J., Beltran, J. R., Boselie, P., Lee Cooke, F.,

Decker, S., DeNisi, A., Dey, P. K., Guest, D., Knoblich,

A. J., Malik, A., Paauwe, J., Papagiannidis, S., Patel, C.,

Pereira, V., Ren, S., … Varma, A. (2023). Human

resource management in the age of generative artificial

intelligence: Perspectives and research directions on

ChatGPT. Human Resource Management Journal,

33(3), 606–659. https://doi.org/10.1111/1748-8583.12

524

Coleman, J. S. (1986). Social Theory, Social Research, and

a Theory of Action. American Journal of Sociology,

91(6), 1309–1335. https://doi.org/10.1086/228423

Gartner, W. B. (2016). Entrepreneurship as organizing:

Selected papers of William B. Gartner. Edward Elgar

Publishing. https://books.google.co.uk/books?hl=zh-

CN&lr=&id=A3ONCwAAQBAJ&oi=fnd&pg=PR1&

dq=(Gartner&ots=KWc8LhO6Bu&sig=OTf7OTJ2i6y

40lBChqxUpPW47X4

Hadi, M. U., Tashi, Q. A., Qureshi, R., Shah, A., Muneer,

A., Irfan, M., Zafar, A., Shaikh, M. B., Akhtar, N., Wu,

J., & Mirjalili, S. (2023a). A Survey on Large Language

Models: Applications, Challenges, Limitations, and

Practical Usage [Preprint]. https://doi.org/10.36227

/techrxiv.23589741.v1

Hadi, M. U., Tashi, Q. A., Qureshi, R., Shah, A., Muneer,

A., Irfan, M., Zafar, A., Shaikh, M. B., Akhtar, N., Wu,

J., & Mirjalili, S. (2023b). Large Language Models: A

Comprehensive Survey of its Applications, Challenges,

FEMIB 2024 - 6th International Conference on Finance, Economics, Management and IT Business

126

Limitations, and Future Prospects [Preprint].

https://doi.org/10.36227/techrxiv.23589741.v4

Khennouche, F., Elmir, Y., Himeur, Y., Djebari, N., &

Amira, A. (2024). Revolutionizing generative pre-

traineds: Insights and challenges in deploying ChatGPT

and generative chatbots for FAQs. Expert Systems with

Applications, 246, 123224.

Lamarre, E., Smaje, K., & Zemmel, R. (2023). Rewired:

The McKinsey guide to outcompeting in the age of

digital and AI. John Wiley & Sons. https://books.

google.co.uk/books?hl=zh-CN&lr=&id=DjDFEAAA

QBAJ&oi=fnd&pg=PA1&dq=(McKinsey,+2023)&ots

=YHxpIC13Jf&sig=MNhupEewOtBtlos86d3Ohfip51s

Mourtzis, D. (2021). Towards the 5th industrial revolution:

A literature review and a framework for process

optimization based on big data analytics and semantics.

Journal of Machine Engineering, 21(3).

https://bibliotekanauki.pl/articles/1833773.pdf

Nagy, P., Frey, S., Sapora, S., Li, K., Calinescu, A., Zohren,

S., & Foerster, J. (2023). Generative AI for End-to-End

Limit Order Book Modelling: A Token-Level

Autoregressive Generative Model of Message Flow

Using a Deep State Space Network. 4th ACM

International Conference on AI in Finance, 91–99.

https://doi.org/10.1145/3604237.3626898

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. Journal of Computer Information

Systems, 1–32. https://doi.org/10.1080/08874417.2023

.2261010

Pasquero, C., & Poletto, M. (2023). Biodesign in the Age of

Artificial Intelligence: Deep Green. Taylor & Francis.

https://books.google.co.uk/books?hl=zh-

CN&lr=&id=2sXDEAAAQBAJ&oi=fnd&pg=PT8&d

q=this+dynamic+landscape+is+shaped+by+tech+giant

s+and+invigorated+by+innovative+AI+startups+like+

Anthropic,+signalling+a+vibrant+and+expanding+fiel

d.&ots=eql9u1-y7S&sig=vE88-

MbQY5AMZgp60Ak1ItbHXVg

Preiksaitis, C., & Rose, C. (2023). Opportunities,

challenges, and future directions of generative artificial

intelligence in medical education: Scoping review.

JMIR Medical Education, 9, e48785.

Ryan, G. (2018). Introduction to positivism, interpretivism

and critical theory. Nurse Researcher, 25(4), 41–49.

Tabassi, E. (2023). Artificial Intelligence Risk Management

Framework (AI RMF 1.0). https://www.nist.gov/

publications/artificial-intelligence-risk-management-

framework-ai-rmf-10?trk=public_post_comment-text

Zermatten, V., Navarro, J. C., Hughes, L., Kellenberger, T.,

& Tuia, D. (2023). Text as a richer source of

supervision in semantic segmentation tasks. IGARSS

2023-2023 IEEE International Geoscience and Remote

Sensing Symposium, 2219–2222. https://ieeexplore.

ieee.org/abstract/document/10282398/?casa_token=rI

wWKhBvi5YAAAAA:GT1viBA2JPi8Kg2K1b4NXU

-9g_fr0DnG_QwSNbUTsEB-eMAaOp2Y5xBOG6Nu

g40khhrgLNMw7A

Generative AI Risk Management in Digital Economy

127