Making tacit knowledge explicit: Generative AI's role in enhancing apprenticeship systems

1.1. Background and motivation

In today's twenty-first-century environment, knowledge has become a critical asset for business organizations as well as academic institutions, playing a significant role in the improvement of sustainability, resilience, and competitive edge. Of the various forms of knowledge, TK is unique in its close ties to the experiential, contextual, and personal dimensions. TK primarily exists as skills, experiential understanding, cognitive frameworks, preconceptions, and beliefs that cannot be easily expressed or conveyed through formal means. In contrast to EK, TK is not codifiable or transmissible, due to its inherent intangible and particularistic characteristics, which pose challenges in situations in which knowledge needs to be transmitted through experience and action-oriented mechanisms, for example, in apprenticeship-type situations. ¹ Given its immense importance, Knowledge Management (KM) is also at the forefront of improving organizational performance, productivity enhancement, and the accumulation of financial capital. The efficient functioning of a KM system cultivates stakeholder relationships and supports the ultimate sustainability of an organization. The successful installation of a KM framework in line with quality assurance standards at University Y laid the foundation for an integration extracted from Joint Application Development (JAD), Knowledge Creating Company, and Intellectual Capital (IC), which in turn guides the organization, successfully installing the KM practice, hence achieving sustainable benefits in the form of competitive advantage and organizational progress. ²

The classic model of apprenticeships depends heavily on the transfer of TK through the (master-apprentice) relationship. It has been effective in developing practical skills, problem-solving abilities, and adaptable minds in professions varying from manufacturing to healthcare. However, while competing with the threats of globalization and innovation and against the new realities of the work environment, the constraints of the traditional model emerge. The inadequate effectiveness and limited scalability of individual mentorship often hamper the ability to meet the rapidly increasing global demand for skilled and qualified workers in the areas of contemporary industrial life. ³ The loss of TK because of employee turnover, retirements, or project-based transience presents a major challenge to the organization. Although TK is recognized as a key enabler for innovation and organizational effectiveness, the process through which it is transmitted tends to be localized and elusive. The paradox has led to a realization at the grassroots level that the organization has to evolve strategies to make TK explicit in order to preserve and disseminate it to wider audiences. ⁴ Generative AI can address this challenge by enabling the conversion of TK into EK. With advanced technologies in the domain of Natural Language Processing (NLP), Machine Learning (ML), and multimodal data interpretation, Generative AI can assist in codifying, structuring, and representing TK in forms that are accessible and shareable. The ability of AI tools to take account of complex information with a context-sensitivity orientation drawn from diverse sources, such as expert performance, video-based explanatory sequences, or real-time workplace mentorship, and to repackage such knowledge in the form of an organized, explicit repository guarantees safeguarding it, as well as sharing it for the purposes of wider learning and enhanced innovation. ⁵ Furthermore, generative AI has the potential to improve apprenticeship programs by increasing the efficiency and effectiveness of knowledge transfer processes. AI-based platforms can offer customized learning experiences that cater to the specific requirements of individuals, giving timely feedback and presenting resources that are specifically designed to assist each learner along their individual development path. The integration of AI in apprenticeship models enables the blending of classic EL approaches and modern KM processes, thus improving learning results and performance measures. ⁶ TK is extremely valuable in various fields, most importantly in knowledge gathering, human resource management, and social psychology. TK plays a critical role in the creation of organizational innovation in fields like live streaming sales, real estate, and tech progress. With organizations relying heavily on KM systems to attain a competitive edge, the use of AI becomes instrumental in the conversion of TK into explicit forms for easy storage, retrieval, and sharing. While addressing issues involving knowledge loss and knowledge gathering inefficiencies, the use of Generative AI becomes one innovative response for organizations to harness their IC and drive sustained growth. ⁷

1.2. Related works

TK, defined as the understanding gained by people within EL contexts, holds great significance in improving organizational effectiveness, creativity, and innovation. The transfer of TK faces great challenges within EL contexts, a fact exemplified through apprenticeship schemes. Recent studies outline many mechanisms that aim to promote the transfer of TK, highlighting the contributions of generative AI, KM systems, and various pedagogical models. Owens ⁸ presented the Synthetic Training Environment Experiential Learning for Readiness Program, or STEEL-R, as a pedagogical model that aims to improve technology-enhanced military training by complementing technological and social knowledge for trainees. Although many pedagogical strategies that focus on the dissemination of explicit technological knowledge dominate the literature, the STEEL-R pedagogical model highlights the transfer of TK as a key feature of improving individual and group performance in real-world settings. This new pedagogical system emphasizes the importance of EL and offers proof of the applicability of AI in personalizing and accelerating the transfer of TK, culminating in positive results for future groups of learners.

In International Joint Ventures (IJVs), Park et al. ⁹ performed an analysis concerning the effects of tacit and EK transfers on the innovative capacity of an organization. The findings indicated that the transfer of EK had a positive impact on innovation, whereas TK did not exhibit a direct connection. The research also revealed that EK serves as a mediator in the TK-innovation relationship, particularly when TK originates from international parent organizations. This study underscores the nuanced interplay in KT in international contexts and adds to the ongoing discussion around the interplay of tacit and EK and their role in promoting innovation. On the contrary, in Baláž et al., ¹⁰ the competencies related to international migration and the importance of tacit and EK in these competencies were investigated. The findings revealed that Slovak migrants placed great emphasis on TK, primarily with respect to individual characteristics such as self-confidence and problem-solving skills. Furthermore, the study distinguished between different forms of TK, such as embodied, embedded, encultured, and enculturated forms, attributed mainly to technical aspects. The categorization clarifies the dynamics of knowledge and its root in fostering progress and versatility, thus having a significant impact on the learning process and organizational structures.

Silva et al. ¹¹ suggested an EK creation and organization model. This model clarifies knowledge into situated and theoretical/normative types. It also examines EK production and documentation strategies. The model can improve the design of high-end KM systems, which organize, spread, and convert TK into explicit forms. Current initiatives aim to convert TK best through KM innovations using AI. Duan et al. ¹² examined how EK and TK concealment affect organizational innovation quality. The results showed that moderate knowledge hiding could initially support innovation, but excessive knowledge hiding would be counterproductive. They also noted that correct knowledge flow within an organization prevents knowledge hiding and promotes innovation. This research emphasizes the importance of organizations’ ability to negotiate the exchange and distribution of tacit and EK to prevent valuable information from being lost and used for innovation and organizational development.

These studies demonstrate the complex relationship between tacit and EK transfer and the importance of generative AI and KM systems in enhancing knowledge flow and organizational performance. AI-based tools can simplify TK transfer, foster innovation, and sustain growth in diverse organizations.

1.3. Research gaps and contribution

The research on TK transfer and the integration of generative AI in ASs highlights several gaps and contributions. While existing studies, such as the STEEL-R project, emphasize the importance of TK transfer in military training, they lack applicability to industries like live-streaming sales, where dynamic, real-time learning is crucial. Additionally, while generative AI has shown promise in converting TK into EK, much of the literature focuses on its technical capabilities, leaving a gap in understanding how AI can be effectively integrated into fast-paced, hands-on environments. Furthermore, the complexity of capturing subjective, context-dependent TK through AI remains a challenge, particularly when dealing with multimodal data integration. There is also limited exploration of the emergent behavior in large AI models and how this can be leveraged to enhance ASs. This study contributes by examining the role of generative AI in transforming TK, particularly in live-streaming sales, and highlights its potential to enhance learning efficiency, scalability, and knowledge retention across industries reliant on apprenticeship-based training.

1.4. Paper structure

This paper is structured as follows: Section 2 outlines the methodology, describing the approach to knowledge transmission and the AI-driven models used to convert TK into EK. Section 3 outlines the results and discussion, including statistical analysis of competency levels, the impact of generative AI on apprenticeship skills, and the limitations faced in the research and its pragmatic implications. This part addresses the issues of making TK explicit and extrapolates the lessons learned from this towards larger organizational learning and KM undertakings. The fourth chapter also includes a summarizing conclusion, summarizing the most essential findings, and considering the present and future relevance of GenAI in apprenticeships and its impact on the future development of the workforce.

2.1. Approach to knowledge transmission

The expressiveness and communicability of EK and TK are what set them apart from each other. EK is contained in documents, books, and other structured formats; it can be shared, replicated, and communicated. Due to its clearness and visibility, the knowledge can be learned and used in different settings. On the other hand, TK is subjective, intangible, and hard to express. Personal experiential knowledge, intuitive judgments, and know-how are acquired through direct experience. Because of its unstructured status, TK is hard to document; effective transfer is separated into EK and TK according to expressiveness and communicability. EK is contained in documents, books, and other structured formats; it can be shared, replicated, and communicated. Due to its clearness and visibility, the knowledge can be learned and used in different settings. On the other hand, TK is subjective, intangible, and hard to express. Personal experiential knowledge, intuitive judgments, and know-how are acquired through direct experience. Because of its unstructured status, TK is hard to document; effective transfer is separated into EK and TK according to expressiveness and communicability. ¹³

TK is the foundation of EK, and transferring TK to explicit forms is the cornerstone of fostering organizational learning and innovation. Almeida Arruda et al. ¹⁴ suggested that TK may be divided into practice-oriented, cognition-oriented, and relationship-oriented TK. These typologies offer a more refined view of TK, which aids its implicitization through generative AI, among others. The transformation of TK into EK helps to increase organizational effectiveness and improve knowledge retention. Generative AI can contribute to this transformation by structuring and explaining TK into a form that can be shared and understood. AI applications help bridge the gap between intuitive, experiential knowledge and formalized knowledge, hence maximizing KT and its use.

The rapid development of generative AI technologies has triggered researchers and practitioners to explore their potential to transform TK into explicit forms, improve apprenticeship programs, and enable the mass dissemination of knowledge and innovation to various fields. TK, with its intangible nature and intuitive understanding, represents a major store of knowing in many professional fields, such as the skillful techniques, instincts, and heuristics practiced by expert practitioners. Although the substantial value placed on this kind of knowledge has frequently been hampered by the inherent challenges surrounding documentation, generative AI has the potential to aid in the externalization of TK, thus converting it into explicit, shareable, and knowable forms. This potential not only mitigates the challenges to the storage and reconversion of knowledge but also supports “training-business integration” through the establishment of learning settings physically integral to real business processes. Such an integration mitigates the challenges of “low learning transfer” and ensures effective training while ensuring alignment with organizational goals. ⁵

Within traditional apprenticeship models, generative AI takes on a twofold role, both facilitating and accelerating the process of KT. By having the ability to systematically classify and analyze the explicit and implicit knowledge of the mentor, AI presents itself as an important factor in understanding previously unidentified knowledge gaps. AI, for example, can help mentors analyze key work processes and create visual models of complex processes, thus reducing complexity and uncertainty in their areas of specialization. AI systems can also provide unique insights and suggestions for improvement by analyzing and profiling the mentor's expertise, which leads to more innovative pedagogical strategies. In industries with extensive TK, like live-streaming, generative AI can analyze and process multimodal information autonomously, structuring and organizing this information into a unified framework. This can then be integrated with an organizational framework for enabling KT, using knowledge graphs and intelligent decision networks as enabling tools for global collaboration and limitless knowledge sharing.

Generative AI is capable of autonomously collecting various forms of data as real-time live feeds: structured or unstructured text, images, video, and audio. The generative AI constructs a multimodal common representation space where the data produced from heterogeneous datasets are represented in a common vector space by merging the data from different modalities (Figure 1). The combination of these diverse datasets allows for end-to-end analysis in addition to knowledge extraction, enabling the determination of patterns and inferences that can dramatically enhance the decision-making process. The AI excels in pinpointing the most effective product presentations or recognizing specific words that play a crucial role in driving sales. Through the application of deep learning models and ML algorithms, the AI discovers hidden data patterns, delivering critical insights that support enhancements in sales tactics.

OPEN IN VIEWER

Figure 1.

Multi-modal representation learning.

Moreover, AI offers personalized learning experiences by analyzing users’ behavior and interests, recommending relevant resources, exercises, and adaptive learning paths. Traditional methods of constructing user profiles based on static labels are enhanced through AI's ability to process natural language and extract dynamic features from text data. This allows for the construction of more accurate and personalized learner profiles (as illustrated in Figure 2). Additionally, AI-powered systems can dynamically adjust the learning content and teaching methods based on real-time feedback and performance, ensuring that each learner's unique needs are met. Interactive guidance through virtual assistants simulates real-life scenarios, empowering apprentices to explore and solve problems autonomously, further reinforcing the concept of personalized and adaptive learning systems.

OPEN IN VIEWER

Figure 2.

AI personalized learning system.

Generative AI extends its role beyond individual KT by integrating TK across entire enterprises, facilitating the construction of a comprehensive organizational knowledge base. This knowledge-sharing platform supports the import and management of various document formats, including text, images, and code snippets, ensuring that organizational knowledge is easily accessible and efficiently managed. Powered by advanced search functions, such as Elastic Search and AI-driven document searches, users can quickly locate and access the information they need. Additionally, AI tools such as ChatGPT provide intelligent customer service support, extracting core information from documents and summarizing it to optimize user interaction with the knowledge base (see Figure 3).

OPEN IN VIEWER

Figure 3.

Intelligent knowledge-sharing platform.

In the context of the digital economy and internet marketing, particularly within the growing field of live-streaming sales, generative AI facilitates the implicitization and efficient transmission of TK. Through multimodal data analysis, natural language generation, and behavioral modeling, AI enhances the externalization of expertise in areas such as on-camera presence, verbal persuasion techniques, audience interaction management, adaptive responsiveness, continuous learning mechanisms, personal branding, and conversion optimization. These advancements in AI-driven knowledge transmission have the potential to revolutionize training in industries where TK is paramount, making expertise more accessible, shareable, and actionable at scale.

The categorization of TK is essential for its implicitization and effective management. TK can be classified into three primary types: practice-based, cognition-based, and relationship-based TK, each requiring different techniques for transformation into EK.

2.2.3. Relationship-based tacit knowledge

TK from interpersonal relations deals with understanding human interaction, organizational culture, and communication skills. This covers the knowledge concerning interpersonal relations, especially in building trust and communicating with stakeholders, and an understanding of organizational culture, including conforming to corporate values. Also included in this category is network knowledge, entailing the building up of professional networks as well as social networks. This type of TK is deeply grounded in personal experience and continuous practice, making it extremely difficult to verbalize and communicate through traditional modes.

As illustrated in Figure 4, generative AI plays a central role in the process of implicitization related to the described types of TK. The technology can autonomously structure and build knowledge graphs based on real-time sales data, thus allowing for the integration of knowledge into an organized and formalized structure. Additionally, AI can automatically create individualized knowledge graphs adapted for specific learners, considering their personal learning experience, skill level, and subjects. This feature can contribute to the development of personalized learning trajectories and increase the efficiency of apprenticeship-type educational models. Serving as an active mentor, the AI system skillfully suggests relevant study materials and tutorial videos. By incorporating interactive learning games and instant Q&A support, the system provides timely feedback on students’ progress, offers targeted support for filling knowledge gaps, inspires motivation in learning, and develops students’ overall potential.

OPEN IN VIEWER

Figure 4.

TK graph in live sales.

AI can independently organize and construct knowledge graphs from available sales data, besides creating customized knowledge graphs reflecting the academic background, individual skill levels, and personal interests of every student, thus providing customized learning trajectories. As a committed educator, relevant study materials and tutorial videos are thoughtfully recommended. With the help of interactive learning exercises and real-time question-and-answer support, the system provides timely feedback on academic progress, provides personalized support while compensating for knowledge gaps, boosts student motivation and competencies, and encourages efficient apprenticeship-style learning.

Experimental design entailed the enrollment of 50 new live-streaming hosts, who were assigned randomly into two separate cohorts: one group was trained using technology-based methods with generative AI, while the other group was trained using traditional methods. The experimental period lasted three months; however, competency assessments were conducted weekly to assess the degree of effectiveness of technology-based training. Measures of effectiveness utilized included performance indicators among seven particular competency areas that included on-camera presence, verbal communication skills, audience engagement management, responsiveness, agility, learning adaptability, personal branding, and sales conversion rates.

The methodology used to gather data was holistic, involving surveys, behavioral tests, and analytical appraisals. Statistical testing of participant competency scores used independent samples t-tests, allowing the exploration of the effect of generative AI tools on training effectiveness delivered in apprenticeship environments. The research design allowed for an intensive exploration of AI applications in the development of multiple skills and competencies in live-streaming sales.

3.1. Statistical analysis of competency scores

To evaluate the effectiveness of generative AI tools in apprenticeship-based training, independent samples t-tests were conducted to compare the competency scores between the EG (AI-assisted training) and the CG (traditional training). The results, summarized in Table 2, demonstrate that the EG consistently outperformed the CG across all competency dimensions, with statistically significant differences observed in each case.

OPEN IN VIEWER

Table 1.

Classification and typical scenarios of TK.

Category	Forming reason	Subclass	Typical scenario	Explicit difficulty
Practice-based TK	Derived from long-term practice and accumulated experience	Operational Knowledge	Skilled workers diagnose faults by sound, and engineers adjust parameters based on intuition	Operational knowledge can be externalized through documentation, but diagnostic and adaptive knowledge require case analysis or simulation
		Diagnostic Knowledge	Identification and resolution of issues during tasks, based on experience	Diagnostic knowledge requires real-world scenarios or simulations
		Adaptive Knowledge	Ability to flexibly adjust to different scenarios	Highly tacit and reliant on individual judgment, requiring complex simulations
Cognition-based TK	Developed from personal observation, understanding, and thought	Intuitive Knowledge	Predicting market trends or technological risks based on intuition	Difficult to formalize due to reliance on personal experiences and heuristics
		Insightful Knowledge	Deep understanding of customer needs or technological advancements	Challenging to represent explicitly, relying on insights gained through experience
		Enlightening Knowledge	Stimulating new ideas or innovations through analogy or associations	Requires creative input, making it hard to document
Relationship-based TK	Involves understanding interpersonal relationships and organizational culture	Human Relationship Knowledge	Effective communication and trust-building in teams and customer interactions	Relational knowledge relies on experience and is difficult to document directly
		Organizational Culture Knowledge	Knowledge of corporate values, norms, and practices	Embedding this in explicit systems requires an in-depth, context-specific understanding
		Network Knowledge	Ability to access, integrate, and leverage resources through social and professional networks	Relies on long-term practice and experience, making it difficult to formalize

OPEN IN VIEWER

Table 2.

Significance levels of AI-assisted learning.

Competency dimension	Experimental group (mean score)	Control group (mean score)	t-tests	p-values
Shot expressiveness	85	70	4.23	<0.01
Verbal expression	88	72	4.56	<0.01
Site control capability	82	68	3.98	<0.01
Ability to make an immediate response	80	65	4.12	<0.01
Learning ability	86	71	4.35	<0.01
People build ability	84	69	4.05	<0.01
Promote single conversion ability	87	73	4.68	<0.01

The p-values for all competency dimensions were consistently below 0.01, indicating that the differences between the EG and CG were statistically significant. This strongly suggests that generative AI tools have a profound impact on enhancing various competencies required in apprenticeship training.

The gains witnessed in the competencies of the EG are of utmost importance as they provide strong evidence for the advantages gained from training supported by AI. The competency domains will be explored in detail in the subsequent sections:

Shot Expressiveness: The mean score for the experiment group was 85, compared to the CG's 70, giving a t-value of 4.23 and a p-value less than 0.01. The statistically significant difference here reveals that the technology-aided training has successfully enhanced the apprentices’ competencies in expressing shot sequences in live broadcasting.

3.3. Effectiveness evaluation of AI-assisted learning

The significant improvements observed across all competency dimensions underscore the effectiveness of AI tools in enhancing apprenticeship-based learning. ⁶ The findings suggest that AI-powered training environments can lead to higher proficiency in both technical and adaptive competencies. These results confirm the potential of generative AI to address the inefficiencies often observed in traditional ASs, particularly in capturing and transferring TK.

3.4. Small-scale fine-tuning and large-scale emergence of the AI model

Besides improving learning outcomes, the implementation of generative AI in the apprenticeship model has benefits related to adaptability and scalability. AI systems can be tailored to suit specific organizational needs, thus increasing the effectiveness of these systems in dynamic training environments. This tailoring enables the models to support specialized niche domain requirements, providing focused solutions that traditional training models cannot provide. A good example of this is the creation of AI models from specialized datasets (like modern sales reports) to enhance their effectiveness for specific roles, thus promoting compliance with industry-specific regulations.

Larger models also perform better when challenged by scale and complexity. As emergence occurs, machine models can expand beyond certain limits to solve complex and multifaceted problems that simpler models cannot.

3.5. Model fine-tuning for business adaptation

After the design of an AI-based framework for the distribution of corporate knowledge, the system can be fine-tuned later to align better with specific organizational requirements. Fine-tuning large-scale models allows for enhanced operational effectiveness by the addition of special-purpose knowledge databases so as to fine-tune search functions or build customized question-and-answer systems that are meaningful to the business environment.

For example, VisualGLM can serve as a generalized multimodal model requiring fine-tuning using domain-specific data sets, like those relevant to medical imaging, to increase its effectiveness in recognizing medical images. This process reflects the concept of hyperparameter optimization employed extensively in ML, as the model parameters are modified to match the intricate complexity of the targeted data set more closely. Additionally, large models can experience the process of repeated fine-tuning, whereby each cycle contributes to the model's overall advancement in efficacy. This cyclical practice guarantees that the model remains perpetually current in light of the advent of new innovations, as well as adapts and changes to suit the evolving demands of the business environment.

Adapter tuning is a specialized form of fine-tuning. Adapter tuning is a specialized form of fine-tuning in which small trainable modules, called adapters, are inserted into selected pre-trained layers of the neural network. When fine-tuning, the parameters of the base model are frozen, and only the adapters are adjusted. Using this mechanism, models can be easily switched to new tasks without changing their underlying architecture. Adapter fine-tuning is also very beneficial in large corporate knowledge transfer systems, where the model would adapt to changing market forces, business objectives, and information requirements.

For instance, one can initiate from a general-purpose text generation model that was initially pre-trained for a wide range of tasks. By purposefully placing adapters at different depths within the model, it is possible to appropriately adapt the model for the task of live-stream sales report generation. During the fine-tuning stage, the parameters of the adapters are solely fine-tuned on live-stream sales data, including appropriate words, styles, and contextual information. As a result, the model can produce more accurate and relevant sales reports while preserving the overall architecture of the backbone (see Figure 5).

OPEN IN VIEWER

Figure 5.

Adapters fine-tune the main architecture. ¹⁵

The process of fine-tuning large AI models offers more than just incremental improvements; it also facilitates the discovery of latent patterns within data that may not have been explicitly identified by the system before fine-tuning. Fine-tuning involves “infusing” additional domain-specific information into the model, which enhances its functionality for particular tasks. As the model is exposed to specialized datasets, it gains a deeper understanding of the domain, thereby improving its performance on tasks such as sentiment analysis, entity recognition, text classification, and dialogue generation.

Recent advancements in AI research have revealed a phenomenon known as emergence, where the expansion of a model, whether in terms of its parameters, the amount of training data, or computational resources, results in a significant leap in the model's capabilities. Emergence occurs when the model reaches a substantial threshold, which in turn produces a sudden increase in its ability to solve a wider range of difficult tasks. Emergence can be noted especially when the model has been trained on diverse datasets and tasks beyond the computational power of smaller models.

Emergence, as described by Rainey et al., ¹⁶ refers to the process where the behaviors in a system develop in an implicit manner rather than as a consequence of intentional design approaches. The change in behavior often results in considerable improvement in the capability of the system; however, undesirable results can also result. The above-mentioned study by Google, with the name “Emergent Abilities of Large Language Models,” illustrates the phenomenon of emergence when a model reaches some threshold, either in size or the amount of data in the dataset it processes. Reaching the said threshold results in the model's capability undergoing a sudden and unforeseen boost, allowing it to go past the performance attained by smaller models in tasks it could not handle well previously. With the size of the model increasing, it develops new behaviors that no one could have anticipated from its previous versions.

Essentially, the improvement in model performance is due to added parameters, a higher amount of training data, and increased computational power, thus allowing the model to solve more sophisticated tasks with increased precision. As depicted in Figure 6, Large Language Models (LLMs) demonstrate emergent capabilities, where the model's performance jumps dramatically after reaching a certain threshold, allowing it to perform tasks that smaller models are unable to manage.

OPEN IN VIEWER

Figure 6.

Eight examples of emergence in the few-shot prompting setting. ¹⁷

Several studies have examined the emergent capabilities exhibited by LLMs. One such study defines emergent capabilities as those “present in larger models but absent in smaller models.” This phenomenon suggests that as an ML model grows in size, its ability to perform specific tasks improves, sometimes dramatically. As illustrated in Figure 6, LLMs exhibit a sudden performance improvement when the model exceeds a certain threshold, enabling it to tackle more complex and diverse tasks that were previously beyond its reach.

The use of generative AI in apprenticeship structures has led to considerable progress in the skill levels of learners and enabled the detection of new patterns and insights. By processing extensive datasets, these AI models can detect TK that is potentially not immediately visible to the human instructor. This new ability for innovation opens the doors to the development of enhanced training systems and organizational procedures.

The ML methods applied in this study demonstrated the ability to discover relationships between novel results and the fundamental physical rules on which they rely, highlighting an importance that renders them extremely beneficial to the domains of optimization and design. These technologies represent a disruptive shift in the domain of AI, thus requiring the formulation of fresh problem-solving strategies and adaptable methodological concepts.

3.8. Discussion

This study has revealed that conversational AI can transform apprenticeship models, particularly in the area of TK transmission. Traditional apprenticeship models, based on face-to-face mentoring for skill imbibition, face challenges in codifying and formalizing TK. Existing models of apprenticeship are increasingly inadequate in responding to industry demands and the scarcity of skilled labor. Generative AI paves the way for a reimagining of the models by representing TK in a richer and shareable form. The research examines the use of sophisticated forms of AI, including ML and NLP, to assess how AI can aid in the capture, organization, and sharing of knowledge to enhance learning results and extend apprenticeship-based training efforts.

The research design applied in the current study includes a controlled experimental design that was applied to 50 novice live-stream hosts, who were systematically divided into two cohorts: the experimental cohort, which used AI-based tools, and the control cohort, which used traditional training methods. The participants were assessed in seven competency areas over a period of three months. The areas included on-camera presence, verbal delivery skills, audience engagement, responsive adaptability, learning agility, personal branding, and sales conversion skills. Independent samples t-tests were used to determine whether the differences found between the two groups were statistically significant. The results were unambiguous: the experimental cohort performed better than the control cohort across all dimensions, with p-values consistently less than 0.01, thus indicating strong statistical significance.

The results of this present study provide conclusive proof of the beneficial impact of generative AI on apprenticeship training. In particular, the EG with AI assistance scored 85 in expressiveness of shots, compared to 70 for the CG, with a t-value of 4.23 and a p-value less than 0.01, thus establishing a significant improvement in presentation skills compared to conventional live streams. Similarly, the EG scored 88 on verbal expression, compared to 72 for the CG, thus establishing a significant improvement in communicative skills. These findings were also replicated across different performance parameters, including technical competence in site administration, response speed, and general learning ability. The statistical findings present definitive proof of the potential of generative AI to drive accelerated growth of both technical and adaptive skills in the context of apprenticeships. The AI-assisted tools not only enhanced tech skills of a specific kind but also developed higher-order adaptive skills that are vitally important in rapidly changing professional contexts.

Despite the promising nature of the results, some limitations are worthy of discussion. A key area of weakness is the challenge of holistically integrating all facets of TK, especially those of an intrinsically personal nature or stemming from particular experiences. Generative AI illustrates its strength in organizing and developing upon more objective facets of TK, i.e., established protocols and technical procedures, but might face challenges in duplicating subtle, instinctual learning imparted by experienced teachers in complex, context-sensitive situations. Moreover, the reliance on AI in apprenticeship models requires careful contemplation of the ethical considerations involved in replacing human mentorship with machine-driven systems. The potential risks of over-reliance on AI-powered solutions can diminish the importance of careful personal attention and mentorship, which are vital components of traditional apprenticeship. Additionally, output generated by research methods that utilize AI is highly dependent on the quality of training data used. Low-quality or corrupted data can produce inferior or flawed knowledge outputs, consequently negatively impacting the effectiveness of AI-powered solutions.

The results of the research have profound consequences for enterprises that harness the apprenticeship scheme, including live-streaming commerce and other hands-on professions. By adopting the epistemic framework for knowledge transfer in the AI space, educational institutions can improve the efficiency of the curriculum implementation, assist in the guarding of vital TK, and systematically strengthen the efficacy of KTs at different levels of the organization. The integration of AI provides the opportunity to offer personalized learning trajectories, adaptive instructional materials, and real-time performance feedback, thereby refining the instructional strategies, as well as adapting them appropriately to the learners’ needs. Additionally, the rise of AI provides the opportunity for enterprises to create centralized repositories of knowledge that are continuously updated and readily accessible, thereby promoting superior KTs and innovation across the organization's solutions. The technology has a leading role in developing multiple aspects of human capital formation, especially in cases where protecting knowledge and fast-tracking skill formation are important for preserving competitive advantages.

Funding project

The first batch of teaching reform projects for higher vocational education in Zhejiang Province during the 14th Five-Year Plan period: The development of the “Live-streaming Sales” textbook based on the Chinese characteristic apprenticeship system(jg20230225).