How organizations can transform institutional knowledge into AI-enabled capability—without losing the human judgment that created it 

By Kiran Viswanatha 
AI Program Leadership | Responsible AI | Research & Governance 

 

 

“The greatest AI asset an organization possesses may not be its models. It may be the expertise already inside its people.” 

For decades, organizations have invested in technology to capture what their people know. 

We turned expertise into: 

• business rules,  

• databases,  

• workflows,  

• SOPs,  

• decision trees,  

• algorithms,  

• knowledge bases,  

• automation scripts,  

• dashboards,  

• and enterprise applications.  

Today, artificial intelligence is taking this transformation one step further. 

AI can interact with organizational knowledge in ways that feel remarkably human: it can summarize, reason over information, identify patterns, answer questions, recommend actions, and increasingly orchestrate workflows. 

This creates a fascinating question for AI leaders: 

What happens when human expertise acquires a digital form? 

We can describe this transformation metaphorically as metempsychosis—the movement from one form into another. 

In this context: 

Human expertise becomes organizational knowledge; organizational knowledge becomes digital logic; digital logic becomes AI-enabled intelligence; and AI-enabled intelligence becomes a new organizational capability. 

But there is an important caveat. 

We should not confuse the digitization of expertise with the digitization of wisdom. 

AI can potentially preserve and amplify knowledge. 

It cannot automatically inherit the judgment, ethics, lived experience, accountability, or human context of the expert who created that knowledge. 

That distinction will define the next generation of enterprise AI. 

 

Did You Know? 

When an experienced employee leaves an organization, the company doesn't simply lose a person. 

It may lose: 

• years of problem-solving experience;  

• understanding of exceptions;  

• customer relationships;  

• operational intuition;  

• knowledge of why processes exist;  

• awareness of historical failures;  

• understanding of undocumented dependencies.  

Some of this knowledge is explicit. 

Much of it is tacit. 

And tacit knowledge is notoriously difficult to capture. 

An experienced project manager may say: 

"I've seen this vendor behavior before. If we don't address it now, we'll have a delivery problem in six weeks." 

That statement may not exist anywhere in the project documentation. 

It is accumulated experience. 

This is where the journey from silicon to "spirit" begins. 

 

1. What Is the "Spirit" of Expertise?

The word spirit here is metaphorical. 

It does not mean that AI possesses consciousness or a human soul. 

Instead, think of the "spirit" as the patterns, principles, experience, and decision knowledge embedded within human expertise. 

Consider an experienced cybersecurity professional. 

Their value isn't merely knowing: 

"Rule X exists." 

Their expertise may include recognizing: 

"This combination of events looks unusual because I've seen a similar attack pattern before." 

That difference is enormous. 

Information tells us what happened. 

Expertise helps us understand what it means. 

AI systems are increasingly capable of working with both—but the quality of their output depends heavily on the quality, context, governance, and boundaries surrounding the knowledge they receive. 

 

2. The Metempsychosis of Expertise

We can visualize the journey like this: 

HUMAN EXPERIENCE 

       ↓ 

Tacit Knowledge 

       ↓ 

Documented Knowledge 

       ↓ 

Business Rules & Data 

       ↓ 

Digital Systems 

       ↓ 

AI Context & Retrieval 

       ↓ 

AI-Assisted Reasoning 

       ↓ 

Organizational Capability 

       ↓ 

Human + AI Decision Making 

This isn't simply technology transformation. 

It is knowledge transformation. 

And that makes it highly relevant to AI leaders, PMO leaders, and project managers. 

 

3. The First Stage: Human Expertise

Every organization starts with people. 

A manufacturing engineer knows how a machine behaves. 

A claims specialist knows which cases require additional investigation. 

A financial analyst recognizes unusual patterns. 

A project manager knows when a seemingly minor dependency could become a major schedule problem. 

A customer-service expert knows when a customer's words don't tell the entire story. 

This knowledge is accumulated through: 

experience + experimentation + mistakes + feedback + context. 

The challenge is that much of it exists inside human minds. 

 

4. The Second Stage: Capture

Organizations then attempt to capture expertise. 

They create: 

• manuals;  

• policies;  

• FAQs;  

• knowledge bases;  

• process maps;  

• training material;  

• decision trees;  

• project documentation.  

This converts: 

Tacit knowledge → explicit knowledge 

But something can be lost during translation. 

A 30-year veteran's experience may become a 10-page procedure. 

Useful—but incomplete. 

The exceptions may not be documented. 

The context may disappear. 

The "why" may be missing. 

 

5. The Third Stage: Encode

Technology then turns knowledge into executable logic. 

For example: 

IF customer meets criteria A 

AND transaction exceeds threshold B 

THEN initiate review C. 

This is powerful. 

Organizations can now automate decisions. 

But rigid rules have a weakness: 

The real world rarely behaves exactly like the rulebook. 

Exceptions appear. 

Contexts change. 

Customers behave unexpectedly. 

Regulations evolve. 

Markets shift. 

This is where AI introduces a new possibility. 

 

6. The Fourth Stage: Contextual Intelligence

Generative AI and modern AI architectures can potentially interact with enterprise information in a more flexible way. 

Instead of only executing: 

IF X → THEN Y 

AI can help interpret: 

"Given this situation, what information, policies, patterns, and historical knowledge are relevant?" 

This is where concepts such as: 

• retrieval-augmented generation;  

• context engineering;  

• knowledge graphs;  

• intelligent workflow orchestration;  

• agentic systems  

become increasingly relevant. 

But flexibility creates responsibility. 

The more flexible the intelligence, the more important governance becomes. 

 

7. Real-World Example: Healthcare

Imagine an experienced healthcare operations specialist who understands how authorization cases move through a complex process. 

Their expertise includes: 

• payer requirements;  

• organizational policies;  

• workflow exceptions;  

• documentation patterns;  

• escalation criteria.  

An AI-enabled system could potentially help staff navigate these processes by bringing together approved knowledge sources and providing contextual assistance. 

Instead of asking employees to search across multiple documents, the system might help answer: 

"What information is missing, which policy applies, and what should happen next?" 

The objective isn't to replace the expert. 

It is to make expertise more accessible. 

Human professionals still need to remain accountable for consequential decisions. 

 

8. Real-World Example: Manufacturing

Consider an experienced maintenance engineer. 

They have learned that certain combinations of: 

• temperature;  

• vibration;  

• operating hours;  

• machine behavior  

often precede equipment problems. 

That knowledge may never have been formally documented. 

With appropriate data collection and validation, an AI system could potentially learn patterns associated with equipment performance and help surface early warning signals. 

The transformation becomes: 

Engineer experience 

→ Operational data 

→ Pattern recognition 

→ AI recommendation 

→ Engineer validation 

→ Maintenance decision 

This is not the replacement of expertise. 

It is expertise becoming scalable. 

 

9. Real-World Example: Project Management

This transformation is particularly interesting for project managers. 

Imagine an organization has completed 5,000 projects. 

Those projects contain information about: 

• schedule delays;  

• vendor problems;  

• resource constraints;  

• dependency failures;  

• change requests;  

• stakeholder conflicts;  

• technology risks;  

• successful interventions.  

Historically, lessons learned might sit inside thousands of documents. 

An AI-enabled PMO could potentially turn that institutional memory into an intelligent knowledge capability. 

A project manager might ask: 

"What risks typically emerge when implementing this type of transformation with this vendor profile and dependency structure?" 

The system could retrieve relevant historical evidence and identify patterns. 

The PM doesn't surrender judgment to AI. 

Instead: 

The PM gains access to organizational memory at the moment of decision. 

 

10. The PMO Could Become an Organizational Memory System

This leads to an exciting possibility. 

The future PMO may not only manage: 

projects → programs → portfolios 

but also manage: 

knowledge → patterns → lessons → organizational intelligence. 

Imagine a PMO where every completed initiative contributes to a continuously improving knowledge ecosystem. 

 

The organization gets smarter with every project. 

 

11. But There Is a Dangerous Assumption

We must be careful with the phrase: 

"Capture human expertise and put it into AI." 

Expertise isn't just information. 

It contains: 

• judgment;  

• uncertainty;  

• ethics;  

• context;  

• experience;  

• intuition;  

• accountability.  

If we simply feed historical decisions into an AI system, we may also encode: 

• historical bias;  

• outdated assumptions;  

• poor decisions;  

• inconsistent practices;  

• organizational blind spots.  

AI can preserve institutional knowledge—and institutional mistakes. 

That is why governance must accompany knowledge transformation. 

 

12. The Responsible AI Filter

Before converting human expertise into AI capability, ask: 

Is the knowledge accurate? 

Is it still relevant? 

Who owns it? 

Is it legally and ethically appropriate to use? 

Does it contain bias? 

What assumptions does it contain? 

What exceptions exist? 

Who validates AI-generated recommendations? 

Who remains accountable? 

How can the recommendation be challenged? 

These are not technical questions alone. 

They are governance questions. 

 

13. From Knowledge Capture to Knowledge Governance

The traditional knowledge-management mindset was: 

"Can we capture what people know?" 

The AI-era question becomes: 

"Can we capture, validate, govern, contextualize, and responsibly operationalize what people know?" 

That is a much higher standard. 

A governance framework should consider: 

Provenance 

Where did the knowledge originate? 

Authority 

Who is authorized to define or approve it? 

Currency 

When was it last validated? 

Context 

Under what circumstances is it applicable? 

Evidence 

What supports the recommendation? 

Accountability 

Who owns the outcome? 

Human oversight 

When must a person intervene? 

 

14. The "Expertise-to-AI" Pipeline

A responsible transformation can follow a structured pathway: 

1. Discover

Identify critical expertise. 

2. Capture

Document explicit and tacit knowledge as far as practical. 

3. Validate

Have domain experts verify the knowledge. 

4. Contextualize

Define where and when it applies. 

5. Govern

Establish ownership, access, controls, and accountability. 

6. Operationalize

Connect the knowledge to AI systems and workflows. 

7. Monitor

Continuously evaluate performance and relevance. 

8. Learn

Feed validated outcomes back into the knowledge system. 

This creates a human-AI learning loop. 

 

15. The Human-in-the-Loop Becomes Human-on-the-Loop

In some applications, humans may directly review every AI recommendation. 

In others, AI may handle low-risk tasks while humans oversee exceptions. 

This creates different governance models. 

Low-risk 

AI assists. 

Moderate-risk 

AI recommends; human approves. 

High-risk 

AI supports; qualified human decides. 

Critical-risk 

AI may be restricted to narrow, controlled functions. 

The appropriate level depends on the consequences of failure. 

Not every expertise-to-AI transformation should become autonomous. 

 

16. AI Should Amplify Expertise—Not Erase It

There is a subtle organizational risk in AI transformation. 

Suppose an organization takes experienced employees' knowledge and creates an AI assistant. 

Leadership might conclude: 

"We don't need as many experts anymore." 

That could become dangerous. 

Because experts don't just provide answers. 

They also: 

• challenge assumptions;  

• recognize anomalies;  

• understand ambiguity;  

• interpret context;  

• mentor others;  

• make ethical judgments.  

If the organization eliminates expertise after digitizing it, it may eventually lose the ability to validate the AI itself. 

The organization needs experts to govern the systems built from their expertise. 

 

17. The Paradox of AI Knowledge

Here's the paradox: 

The better we become at encoding expertise into AI, the more important it becomes to preserve human expertise. 

Why? 

Because AI systems require: 

• validation;  

• supervision;  

• updating;  

• challenge;  

• contextual interpretation.  

Therefore, the goal should not be: 

Human expertise → AI replacement 

It should be: 

Human expertise → AI augmentation → stronger human capability 

 

18. What This Means for AI Leaders

AI leadership needs to move beyond: 

"Which model should we use?" 

and toward: 

"What organizational expertise should we augment, and under what governance conditions?" 

Before selecting an LLM, ask: 

• What problem are we solving?  

• Whose expertise are we operationalizing?  

• What knowledge sources will the AI use?  

• How will we validate them?  

• What should AI never decide?  

• Where does human authority remain?  

• How will we monitor drift?  

• What happens when expert knowledge conflicts?  

These questions can prevent an AI initiative from becoming an expensive technology experiment. 

 

19. What This Means for Project Managers

Project managers can play a critical role in this transformation. 

When initiating an AI project, add an Expertise Map to your project plan. 

Identify: 

Expertise	Owner	Source	Context	Risk	AI Role
Claims processing	SME	SOP + cases	Claims	High	Recommend
Equipment maintenance	Engineer	Logs + manuals	Factory	Medium	Assist
Project risk	PMO	Historical projects	Portfolio	Medium	Predict
Customer service	Service team	Knowledge base	Support	Low/Medium	Assist

This turns an abstract AI strategy into an actionable program. 

20. The New AI Project Dependency

Traditional AI project dependencies might include: 

• data;  

• infrastructure;  

• APIs;  

• models;  

• security.  

Add another: 

Expert dependency. 

Ask: 

Who understands the business well enough to tell us when the AI is wrong? 

This person—or group—can become part of the AI governance structure. 

 

21. From "Human-in-the-Loop" to "Expertise-in-the-Loop"

The phrase human-in-the-loop is often used to describe oversight. 

But perhaps we need a richer concept: 

Expertise-in-the-loop. 

The objective isn't simply to have a human click Approve. 

The objective is to bring meaningful domain knowledge into: 

• design;  

• validation;  

• evaluation;  

• deployment;  

• monitoring;  

• continuous improvement.  

That's much stronger governance. 

22. The Silicon-to-Spirit Architecture

A responsible enterprise architecture might look like this: 

HUMAN EXPERTISE 

                       ↓ 

              KNOWLEDGE CAPTURE 

                       ↓ 

             VALIDATION & CURATION 

                       ↓ 

              GOVERNANCE LAYER 

                       ↓ 

        ┌──────────────┼──────────────┐ 

        ↓              ↓              ↓ 

      DATA          RULES         KNOWLEDGE 

        ↓              ↓              ↓ 

        └──────────────┼──────────────┘ 

                       ↓ 

                AI INTELLIGENCE 

                       ↓ 

                 CONTEXT ENGINE 

                       ↓ 

             RECOMMEND / ASSIST 

                       ↓ 

                HUMAN JUDGMENT 

                       ↓ 

                   OUTCOME 

                       ↓ 

                VALIDATED LEARNING 

                       ↺ 

Notice something important. 

AI is not the final authority. 

It sits within a governed system. 

 

23. The Business Benefits

Done responsibly, expertise-to-AI transformation can create significant organizational benefits. 

Faster onboarding 

New employees gain access to institutional knowledge more quickly. 

Better decision support 

Teams can access relevant historical information at the point of need. 

Reduced knowledge loss 

Critical organizational knowledge can become less dependent on individual memory. 

Greater consistency 

Approved policies and practices can be surfaced consistently. 

Increased productivity 

Employees spend less time searching for information. 

Better scalability 

One expert's validated knowledge can potentially support thousands of employees. 

Stronger organizational learning 

Lessons from previous work can inform future decisions. 

But these benefits depend on quality and governance. 

 

24. The Risk of "Digitalizing the Wrong Expertise"

One of the most important questions is: 

Are we capturing best practice—or merely existing practice? 

Organizations sometimes confuse: 

"This is how we've always done it." 

with: 

"This is the best way to do it." 

AI can make the distinction more difficult because it can scale historical patterns very efficiently. 

If the historical process is flawed, AI may simply make the flaw faster. 

Automation amplifies process quality. 

AI can amplify organizational assumptions. 

Therefore: 

Before scaling expertise, validate the expertise. 

 

25. The Metempsychosis Test

Before transforming human expertise into AI capability, ask seven questions: 

1. What knowledge are we transferring?
2. What context gives that knowledge meaning?
3. What assumptions are embedded in it?
4. What evidence validates it?
5. What should AI be allowed to do with it?
6. Where must humans retain authority?
7. How will the knowledge evolve over time?

If these questions don't have good answers, the organization may not be ready to operationalize the expertise. 

 

26. The Future: Organizations That Remember

One of the most exciting possibilities of enterprise AI is the creation of organizations that can remember. 

Not simply remember documents. 

But remember: 

• what worked;  

• what failed;  

• why decisions were made;  

• what risks emerged;  

• which assumptions proved wrong;  

• which approaches created value.  

Imagine an organization where every project contributes validated learning to future initiatives. 

Every customer interaction improves understanding. 

Every operational incident improves resilience. 

Every AI deployment contributes governance lessons. 

The organization develops a form of institutional memory that is continuously learning. 

 

27. But Memory Needs a Conscience

An organization that remembers everything isn't necessarily a responsible organization. 

It must also know: 

• what should not be remembered;  

• what should be deleted;  

• what should be restricted;  

• what should not be inferred;  

• what should not be automated.  

That is why: 

Data governance + AI governance + knowledge governance 

must increasingly work together. 

The future enterprise isn't simply an intelligent organization. 

It must be a governed intelligent organization. 

 

28. The New Leadership Responsibility

AI leaders have an opportunity to redefine what transformation means. 

Instead of asking: 

"How can AI replace this task?" 

Ask: 

"How can AI preserve, augment, and responsibly scale the expertise behind this task?" 

Instead of: 

"How many employees can AI eliminate?" 

Ask: 

"How many employees can AI empower to make better decisions?" 

Instead of: 

"How autonomous can we make the system?" 

Ask: 

"What level of autonomy is appropriate for the risk involved?" 

That is the difference between an AI implementation and an AI transformation strategy. 

 

Conclusion: From Silicon to Spirit 

The journey from human expertise to AI is not simply: 

People → Machines. 

It is: 

Experience → Knowledge → Context → Intelligence → Capability. 

Silicon provides the computational foundation. 

Data provides evidence. 

Business logic provides structure. 

Context provides meaning. 

AI provides scalable intelligence. 

And humans provide something that remains extraordinarily difficult to automate: 

judgment, accountability, purpose, and responsibility. 

That is why the metaphor of metempsychosis is useful. 

Human expertise can take new digital forms. 

But the objective should never be to transfer the "soul" of expertise into AI and walk away. 

The objective is to create a partnership where human expertise becomes more scalable without becoming less human. 

For project managers and AI leaders, the opportunity is enormous: 

Don't just digitize what your experts know. Understand why they know it, when it applies, how it should be governed, and where AI can responsibly amplify it. 

The organizations that master this transition may gain something far more valuable than another AI application. 

They may build an organization that can: 

remember → learn → reason → adapt → govern → improve. 

And perhaps that is the real journey: 

From Silicon to Spirit. 

Not replacing human expertise. 
Not freezing it in software. 
But giving it a responsible new life.