Crystallized and Fluid Intelligence: AI vs Human Intelligence

 

Crystallized and Fluid Intelligence: AI vs Human Intelligence

AI inference has become dramatically cheaper. Stanford's 2025 AI Index reported that the cost of querying a model with GPT-3.5-level performance on the MMLU(Massive Multitask Language Understanding) benchmark fell by more than 280 times between November 2022 and October 2024.

That number represents more than cheaper software.

It shows how quickly artificial intelligence is moving from an experimental technology into an everyday intelligence tool.

But an important question remains:

Is artificial intelligence becoming intelligent in the same way humans are intelligent?

To answer that question, we need to look beyond AI benchmark scores and understand two important concepts from psychology: crystallized intelligence and fluid intelligence.

Humans use both.

AI systems are also beginning to demonstrate capabilities that look somewhat similar to both.

However, the similarities can be misleading.

AI can know enormous amounts of information without understanding the world like a human. It can solve difficult mathematical problems while occasionally making surprisingly simple mistakes.

So, how smart is AI really?

And how does the intelligence of systems such as GPT-6 Astra, Claude and Gemini compare with human intelligence?

What Is Crystallized Intelligence?

Crystallized intelligence is the ability to use knowledge, information and skills that we have learned through education and experience.

Think about something you learned several years ago.

You know that Paris is the capital of France.

You know what HTTP means.

You know how to calculate the area of a rectangle.

You know how to write a SQL query because you have used SQL many times.

When you answer these questions using knowledge you already acquired, you are using crystallized intelligence.

Simple example

Someone asks:

What is the capital of Japan?

If you already know the answer is Tokyo, you do not need to solve a new problem.

You retrieve existing knowledge.

That is crystallized intelligence.

Another example:

Who wrote Hamlet?

You do not need to reason from first principles.

You simply retrieve learned information.

Crystallized intelligence generally grows with education, vocabulary, professional experience and accumulated knowledge.

This makes it particularly interesting when we compare humans with large AI models.

What Is Fluid Intelligence?

Fluid intelligence is the ability to solve new problems, recognize patterns, reason and adapt when the solution is not already known.

Suppose somebody gives you a puzzle you have never seen before.

You cannot simply remember the answer.

You have to analyze it.

You may need to:

• Understand the problem.
• Identify patterns.
• Test different possibilities.
• Remove incorrect possibilities.
• Build a solution.
• Check whether the solution makes sense.

That is closer to fluid intelligence.

Simple example

Consider:

2, 4, 8, 16, ?

You identify the pattern and conclude that the answer is 32.

Now imagine a much more complicated puzzle where you have never encountered the same structure before.

Your ability to figure it out is more closely related to fluid intelligence.

This gives us an important distinction:

Knowing an answer and figuring out an answer are not the same thing.

And that distinction is becoming extremely important in artificial intelligence.

Crystallized vs Fluid Intelligence

Intelligence	What it means	Human example	AI equivalent
Crystallized intelligence	Using learned knowledge	Remembering historical facts	Learned model knowledge
Fluid intelligence	Solving unfamiliar problems	Solving a new puzzle	Reasoning
Experience	Applying past knowledge	Solving a familiar business problem	Patterns learned from data
Adaptation	Changing strategy	Responding to a new situation	Contextual reasoning
Knowledge retrieval	Finding known information	Remembering a formula	Generating or retrieving information

This comparison is not perfect.

Human intelligence developed through biology, physical experience, social relationships and years of interaction with the real world.

AI intelligence is created through training data, algorithms, reinforcement learning, computational resources and increasingly sophisticated tools.

Still, the framework is useful because it allows us to ask a better question:

How much of AI intelligence is knowledge, and how much is reasoning?

Artificial Intelligence and Crystallized Intelligence

Large language models contain an extraordinary amount of learned information.

During training, AI models process huge amounts of text, code, images and other data.

This creates something that resembles a machine version of crystallized intelligence.

Ask an advanced AI model about:

• History
• Programming
• Mathematics
• Marketing
• Economics
• Biology
• Finance
• Business strategy
• Literature

and it can often produce an answer within seconds.

This is one of AI's biggest advantages.

A human expert may spend decades developing expertise in one area.

An AI model can provide useful information across hundreds of domains.

But there is an important limitation.

A knowledgeable answer does not automatically prove genuine understanding.

AI can produce a convincing explanation and still make an incorrect assumption.

This is why AI hallucinations remain an important issue.

The model may appear extremely intelligent because its crystallized knowledge is enormous, while its reasoning can still fail under certain conditions.

Artificial Intelligence and Fluid Intelligence

The more interesting development is happening with reasoning.

Early generative AI was often judged by how well it could answer questions.

Modern frontier AI is increasingly judged by what it can figure out.

Models are now being developed to handle:

• Mathematical reasoning
• Software engineering
• Complex research
• Planning
• Data analysis
• Tool use
• Computer interaction
• Scientific problems
• Multi-step workflows

This represents a major change.

The question is no longer simply:

"What does AI know?"

It is increasingly:

"What can AI discover or solve when it encounters something new?"

That is much closer to the idea of fluid intelligence.

Knowledge vs Reasoning: The Real AI Intelligence Test

Consider two questions.

Question 1

Who discovered penicillin?

This primarily tests knowledge.

Question 2

A company has declining revenue, increasing customer acquisition costs and falling retention. Based on these constraints, what should management investigate first and why?

This is different.

The AI needs to:

Understand the situation.

Identify relationships.

Prioritize variables.

Consider alternatives.

Reason about possible causes.

Recommend an action.

Explain the reasoning.

The second problem is much closer to reasoning.

This gives us a useful framework:

AI capability = Knowledge + Reasoning + Context + Tools + Verification

The balance between these components determines how useful an AI system is.

GPT vs Claude vs Gemini: Which Is More Intelligent?

The competition between GPT, Claude and Gemini demonstrates why measuring AI intelligence is difficult.

There is no single number that perfectly measures intelligence.

A model can outperform another on coding but perform worse on a particular reasoning benchmark.

Another model may be stronger at multimodal tasks.

Another may be better at long-context analysis.

So rather than asking:

"Which AI is smartest?"

we should ask:

"Which AI is strongest for this type of intelligence?"

GPT-6 Astra: From Answering Questions to Doing Work

GPT-6 Astra is particularly important to this discussion because it represents a shift from conversational AI toward AI that can perform complex work.

OpenAI describes GPT-6 Astra as its most capable model and reports state-of-the-art performance in computer use, browsing, software engineering, cybersecurity, science and professional work.

OpenAI reports 98% on FrontierMath Tier 4 and 99.9% on ARC-AGI-3, although these are specific benchmark measurements rather than a universal measure of intelligence.

Astra can also work with computers, browsers and professional applications.

For example, OpenAI says Astra can handle tasks such as:

• Filling online forms
• Updating CRM records
• Conducting online research
• Creating websites
• Analyzing scientific data
• Generating plots
• Running frontend quality checks
• Troubleshooting software

This is important.

Imagine asking AI:

"How do I update this CRM?"

That is traditional AI assistance.

Now imagine saying:

"Find the customer, review their account, update the CRM record, check the result and prepare a summary."

That requires considerably more than knowledge.

It requires:

Understanding → Planning → Reasoning → Computer use → Verification

This is where the boundary between crystallized and fluid intelligence becomes particularly interesting.

Claude: Reasoning and Long-Context Knowledge Work

Claude has also become an important competitor in reasoning, coding and knowledge work.

Anthropic's current Claude systems emphasize complex reasoning, coding, long-context tasks and agentic workflows.

This makes Claude another useful example of how AI is moving beyond simple knowledge retrieval.

A long document may contain thousands of pieces of information.

The AI must determine:

• Which information matters?
• Which information conflicts?
• What evidence supports a conclusion?
• What information is missing?
• What should happen next?

That is not simply a memory problem.

It is a reasoning problem.

And it highlights an important principle:

More knowledge does not automatically mean better intelligence.

The ability to select relevant knowledge is equally important.

Gemini: Multimodal Intelligence

Gemini represents another dimension of AI intelligence.

Humans do not experience the world through text alone.

We see.

We hear.

We observe movement.

We recognize objects.

We combine visual information with language and experience.

Gemini's multimodal capabilities are therefore important when considering the future of AI intelligence.

A system that can combine:

Text + Images + Audio + Video + Context + Tools

has a much richer information environment than an AI system that only processes text.

This could become increasingly important as AI moves into robotics, scientific research, autonomous systems and real-world environments.

GPT vs Claude vs Gemini Intelligence Comparison

A high-level comparison looks like this:

Capability	GPT	Claude	Gemini
General knowledge	Very strong	Very strong	Very strong
Reasoning	Very strong	Very strong	Very strong
Coding	Very strong	Very strong	Very strong
Long-context work	Strong	Very strong	Very strong
Multimodal capability	Strong	Strong	Very strong
Tool use	Very strong	Very strong	Very strong
Computer/agent tasks	Very strong	Very strong	Very strong
Research workflows	Very strong	Very strong	Very strong

This should not be treated as a permanent ranking.

AI models evolve rapidly, and benchmark results measure specific capabilities rather than human intelligence as a whole.

How Smart Is AI Really?

This is where the discussion becomes more interesting.

AI can already outperform humans in specific tasks involving:

• Large-scale information processing
• Calculation
• Pattern recognition
• Information retrieval
• Certain coding tasks
• Certain mathematical problems
• Some benchmark reasoning tasks

But this does not mean AI has completely reproduced human intelligence.

There are still major differences.

Human Intelligence Includes Experience

Humans learn from physical and social experience.

A child touches something hot and learns not to touch it again.

A business owner loses money on a product and changes strategy.

A manager experiences a failed negotiation and develops better instincts.

A doctor develops judgment after years of observing patients.

These experiences become part of human intelligence.

AI does not experience the world in exactly the same way.

It can process descriptions of millions of experiences without actually living those experiences.

That distinction matters.

AI Can Reason and Still Make Strange Mistakes

One of the most fascinating characteristics of modern AI is that it can solve very difficult problems and still make surprisingly simple mistakes.

This creates an AI paradox:

Advanced capability does not guarantee universal reliability.

A model might solve a complex programming problem and then misunderstand an ambiguous instruction.

It may analyze a large dataset but overlook an obvious business constraint.

It may provide a sophisticated explanation containing one incorrect assumption.

This is why AI systems need evaluation and verification.

Why Explainability Matters

The problem becomes even more important when AI makes decisions.

Suppose an AI system predicts that a customer has an 82% probability of buying.

A manager may ask:

Why 82%?

Simply knowing the prediction is not enough.

The decision-maker wants to understand what influenced it.

This is where explainable AI becomes important.

Kovendo's guide on SHAP analysis and its sales and churn examples explores how SHAP can show which features pushed a prediction higher or lower.

This is an important connection between artificial and human intelligence.

Humans generally want reasons.

We do not simply want:

"Do this."

We want:

"Why?"

Explainability helps transform AI from a black-box prediction system into something that humans can evaluate, challenge and use.

Human Creativity vs AI Creativity

Another important question is creativity.

Can AI really be creative?

AI can generate:

• Stories
• Images
• Music
• Marketing campaigns
• Product ideas
• Software designs
• Business strategies

But human creativity has a different foundation.

Humans combine experience, emotions, goals, curiosity, cultural understanding and personal meaning.

AI generates outputs based on learned patterns and instructions.

This does not mean AI creativity is useless.

Quite the opposite.

AI can dramatically increase the number of ideas humans can explore.

Kovendo's article on the power of creative thinking explores how creative thinking connects with problem solving, machine learning and innovation.

The emerging model may therefore be:

Human creativity → AI idea generation → Human evaluation → AI refinement → Human decision

This is another form of hybrid intelligence.

AI Can Also Influence Human Intelligence

There is another side of the relationship.

Humans are not only using AI.

AI is increasingly influencing how humans think.

Recommendation systems decide what information we see.

Search engines decide which information appears first.

Social platforms personalize our feeds.

AI systems predict what content we are likely to engage with.

Kovendo's analysis of social media echo chambers and AI-driven reinforcement explains how machine-learning systems can analyze clicks, likes, shares, watch time and other signals to personalize what users see.

This creates a fascinating feedback loop:

Human behavior → AI learns → AI recommends → Human consumes → Human behavior changes → AI learns again

So the future of intelligence is not only about AI becoming smarter.

It is also about AI changing human decision-making.

AI + Human Intelligence: A Practical Case Study

Consider an e-commerce company.

The marketing team has:

• 250,000 customers
• Purchase history
• Website activity
• Email engagement
• Product preferences
• Cart behavior

A human marketing manager understands the business objectives.

AI analyzes the enormous volume of customer signals.

The workflow becomes:

Customer data → AI analysis → Customer segments → Prediction → Human strategy → AI execution → Results → Human review

This is not simply AI replacing a marketer.

The human provides:

• Business objectives
• Brand judgment
• Risk boundaries
• Strategy
• Final decisions

AI provides:

• Scale
• Pattern detection
• Data processing
• Prediction
• Automation

Kovendo's practical guide on deploying AI-powered digital marketing uses this type of framework: business problem → data → AI → workflow → experiment → KPI → financial impact → scale.

That is an excellent example of hybrid intelligence in practice.

From AI Tools to AI Decision Systems

The biggest change may be moving from individual AI tools to complete AI-powered systems.

Instead of asking:

"Can AI write an advertisement?"

companies will ask:

"Can AI improve the entire marketing decision process?"

That means connecting:

Customer data → AI segmentation → prediction → content → advertising → personalization → conversion → revenue analysis

Kovendo's AI-powered digital marketing growth strategy explores this broader approach, arguing that the competitive advantage is not simply using ChatGPT but connecting AI with customer data, automation, predictive analytics, experimentation and revenue attribution.

This idea applies far beyond marketing.

The same principle can be used in:

• Finance
• Healthcare
• Software engineering
• Manufacturing
• Cybersecurity
• Supply chains
• Scientific research
• Customer service

AI becomes an intelligence layer across the organization.

A New Definition of AI Intelligence

Perhaps we should stop thinking about AI intelligence as a single number.

Instead, consider an intelligence stack:

Level 1: Knowledge

Can the AI retrieve and generate useful information?

Level 2: Reasoning

Can it solve problems it has not directly encountered?

Level 3: Context

Can it understand the specific situation?

Level 4: Tool use

Can it use software, databases and external information?

Level 5: Planning

Can it break a complicated objective into smaller tasks?

Level 6: Adaptation

Can it change its strategy when something fails?

Level 7: Judgment

Can it recognize uncertainty and know when it should not act?

Level 8: Autonomy

Can it complete meaningful objectives with limited human intervention?

This is where models such as GPT-6 Astra become particularly interesting.

Astra is explicitly designed for multistep workflows involving reasoning, computer use, coding and professional work. OpenAI reports that it can preserve context across long coding sessions and handle changing requirements while continuing a task.

What Will AI Intelligence Look Like in the Future?

The future is unlikely to be determined by bigger models alone.

Five developments are particularly important.

1. Better reasoning

AI will become better at solving unfamiliar problems.

2. Better memory

AI systems will retain useful context across longer projects.

3. Better tools

AI will connect to enterprise applications, databases, browsers and specialized software.

4. Better multimodal understanding

AI will combine text, images, audio, video and real-world information.

5. Better agents

Instead of asking AI one question at a time, people will delegate complete objectives.

For example:

"Analyze our competitors, identify important pricing changes, update our report and alert me if something requires attention."

That is much closer to an intelligent assistant than a chatbot.

Will AI Become More Intelligent Than Humans?

The answer depends on how we define intelligence.

If intelligence means:

• Processing information
• Calculating
• Recognizing patterns
• Retrieving knowledge
• Performing certain reasoning tasks

AI is already superior to humans in some areas.

If intelligence means:

• Physical experience
• Emotional understanding
• Social judgment
• Common sense
• Conscious experience
• Lifelong adaptation
• Moral responsibility

the comparison is much more complicated.

AI and humans are not simply competing on one intelligence scale.

They have different strengths.

The Future Is Probably Hybrid Intelligence

The most important competition may not be:

Human vs AI

It may be:

Human + AI vs Human without AI

Imagine a scientist using AI to analyze thousands of research papers.

A developer using AI to inspect millions of lines of code.

A financial analyst using AI to identify patterns across thousands of companies.

A doctor using AI to compare a patient's information with a huge body of medical knowledge.

A CEO using AI to monitor competitors, customers and business performance.

In each case, AI provides scale.

Humans provide:

• Goals
• Context
• Judgment
• Responsibility
• Values
• Real-world understanding

That combination could be much more powerful than either intelligence working alone.

The Real Meaning of Crystallized and Fluid Intelligence in AI

Crystallized intelligence helps us understand what AI knows.

Fluid intelligence helps us understand what AI can figure out.

This distinction is becoming increasingly important as AI evolves.

Older AI systems were often impressive because of their ability to retrieve patterns from huge datasets.

Modern frontier systems increasingly combine:

Knowledge + reasoning + tools + planning + action

GPT-6 Astra is a strong example of this transition. OpenAI reports that it is designed for complex reasoning, computer use, software engineering, scientific work and professional workflows. (OpenAI)

Claude and Gemini are also pushing the boundaries of reasoning, multimodal understanding, long-context work and agentic capabilities.

But none of this means AI has simply become a digital human.

AI has a different form of intelligence.

It has extraordinary strengths and unusual weaknesses.

And that may be the most important insight of all.

FAQs

Is AI more intelligent than humans?

AI already exceeds humans in some specific tasks involving computation, information processing, pattern recognition and selected reasoning benchmarks. However, human intelligence remains broader, particularly in areas involving physical experience, social understanding, values, judgment and general adaptation.

What is the difference between crystallized and fluid intelligence in AI?

Crystallized intelligence refers to AI's learned knowledge and ability to use existing information. Fluid intelligence refers to its ability to reason through unfamiliar problems, identify patterns, adapt strategies and develop solutions when the answer is not simply retrieved.

Conclusion

The debate about artificial intelligence is moving beyond the question:

"Can AI think?"

A better question is:

"What type of intelligence does AI have?"

Crystallized intelligence helps explain AI's enormous knowledge base.

Fluid intelligence helps us understand its increasingly powerful reasoning capabilities.

GPT, Claude and Gemini demonstrate that modern AI can combine knowledge, reasoning, tools, multimodal information and increasingly autonomous workflows.

GPT-6 Astra takes this further by combining advanced reasoning with computer use and professional task execution. OpenAI's current evaluations show significant progress on mathematical reasoning, computer use, software engineering and other demanding tasks.

Yet AI is not simply a digital human brain.

It can solve a difficult problem and still make a simple mistake.

It can know more facts than an individual human while lacking human lived experience.

It can generate creative ideas while relying on patterns learned from data.

It can make predictions while still requiring humans to question and verify those predictions.

The future may therefore not belong to AI alone.

It may belong to people and organizations that learn how to combine crystallized human knowledge, fluid human reasoning, machine intelligence and human judgment.

The most important question of the AI era may not be:

"Will AI replace human intelligence?"

It may be:

"What can humans accomplish when human intelligence and artificial intelligence work together?"