AI Has More Knowledge Than Ever. But Is That Expertise?

Expertise is sometimes treated as another word for knowledge, skill, or competence. They are related, but they are not the same.

Knowledge provides an understanding of information, principles, concepts, and procedures.

Skills enable someone to perform particular activities.

Competence is the demonstrated ability to apply relevant knowledge and skills to perform to an expected standard.

Experience comes from encountering and dealing with situations over time.

Expertise  is the ability to apply that knowledge effectively in situations that are often complex, uncertain, or constantly changing.    Expertise develops when experience  progressively transforms how knowledge is organized, retrieved, interpreted, and applied.

Experts have typically encountered far more than routine situations.

They have dealt with exceptions, failures, unusual combinations of circumstances, conflicting information, and problems for which established procedures provide no straightforward answer. They have seen the consequences of decisions, received feedback, adjusted their understanding, and learned what matters in different contexts.

That accumulated experience enables experts to distinguish important signals from irrelevant information, anticipate consequences, identify exceptions, and make judgments when information is incomplete.

It also gives them something particularly important: a much richer understanding of when the usual answer does not apply.

An experienced quality engineer, clinician, plant operator, or technical specialist may therefore recognize that something is wrong before they can fully articulate why.

That is one reason expertise can be both difficult to replace and difficult to capture.

When experts encounter a problem, they draw on knowledge accumulated and transformed through previous experience.

They recognize potentially relevant patterns, retrieve associated knowledge, and assess whether it fits the current situation.

Where the pattern fits, existing knowledge can be applied. The outcome and consequences provide further experience and reinforce, refine, or challenge what is already known.

But sometimes the pattern does not fit.   Something is different.

The expert must identify what has changed, reconsider assumptions, assess possible consequences, and develop or adapt a solution.

That process transforms existing knowledge.

What has been learned then becomes part of the expert’s accumulated knowledge, ready to be retrieved, applied – or questioned when another problem arises.

Pattern recognition is one of the defining characteristics of expertise, but it is easily misunderstood.

It is not simply:    “I’ve seen this before.”

Accumulated experience gives experts a richer set of patterns against which to interpret a new situation. More importantly, it helps them recognize when the apparent pattern is misleading.

Expert pattern recognition can mean:

Recognizing a match – identifying a familiar pattern and retrieving relevant knowledge quickly.
Spotting a difference – recognizing that a familiar-looking situation contains an important contextual difference.
Identifying an exception – knowing when an established rule or usual response does not apply.
Detecting an anomaly or weak signal – noticing something that appears insignificant but may indicate a larger problem.
Judging incomplete or conflicting information – determining what matters when the evidence does not provide a straightforward answer.
Questioning assumptions  – recognizing that something being treated as fact may not be valid.
Reframing the problem  – recognizing that the problem itself has been incorrectly defined.

This last capability is particularly significant.   Expertise is not simply knowing the right answer. Sometimes expertise means recognizing that the obvious answer—or even the question itself—is wrong.

AI can already perform many activities once associated with substantial human expertise.

It can search and synthesize enormous bodies of recorded knowledge, identify complex patterns, retrieve relevant information, analyze data, compare alternatives, reason through problems, and generate potential solutions. In some defined tasks, AI systems already perform at or above the level of many human specialists.

And its capabilities will continue to advance.

AI also has an advantage no individual person can match: it can draw upon quantities of recorded knowledge that no human could acquire or retain in a lifetime.

But there is a catch.  Scale of knowledge is not the same as completeness of expertise.

AI depends on the information, examples, and feedback available to it. However extensive its training data becomes, it represents what has been captured – not necessarily what is true, what is relevant, or everything that people know.

AI can only work with the knowledge, data, context, and patterns available to it.   And those are incomplete.

Published information can be outdated, biased, contradictory, or simply wrong. Commercially valuable expertise may never be published. Organizations protect proprietary knowledge. And much expertise exists only in the accumulated experience of people working in particular contexts.

This exposes a critical limitation:   AI does not reliably know what it doesn’t know.

If an important variable, exception, experience, or piece of context is missing, AI can still identify the closest available pattern and produce a convincing response.

An expert may recognize something different.

The pattern doesn’t quite fit.  Something is unusual. A normally reliable assumption may not apply. Important information may be missing. A familiar-looking situation may differ in one critical respect.

Sometimes the expert response isn’t an answer at all – Something is missing or  we’re asking the wrong question.

This matters most when situations fall outside established patterns. Strong performance on familiar problems does not guarantee the same performance when something genuinely unusual occurs.

Nor does access to general knowledge provide the context accumulated by people working within a particular organization, process, environment, or system.

Organizations can give AI more of that context. They can capture expert knowledge, provide proprietary data, and build AI systems around their own information.

But then another question arises – Who identifies the knowledge worth capturing, the exceptions that matter, and what the AI is missing?

That still requires expertise.

The companies developing frontier AI provide an interesting test of the distinction between knowledge and expertise.

Anthropic and OpenAI have recruited specialists in areas including weapons, explosives, and chemical and biological risks to help evaluate what their models can do and prevent dangerous information from being made readily accessible.

AI makes substantial amounts of publicly available technical knowledge easier to find, combine, and apply. But the most advanced expertise may never have been available in its training data at all. It may be classified, commercially protected, unpublished, inaccessible, or held only in the experience of a small number of specialists.

Human experts are therefore needed to recognize both what the AI knows and what may be missing, identify dangerous capabilities and edge cases, and determine where safeguards are required.

The example exposes both sides of the expertise problem:

AI can make existing knowledge dramatically more accessible. But access to knowledge is still not the same as possessing the expertise required to judge its limits, consequences, or appropriate use.

A general AI system may have access to an extraordinary amount of knowledge about engineering, healthcare, manufacturing, finance, technology, or almost any other domain.

It does not follow that it possesses your organization’s expertise.

Every organization develops knowledge through experience.

People learn how particular systems behave, why processes evolved in particular ways, which exceptions matter, which apparently minor signals precede larger problems, which solutions have already been tried, and what happens when established procedures meet real-world circumstances.

Much of this knowledge never appears in a procedure, training course, database, or document.

And it does not necessarily reside with people who have “expert” in their title.

Critical expertise may sit with a highly qualified specialist. But it may equally sit with an experienced operator, technician, nurse, supervisor, salesperson, administrator, or anyone else who has accumulated deep knowledge of a particular task or context.

Today’s experts were once beginners.  They became experts by doing the work.

They applied theoretical knowledge to straightforward problems, observed experienced colleagues, made decisions, made mistakes, received feedback, encountered exceptions, and gradually tackled more complex situations.

Yet much of the work AI can most readily automate sits near the beginning of this pathway – Research. Analysis. Drafting. Routine problem solving. Preliminary interpretation.

These aren’t simply outputs organizations need produced cheaply. They are also how people learn.

If AI removes them, organizations may gain productivity while simultaneously removing some of the experiences through which future expertise develops.

Giving a novice the same AI used by an expert doesn’t solve the problem.

The expert brings accumulated knowledge and experience to the interaction. They are better equipped to challenge an output, spot an exception, detect a missing assumption, or recognize that an apparently convincing answer simply doesn’t make sense.

The novice may be the person least equipped to recognize when AI is wrong.

So there is a critical workforce question: If people no longer learn by doing the work, how will they acquire the experience required to become experts?

Automating work can also automate away learning opportunities.

Organizations have often allowed expertise to develop relatively organically. People learn the fundamentals, do the work, encounter harder situations, learn from experienced colleagues, and gradually become the people others turn to when something unusual happens.

AI may disrupt that pathway.

If developmental work disappears, organizations will need to deliberately create the experiences through which expertise develops.

Training isn’t enough. Neither is access to information.

Expertise requires people to apply knowledge, encounter variation and exceptions, experience consequences, receive feedback, and exercise increasing judgment in more complex situations.

That means organizations need to connect processes that have traditionally been managed separately:

• Capability requirements — defining the knowledge, skills, competence, and experience required for effective performance.
• Competency assessment – establishing whether people can actually apply what they know.
• Developmental experience – deliberately exposing people to different contexts, problems, exceptions, and increasing complexity.
• Coaching and observation – enabling developing practitioners to see not only what experts do, but how they interpret situations.
• Feedback and reflection – turning experience into learning rather than simply time served.
• Progressive responsibility – increasing autonomy and judgment as capability develops.
• Development and career pathways – creating progression through experiences that build deeper capability.
• Knowledge transfer and succession – identifying expertise that is vulnerable to loss and creating opportunities to transfer it.

But developing expertise also requires knowing where the organization stands today.

Where does critical expertise reside? Where is it concentrated in too few people? Who is developing toward it? What experience do they need next? Which expertise is vulnerable to retirement, resignation, or restructuring? And where might automation remove the very experiences needed to develop its successors?

These are workforce capability questions:

What’s expected?  What knowledge, competence, experience, and expertise does the work require?

Where are we now?   Who can demonstrate that capability? Where does critical expertise reside, and where are the gaps and risks?

What’s next?  What experience, development, coaching, career moves, and knowledge transfer will build the capability required next?