Digital humans as knowledge carriers: How AI avatars transform knowledge storage and transfer
During my internship at ByteDance, I was responsible for recruiting candidates with expertise in computer graphics for the development of social games on Douyin. This experience introduced me to the world of digital humans and avatar customization, concepts that extend far beyond gaming. I quickly realized that these virtual entities are not just about entertainment—they are also deeply connected to how organizations store, transfer, and manage knowledge.
Traditionally, knowledge management (KM) has relied on structured repositories such as databases, manuals, and training programs. While effective in storing explicit knowledge, these methods often struggle to capture the rich, experience-based tacit knowledge that resides in experts’ minds. What if AI-driven digital humans could bridge this gap? Imagine a future where employees could consult a virtual mentor—one that not only recalls technical processes but also adapts its responses based on years of accumulated experience.
Digital humans, powered by AI and machine learning, are emerging as potential knowledge carriers in businesses, education, and even customer service. Companies like Microsoft and Nvidia are already leveraging AI avatars for workplace collaboration and training, while industries from healthcare to consulting are exploring how virtual assistants can capture and relay expertise dynamically. This raises a crucial question: Can digital humans become more than just virtual assistants and evolve into living repositories of knowledge?
This article explores how AI-powered digital humans can store, transfer, and manage knowledge in organizations, particularly in industries where expertise is highly valuable. By examining real-world applications and potential challenges, we will assess whether these virtual beings represent the next stage in the evolution of knowledge management—or if they remain just another technological experiment.
Digital humans as knowledge repositories: storing expertise beyond databases
Traditional knowledge management systems typically depend upon explicit repositories such as databases and manuals to hold explicit knowledge—straight facts and rule-bound directives that are easy to codify. But it has always acted as an impediment to encode the tacit knowledge—deep, experience-based understandings contained in the brains of individuals. Tacit knowledge is largely unwritten, learned through performance, and difficult to formalize or transfer.
AI-powered digital human beings offer a creative solution to this problem by acting as dynamic, interactive repositories that are able to absorb and distribute explicit as well as tacit knowledge. In contrast to static databases, these AI entities are able to carry out conversations in real-time, constantly learning from conversations so that they can refine their knowledge base. This allows them to simulate human-like understanding and provide context-specific responses, essentially bridging the distance between recorded data and practical expertise.
For example, Microsoft’s Mesh platform1 has AI-infused avatars in work collaboration products, facilitating immersive, real-time conversations that facilitate the exchange of intricate knowledge in an effortless manner. The technology allows individuals to communicate with replicas of specialists, facilitating an easy exchange of explicit as well as tacit knowledge.
Similarly, SenseTime2, one of the prominent AI companies, has developed digital human technologies that are intelligent, interactive repositories. A notable case is their AI receptionist, which was able to recall and deliver service-related knowledge in real-time, illustrating the storage of operational expertise through conversational interaction. During a course-organized visit to SenseTime’s Shenzhen office in 2022, their HR team also demonstrated how these digital individuals are implemented to both store organizational knowledge as well as convey it. Through interactions with people and real-time inputs, these AI-backed avatars are able to provide personalized responses as well as provide training, basically retaining the nuanced expertise that other documentation processes lack.
Integrating digital individuals into knowledge management systems is an essential leap, one that allows for an expanded approach toward retaining organizational expertise and distributing it. By marrying the data-storing functions of traditional databases with the adaptive, interactive functions of AI, corporations are able to retain and utilize the full scope of their intellectual capital.
The role of digital humans in knowledge transfer and learning
While digital humans are promising as units of information, their potential to revolutionize knowledge transfer and learning is even more profound. Traditional corporate learning methods—such as instructor-led sessions, mentorship, and classroom training—face challenges in scalability, consistency, and personalization. Knowledge transfer also proves uneven—multiple trainers interpret and transfer the same content differently, leading to inconsistent learning outcomes. Moreover, static materials like video lectures and manuals lack real-time interaction, limiting their effectiveness for diverse learners.
Digital humans address these concerns by offering scalable, adaptive, and interactive learning experiences. Compared to recorded sessions, AI-powered mentors can engage in real-time conversations, tailoring explanations based on an individual’s prior knowledge, learning pace, and specific queries. This transforms passive content consumption into dynamic learning, making the process stimulating and efficient.
A notable example is Sandvik Mining & Rock Solutions’ collaboration with Guildhawk3 to implement multilingual digital human avatars for employee safety training. These AI-driven avatars deliver culturally relevant and engaging modules that offer key advantages over traditional training methods—around-the-clock availability, multilingual support, real-time interaction and cost efficiency at scale, making L&D training more accessible and effective for a global workforce.
However, the effectiveness of digital humans in knowledge transfer is not guaranteed. Challenges remain in their ability to interpret nuanced user inquiries, generate contextually appropriate responses, and win over human employees’ trust.
Digital humans and tacit knowledge transfer: the SECI model perspective
Tacit knowledge plays a crucial role in organizational learning, particularly in experiential and interpersonal contexts. To better analyze how such knowledge might be transferred through digital means, Nonaka and Takeuchi’s SECI model4 offers a useful framework. Its four modes—socialization, externalization, combination, and internalization—highlight how tacit and explicit knowledge interact in dynamic cycles. Socialization, the mode through which tacit knowledge is shared via shared experience and observation, is traditionally limited by the need for close personal interaction—often informal and unstructured. This makes it difficult to scale, especially in large or remote organizations.
AI-driven digital individuals may offer a partial solution by enabling elements of socialization to occur in virtual environments. Through humanlike conversation, case-based learning, and simulated decision-making scenarios, these systems can reproduce aspects of experiential learning. In leadership development and professional training, AI avatars can replicate elements of the user’s work environment and offer structured feedback, supporting employees in absorbing complex, judgment-based insights. While not replicating true immersion, these tools help make elements of tacit knowledge more accessible and repeatable across contexts.
That said, the ability of digital humans to facilitate tacit knowledge transfer remains an evolving frontier. Because tacit knowledge is deeply context-bound and shaped by intuition, emotion, and situational judgment, AI systems—despite advances in machine learning—may still struggle to interpret and respond meaningfully in more nuanced situations. For instance, they may not consistently recognize subtle patterns in individual learning preferences or emotional cues. Rather than outright replacements for human mentors, AI-powered avatars may be better viewed as complementary assistants, enhancing but not fully replicating human-led knowledge transfer. Their continued advancement will likely require close integration with evolving technologies such as adaptive learning systems and human-in-the-loop feedback processes, though the extent of such success remains to be seen.
This invites a broader reflection: will digital humans ever become capable of transferring deeply held experiential knowledge in a truly human way, or will their role continue to be that of scalable support tools that complement, rather than replace, human mentorship?
Conclusion: rethinking the role of digital humans in knowledge management
Despite their ability to store, retrieve, and even contextualize information, digital humans are constrained by essential limitations from becoming authentic facilitators of knowledge. Among the strongest hindrances is that they cannot match the profoundly human quality of knowledge exchange. Knowledge management encompasses not only the conveyance of facts—it requires collaboration, interpretation, and emotional intelligence accrued from human-to-human exchange.
In mentorship, for example, experienced practitioners do more than deliver instructions—they interpret subtle contextual cues, adapt their guidance on the fly, and foster growth through relational presence. While AI systems have made progress in natural language processing and adaptive learning, they still operate within structured parameters and may struggle with the spontaneity, ambiguity, and emotional dynamics that human mentors navigate instinctively.
Although digital humans may not yet rival human expertise in conveying deeply embedded knowledge, they offer promising potential as amplifiers of knowledge access. With ongoing advances in AI reasoning, immersive interfaces, and real-time learning, these systems are becoming increasingly capable of supporting interactive and scalable learning environments. Recent discussions, such as those by the World Economic Forum5, emphasize the importance of making AI tools broadly accessible rather than concentrated in the hands of a few. In this context, national initiatives like Singapore’s Smart Nation provide a glimpse into how AI-powered knowledge systems can help close—not widen—knowledge gaps across sectors such as education, public service, and industry. For this potential to be realized, AI in knowledge management should be guided by principles of ethical governance, ensuring that human insight is not diminished but amplified.
At its core, knowledge management is concerned with enabling the effective creation and application of knowledge within social and organizational processes—not merely the storage and dissemination of information. Digital humans may help organizations scale aspects of this process, but the richness of mutual experience and interpretive judgment remains inherently human. The future of knowledge management has nothing to do with AI versus human, but has absolutely everything to do with building an environment wherein people are empowered by, not alienated by, technology.
Article source: Adapted from Digital humans as knowledge carriers: How AI avatars transform knowledge storage and transfer, prepared as part of the requirements for completion of course KM6304 Knowledge Management Strategies and Policies in the Nanyang Technological University Singapore Master of Science in Knowledge Management (KM).
Artificial intelligence (AI) statement: To assist the preparation of this article, Wang Qiwei used ChatGPT to help identify relevant real-world examples of how digital humans are applied in knowledge management, which then served as the basis for contextual analysis and case integration.
Header image source: Created by Wang Qiwei with Microsoft Designer Image Creator.
References:
- Microsoft. (2024, January 24). Bring virtual connections to life with Microsoft Mesh, now generally available in Microsoft Teams. Microsoft 365 Blog. ↩
- SenseTime. (2021, January 13). SenseTime’s AI “Digital Human” Receptionist Adds Intelligences to Customer Service. SenseTime Newsroom. ↩
- Guildhawk. (2024, March 27). How to use Digital Humans in L&D Staff Training. Guildhawk Insights. ↩
- Nonaka, I. (1994). A Dynamic Theory of Organizational Knowledge Creation. Organization Science, 5(1), 14-37. ↩
- Li, C. (2025, February 24). Commentary: AI and the edge of possibilities. Channel News Asia. ↩
Dear Ms Qiwei, I found your article courageous, as you express your opinion on the use of AI for translating implicit knowledge. Implicit knowledge includes feelings that we cannot easily codify, as well as contexts that AI does not understand as a whole because it has not ‘experienced’ them. So, how could digital humans codify this if they are merely a means of codifying knowledge? How could digital humans understand feelings or physical sensations? I believe that a lot of implicit knowledge would be lost or misunderstood as implicit, when in reality it would just be the transcription of audio recordings and/or decisions we make with the tool. Well, that’s exactly what we already do when we create explicit knowledge.
The discussion is indeed necessary, and analysing the pros and cons is an important exercise for us in order to be able to make decisions about the use of AI in the future.
Dear Ms Carolina, thank you so much for taking the time to read my article and for sharing such thoughtful reflections. I truly appreciate your point about the limitations of AI in capturing the depth of implicit knowledge—especially when it comes to emotions, embodied experience, and unspoken context.
Your question—how could digital humans codify something they haven’t “experienced”—gets right to the heart of the dilemma. I fully agree that much of what we consider tacit knowledge risks being oversimplified or even misrepresented when processed through AI systems. In this article, my intention was not to suggest that digital humans could fully replicate tacit knowledge, but rather to explore how they might support certain aspects of its transfer in scalable contexts, especially in cases where access to expert mentorship is limited.
I also share your concern that transcription or simulation should not be mistaken for true understanding. As you rightly noted, this is precisely why the distinction between tacit and explicit knowledge remains critical in knowledge management. The hope is that by recognising these limitations early, we can better shape how AI is designed and deployed—ideally to augment human insight rather than replace it.
Thank you again for opening this important conversation. I’m very grateful for your perspective.