Extracting Implicit Knowledge with the Help of Generative AI

Implicit knowledge, the deeply ingrained insights and expertise that individuals possess, has long been recognized as a vital asset for organizations. However, extracting and leveraging this knowledge is often difficult and sometimes even an elusive goal. With recent breakthroughs in generative AI, we may now have the right tools to unlock implicit knowledge on an unprecedented scale.

This week, Turtle's AI newsletter explores the conceptual underpinnings of implicit knowledge, why accessing it is so valuable yet difficult, and how innovations in large language models (LLMs) and other generative AI technologies could provide new means for organizations to tap into this knowledge.

The Theoretical Basis of Implicit Knowledge

Philosophers as early as Aristotle recognized that some knowledge defies easy articulation. In his seminal work The Nicomachean Ethics, Aristotle distinguished between two types of knowledge: episteme, know-that knowledge of facts and universal truths; and techne, know-how knowledge mastered through practice and experience. The latter corresponds closely to modern notions of implicit knowledge.

In the 20th century, philosopher Michael Polanyi fully developed the theory of implicit knowledge in his book The Tacit Dimension. Polanyi asserted that “we can know more than we can tell”. He introduced the phrase “tacit knowing” to describe the act of understanding something without relying on explicit rules or knowledge. As an example, Polanyi noted how we recognize faces intuitively, without being able to pinpoint exactly how we perform this feat of perception and memory.

Sociologist Ikujiro Nonaka built upon Polanyi’s work to explore knowledge creation within organizations. In his SECI model, Nonaka described how knowledge conversion occurs along two dimensions: tacit to explicit, and individual to collective. A key premise is that tacit knowledge can be extracted and codified through social interactions and collaboration.

Psychologist Gary Klein has also examined intuitive expertise, which aligns with implicit knowledge. His research found that experienced professionals often make rapid decisions by drawing on deeply ingrained mental models rather than deliberate analysis. While difficult to externalize, this tacit knowledge is vital for excelling in many fields.

The common thread running through these perspectives is that implicit knowledge represents a valuable yet elusive asset. It develops gradually through immersive experience. It resides in the minds and behaviours of individuals, encoded in perceptions, judgements, and physical skills that operate beneath conscious awareness. Tapping into this resource presents organizations with both tremendous opportunities and difficulties, as the next section explores.

The Promise and Challenge of Leveraging Implicit Knowledge

Across industries, leading organizations recognize that their most valuable assets walk out the door each evening. Their competitive advantage lies in the expertise and creativity of their employees. However, translating this individual knowledge into collective benefits has proven difficult. Surveys show that up to 90% of workplace knowledge remains underutilized. Why does this challenge persist, and how could accessing implicit knowledge provide value?

First, implicit knowledge does not automatically diffuse through an organization. Unlike explicit information contained in documents and databases, it sticks to the people who developed it through years of experience. They often struggle to articulate their intuitions and thought processes. Critical insights remain trapped inside someone's head.

Second, attempts to extract and codify implicit knowledge do not always succeed. Debriefing experts through interviews and observation can be time-consuming and inaccurate. People may not even be conscious of their own mental models, heuristics, and ingrained skills. This creates barriers for translating knowledge across individuals and groups.

However, accessing and integrating high-quality implicit knowledge provides several advantages. It avoids redundancy of efforts, as people do not have to keep reinventing the wheel. It reduces errors and mistakes, leveraging learned lessons and best practices. Innovation flourishes when individuals share their unique perspectives. Implicit knowledge also propagates a culture of excellence, raising the competency bar across an organization.

In summary, implicit knowledge is a core ingredient for organizational success, but traditional extraction methods fall short. Companies need new approaches to unlock this vital asset. This brings us to the transformational potential of generative AI.

Philosophical Perspectives on Generative AI and Implicit Knowledge

Generative AI has sparked profound questions about the nature of knowledge itself. Leading philosophers have put forth views on how technology is reshaping our relationship with knowledge. These perspectives provide context on both the promise and perils of using generative AI to extract human wisdom.

Hubert Dreyfus, renowned for critiquing early AI, argued that knowledge acquisition requires embodied skills learned through real-world experience. Purely cerebral approaches cannot capture context-dependent intuition. While acknowledging impressive advances, Dreyfus would likely question whether disembodied AI systems could encapsulate embodied implicit knowledge effectively.

John Searle developed the Chinese Room thought experiment to expose limitations in AI's understanding. He asserted that processing symbols is not equivalent to comprehension. This raises concerns about generative AI's ability to reproduce knowledge rather than merely mimic it. However, Searle's views have also been critiqued for setting unreasonably high bars for machine knowledge.

Hilary Putnam espoused an externalist perspective in which knowledge extends beyond what is in the mind. In this view, AI systems that leverage external data sources are not confined to internal cognition. Some philosophers argue that generative models connecting diverse inputs exhibit extended knowledge.

These perspectives highlight debates surrounding AI and knowledge. Critics see generative technologies as merely manipulating symbols without true understanding. Advocates counter that AI represents extended knowledge networks reflecting collective intelligence. Regardless, extracting implicit knowledge raises crucial questions about replication versus comprehension.

Sociological Implications of Extracting Implicit Knowledge with AI

Applying generative AI to unlock human knowledge also introduces societal impacts. Sociologists have put forth several theories regarding how AI could affect knowledge sharing in communities. These provide insights into principles for responsibly extracting implicit knowledge.

Habermas introduced colonization theory, critiquing society's tendency to favor instrumental technical knowledge over communicative wisdom. As engineers develop systems to extract human knowledge, they may need to prioritize meaningful discourse and mutual understanding.

Frickel theorized about epistemic communities. To avoid fragmented knowledge, engineers should see themselves as embedded within social systems rather than taking purely technical perspectives. Generative AI should enhance collective knowledge, not isolate it.

Sociologists also examine deskilling, recognizing that past automation technologies have often degraded human expertise. To counteract this, companies should ensure generative AI augments professionals to enrich skills rather than replace them.

Finally, theories on the social construction of technology emphasize that societal values become embedded in designs. Generative AI should therefore incorporate principles of transparency, accountability, and human dignity when extracting knowledge.

In summary, philosophers raise epistemic questions about machine knowledge, while sociologists highlight complex social dynamics surrounding AI and knowledge sharing. By considering diverse viewpoints, organizations can develop ethical, human-centred approaches as they leverage generative technologies to extract implicit knowledge.

Practical Applications of Generative AI to Extract Implicit Knowledge

While debates continue regarding the philosophical and social implications, generative AI offers tangible value today for extracting implicit knowledge in practice. LLMs, combined with fine-tuning techniques, are unlocking use cases that were previously intractable. Here we present five examples of applying generative AI to extract high-quality implicit knowledge in domains from engineering to healthcare:

1. Intelligent Expert Systems. With its ability to generate human-like text, GPT-3 and newer models (e.g. Claude, Bard, LLaMa...) can power conversational agents that provide expert assistance. After ingesting technical manuals and accumulating domain expertise through practice questions, these AI tutors deliver insights tailored to users' needs. Immersive simulations further enrich their implicit skills.
2. Design Pattern Mining. By analyzing codebases, tools like GitHub Copilot can extract embedded design patterns. This makes best practices reusable across the organization instead of siloed in individuals' habits. StructuredParsing techniques also elicit abstract representations of code, surfacing knowledge.
3. Patient Therapeutic Insights. Doctors develop clinical acumen through years of practice. By learning from anonymized records, AI models like Anthropic's Claude can deliver personalized therapeutic insights to patients, capturing physicians' implicit skills.
4. Fraud Detection. Experienced fraud analysts notice subtle suspicious patterns. But these instincts stay private. Anti-fraud AI systems leverage past examples to recognize fraud heuristics. They expose insights from analysts’ minds to the whole team.
5. Scientific Knowledge Extraction. AI can mine scientific papers to uncover hidden connections. Modern LLMs scan full texts to link concepts and surface testable hypotheses, instead of relying on human-defined keywords. This paradigm shift accelerates discovery.

These examples demonstrate generative AI’s versatility for extracting diverse forms of implicit knowledge. But like any technology, its application requires forethought and care. Next we turn to practical strategies organizations should employ to implement AI knowledge extraction responsibly and effectively.

Responsible Approaches for Extracting Implicit Knowledge with AI

While promising, applying generative AI to unlock implicit knowledge demands diligence. Organizations should adhere to principles of transparency, ethics, and human partnership:

First, establish clear documentation and communication regarding the AI system's provenance, training process, capabilities, and limitations. Proactively mitigate risks such as bias propagation.

Second, implement monitoring procedures to regularly review model behaviour and correct errors. Subject outputs to human evaluation before dissemination.

Third, position AI tools as assistants rather than replacements for human experts, preserving professional skills and judgement. Avoid overreliance on AI suggestions.

Fourth, take an iterative approach by starting with limited use cases and data samples. Expand the application incrementally based on feedback and demonstrated value. Move cautiously instead of overextending AI systems beyond their capabilities.

Fifth, implement thorough testing protocols to validate performance on corner cases and identify potential harms. Rigorously scrutinize whether outputs match truth and wisdom criteria. Refine the systems based on insights from validation.

Sixth, engineer AI systems to provide explanations regarding their reasoning, influences and limitations. Build interpretability directly into the models rather than treating them as black boxes. Explainability supports trust and accountability.

Seventh, give professionals visibility into the training data and processes shaping AI systems. Enable them to audit algorithms and verify alignment with experiential knowledge from practice. Transparency empowers users.

Finally, continually monitor the impacts of AI knowledge extraction on workplace culture and effectiveness. Survey end users regularly to gauge perceptions. Adapt strategies based on human feedback to prevent deskilling and maintain organizational health.

In closing, generative AI shows immense potential for unlocking the wealth of implicit knowledge within organizations' walls. But prudent programs demand proactive mitigation of risks and alignment with human needs.

Managers should approach this technology thoughtfully, elevating skilled expertise rather than seeking to replace it.

What experiences can you share regarding unlocking your organization's knowledge? What concerns do you want to discuss? Let's keep the conversation going to steer these technologies toward empowering professionals and enriching collective wisdom.

Duke Rem ?? and Turtle's AI team

?