PART II — UNDERSTANDING THE OTHER INTELLIGENCE
What AI Brings to the Relationship
Chapter 8 — What AI Does Exceptionally Well
AI does not have to think exactly like us to be extraordinarily useful to us. Its distinctive advantage lies in doing certain kinds of cognitive work at a speed, scale, breadth and repeatability that humans alone cannot easily match.
Imagine walking into a library containing millions of pages.
You have ten minutes.
Your task is to:
find recurring themes,
compare competing explanations,
extract important differences,
summarize the material,
suggest possible connections,
translate the findings for three different audiences,
and generate questions worth investigating next.
For a human being, this could require days, weeks or an entire team.
AI can sometimes help with parts of this process in seconds or minutes.
That does not mean AI has understood every page perfectly.
It does not mean its conclusions are necessarily correct.
And it certainly does not mean the human expert has become unnecessary.
It reveals something else:
AI has a different capability profile from human intelligence.
Understanding that profile is essential.
If Chapter 6 asked:
What kind of technology is AI?
and Chapter 7 asked:
Where do patterns appear?
Chapter 8 asks:
What does this other intelligence do exceptionally well?
1. Stop Asking Whether AI Is “Smarter”
We often ask:
“Is AI smarter than humans?”
The question sounds reasonable.
But it can be misleading.
Imagine asking:
“Is a calculator smarter than a mathematician?”
The calculator can multiply two enormous numbers faster than almost any human.
The mathematician can decide:
which calculation matters,
why it matters,
whether the assumptions are sensible,
and what the result means.
Which is smarter?
The comparison collapses several different capabilities into one word.
AI is similar.
Current systems can perform remarkably well on some tasks while remaining surprisingly unreliable on others. Stanford's 2026 AI Index calls attention to this “jagged” capability profile: leading systems can perform at extraordinary levels in competition mathematics while struggling with apparently simpler tasks such as reading analogue clocks. (Stanford HAI)
So the useful question is not:
Human or AI—which is more intelligent?
It is:
What is each exceptionally good at?
That question opens the door to collaboration.
2. AI Is Exceptionally Good at Scale
Imagine reading 10 customer reviews.
Easy.
Now:
Possible.
10,000?
Difficult.
One million?
No individual human can meaningfully inspect them all manually.
AI systems can process enormous amounts of machine-readable information and help identify:
themes,
similarities,
differences,
clusters,
recurring language,
and candidate anomalies.
This is one of AI's most important strengths:
PATTERN RECOGNITION AT SCALE
The human sees individual trees.
AI can help scan the forest.
But remember:
Seeing a pattern across a million records does not automatically tell us what the pattern means.
Scale is an AI advantage.
Meaning remains a joint problem.
3. The Customer Complaint Mountain
Imagine a company receives 50,000 customer comments.
A manager asks:
“Why are customers unhappy?”
A traditional approach might involve sampling comments manually.
But perhaps the sample misses an emerging issue.
AI can help organize the full collection into themes such as:
Delivery delay
Billing confusion
Difficult onboarding
Product reliability
Poor support response
Then it can compare:
this month versus last month,
one region versus another,
one product versus another.
Suddenly, an overwhelming pile of text becomes a navigable landscape.
That is valuable.
But the manager must still ask:
Are these categories accurate?
Are minority issues disappearing inside large categories?
Which problems matter most?
What evidence supports the conclusion?
What should we change?
AI converts:
VOLUME → STRUCTURE
Human intelligence must help convert:
STRUCTURE → MEANING → ACTION
4. AI Is Exceptionally Fast
Consider a teacher preparing tomorrow's class.
She wants:
five examples,
three misconceptions,
ten quiz questions,
two case studies,
and one alternative explanation for struggling students.
Producing all of this manually takes time.
Generative AI can produce drafts quickly.
That speed changes the economics of intellectual experimentation.
Previously, the teacher might think:
“I don't have time to create five different explanations.”
Now she can ask:
“Give me five ways of explaining this concept.”
The scarce resource changes.
Generating alternatives becomes cheaper.
Selecting the right alternative becomes more important.
So:
When generation becomes fast, judgment becomes more valuable.
5. AI Is Exceptionally Good at Generating Alternatives
Suppose you are designing a new service.
Ask a human team:
“Give me ten possible customer segments.”
They may produce several.
Ask AI and you may receive ten immediately.
Then:
“Give me ten completely different ones.”
Then:
“Now identify unconventional segments.”
Then:
“Generate a business model for each.”
This does not mean any of those ideas are good.
That is precisely the point.
AI can dramatically lower the cost of possibility generation.
We can express this as:
ONE PROBLEM → MANY POSSIBILITIES
This is especially useful during:
brainstorming,
planning,
writing,
design,
scenario exploration,
and problem solving.
6. Divergence Becomes Cheap
Human creativity often involves two broad movements.
DIVERGENCE
Generate possibilities.
CONVERGENCE
Select and develop the promising ones.
AI can be extremely powerful during divergence.
Give me:
20 headlines.
10 explanations.
8 strategies.
5 analogies.
3 alternative architectures.
4 counterarguments.
Previously, generating this variety could consume substantial human effort.
Now it can happen quickly.
But there is a danger.
If divergence becomes almost free, we can drown in possibilities.
Therefore:
AI expands. Humans must discriminate.
The better AI becomes at generating options, the more valuable our ability to choose among them becomes.
7. AI Is Exceptionally Good at Transformation
Give AI a paragraph.
Ask:
“Make this simpler.”
Then:
“Make it suitable for a CEO.”
Then:
“Explain it to a 12-year-old.”
Then:
“Convert it into five questions.”
Then:
“Turn it into a table.”
Then:
“Translate it.”
Then:
“Convert it into a presentation outline.”
The underlying information may remain similar.
The representation changes.
This is a major AI capability:
CONTENT → DIFFERENT REPRESENTATION
AI can transform:
long → short,
technical → simple,
prose → table,
notes → draft,
concept → example,
text → questions,
code → explanation,
and one language → another.
This is not merely a convenience.
Representation can change understanding.
8. The Same Idea, Five Ways
Suppose a student cannot understand overfitting in machine learning.
Ask AI:
“Explain overfitting technically.”
Still unclear.
“Explain it using a student memorizing examination answers.”
Still unclear.
“Explain it visually using a curve.”
Then:
“Give me a real-world example.”
Then:
“Give me a wrong explanation and ask me to correct it.”
The concept has not changed.
The representation has.
Sometimes learning difficulty is not:
I cannot understand this idea.
It is:
I have not yet encountered the representation through which I can understand it.
AI can generate representations cheaply.
That can make it a powerful learning partner.
9. AI Is Exceptionally Good at Recombination
Imagine you ask:
“Explain cybersecurity risk using the idea of a castle.”
AI can combine:
cybersecurity
with
fortification.
Or:
“Explain neural networks using an orchestra.”
Or:
“Apply the bottleneck pattern to university administration.”
Or:
“What can project management learn from air-traffic control?”
AI can explore connections among domains rapidly because its learned representations span enormous amounts of diverse material.
This makes AI especially useful for:
PATTERN TRANSFER
The connection may be insightful.
Or superficial.
So once again:
AI proposes.
Human judges.
10. AI Is Exceptionally Good at First Drafts
The blank page can be expensive.
Imagine writing:
a report,
proposal,
lesson plan,
job description,
email,
presentation,
or project outline.
Often the hardest moment is:
Where do I begin?
AI can turn:
nothing
into
something.
And something can be:
criticized,
reorganized,
corrected,
expanded,
shortened,
or rejected.
This creates an important shift:
BLANK PAGE → DRAFT → HUMAN IMPROVEMENT
AI does not have to produce the final answer to create value.
Sometimes its greatest contribution is simply eliminating the blank page.
11. Drafting Is Not Deciding
But this distinction matters enormously.
Suppose AI drafts a business proposal.
It may sound excellent.
Professional vocabulary.
Perfect headings.
Convincing arguments.
But:
Are the assumptions realistic?
Are the numbers correct?
Does the proposal fit the organization?
Are the risks understood?
Will employees accept it?
Is the recommendation wise?
A polished draft can create the illusion of completed thinking.
Therefore:
Fluency should never be confused with validity.
AI can be exceptional at expression.
The human must remain responsible for substance.
12. AI Is Exceptionally Patient
Imagine asking a human teacher:
“Explain this again.”
Then:
“Again.”
Then:
“Simpler.”
Then:
“Use another example.”
Then:
“I still don't understand.”
Human teachers may be wonderfully patient.
But they have:
limited time,
many students,
fatigue,
and other responsibilities.
AI does not become embarrassed because you ask the same basic question ten times.
This creates an important educational possibility.
A learner can say:
“Keep changing the explanation until I understand.”
That can reduce one barrier to learning:
fear of appearing ignorant.
13. AI Can Personalize the Interaction
Suppose two students are learning regression.
Student A understands mathematics but struggles with intuition.
Student B understands the intuition but struggles with equations.
A static textbook gives both largely the same page.
AI can respond differently.
To Student A:
“Let's build intuition before returning to the equation.”
To Student B:
“Let's derive the equation step by step from what you already understand.”
This is not perfect personalized education.
AI does not automatically know the learner accurately.
But conversational interaction allows explanations, examples and difficulty to be adapted rapidly.
That is a meaningful capability.
14. AI Is Exceptionally Good at Iteration
Human work often follows:
Create → Review → Improve
But review cycles can be slow.
AI can make iteration cheap.
Write a paragraph.
“Find weaknesses.”
Revise.
“Challenge the argument.”
Revise.
“Make the structure clearer.”
Revise.
“What would a skeptical reader object to?”
Revise.
Five cycles may happen in the time previously required for one.
This gives AI another major advantage:
FAST FEEDBACK LOOPS
And as Chapter 4 argued, feedback loops can support capability growth when the human remains cognitively engaged.
15. The Programmer's Loop
Imagine writing code.
Without assistance:
Write → Run → Error → Search → Read → Modify → Run
With AI:
Write → Run → Error → Ask → Hypothesis → Modify → Run
Potentially much faster.
AI can:
explain an error,
suggest likely causes,
generate test cases,
compare approaches,
or explain unfamiliar code.
Stanford's 2026 AI Index documents striking recent gains in autonomous software-engineering benchmarks; on SWE-bench Verified, frontier performance moved from roughly 60% toward near-saturation over a year. Yet benchmark performance should not be equated with flawless real-world software engineering. (Stanford HAI)
So the programmer's role changes.
Less time may be spent searching for syntax.
More attention can move toward:
architecture,
requirements,
testing,
security,
verification,
and judgment.
16. AI Is Exceptionally Good at Comparison
Humans often struggle when many dimensions must be compared simultaneously.
Imagine evaluating five proposals across:
cost,
time,
risk,
skills,
scalability,
and strategic fit.
AI can quickly create:
a comparison matrix,
trade-off summary,
decision criteria,
or alternative ranking schemes.
For example:
“Compare these proposals using cost, implementation difficulty, expected impact and reversibility. Do not recommend one yet.”
This last instruction matters.
First:
STRUCTURE THE COMPARISON
Then:
MAKE THE JUDGMENT
AI can help organize the decision space.
Humans should own consequential decisions.
17. AI Is Exceptionally Good at Summarization
Long documents create a familiar problem.
There is more information than attention.
AI can help:
identify themes,
extract key points,
compare sections,
produce executive summaries,
or answer questions about supplied material.
This can be enormously useful.
But summarization always involves compression.
And compression involves loss.
The summary may omit:
qualifications,
exceptions,
minority viewpoints,
methodological limitations,
or context.
Therefore:
A summary is a map of the document—not the document itself.
Use AI summaries to navigate.
Return to the source when the stakes demand it.
18. AI Is Exceptionally Good at Finding Candidate Patterns
Suppose you give AI:
customer comments,
survey responses,
meeting notes,
student feedback,
or incident reports.
Ask:
“What themes recur?”
AI can propose structures quickly.
This connects directly to Chapter 7.
AI can function as a:
PatternScope
It can help us move from:
many observations
to
candidate patterns.
But the word candidate matters.
AI is not an infallible pattern detector.
It can invent connections too.
So:
AI DISCOVERS CANDIDATES → HUMAN TESTS THEM
19. AI Is Exceptionally Good at Simulation
Imagine preparing for an interview.
You could read:
“Top 20 interview questions.”
Or say:
“Interview me for a product-manager role. Ask one question at a time. Challenge vague answers.”
Now AI becomes:
interviewer.
Preparing for negotiation?
“Play a difficult customer who thinks my price is too high.”
Learning medicine?
“Give me a fictional case and reveal information only when I ask appropriate questions.”
Teaching?
“Act like a student who holds a common misconception.”
Management?
“Simulate an employee who disagrees with my feedback.”
AI can create low-cost cognitive rehearsal.
That may be one of its most underused strengths.
20. Practice Can Become Nearly Unlimited
Historically, practice opportunities were scarce.
A student needed:
questions.
A salesperson needed:
customers.
A manager needed:
situations.
A speaker needed:
audiences.
AI can generate artificial practice environments.
Again and again.
Change difficulty.
Change context.
Introduce ambiguity.
Challenge assumptions.
Provide feedback.
This turns AI into the Intelligence Gym introduced earlier.
AI's ability to generate endless variation can become human practice capacity.
That is an extraordinary opportunity.
21. AI Is Exceptionally Good at Being a Challenger—If Asked
AI is often used as an agreeable assistant.
But consider:
“Find five reasons my strategy could fail.”
Or:
“Construct the strongest argument against my position.”
Or:
“What evidence would change my conclusion?”
Or:
“Pretend you are the CFO rejecting this proposal.”
Or:
“Find assumptions I have not justified.”
AI can cheaply create intellectual opposition.
That is valuable because human beings naturally risk becoming attached to their own ideas.
A useful AI does not always say:
“Here is your improved answer.”
Sometimes it should say:
“Have you considered this?”
22. AI Is Exceptionally Good at Question Generation
This may be more valuable than answer generation.
Suppose you are considering investing in a new project.
Ask:
“What questions should I answer before making this decision?”
AI might organize questions around:
market,
customer,
finance,
operations,
risk,
technology,
competition,
and execution.
Now the AI has not made the decision.
It has expanded your question space.
This is an important principle:
Intelligence is not merely having more answers. It is knowing what deserves questioning.
AI can help us discover those questions.
23. AI Is Exceptionally Good at Making Expertise More Accessible
Imagine a new customer-support employee.
Experienced colleagues have accumulated years of tacit knowledge:
how to diagnose problems,
how to phrase explanations,
how to calm frustrated customers,
how to navigate unusual cases.
Can AI help make some of those patterns more accessible?
A major field study by Erik Brynjolfsson, Danielle Li and Lindsey Raymond examined 5,179 customer-support agents using a generative-AI assistant. Access to the system increased issues resolved per hour by about 14% on average, with substantially larger gains among novice and lower-skilled workers and little benefit for the most experienced workers. The authors found suggestive evidence that AI helped diffuse patterns associated with stronger workers' practices. (National Bureau of Economic Research)
This is important.
AI may not merely automate expertise.
It may sometimes help distribute parts of expertise.
24. The Novice Can Reach Further
Imagine two new employees.
One learns only through trial, documentation and occasional access to senior colleagues.
The other has those same resources plus an AI assistant that can:
explain procedures,
surface relevant examples,
suggest next questions,
and provide immediate guidance.
The second employee may move through some parts of the learning curve faster.
The customer-support evidence suggests exactly this kind of heterogeneous effect: less-experienced workers benefited more than already high-performing workers. (National Bureau of Economic Research)
But we must distinguish:
performing better with AI
from
becoming independently expert.
Those are related possibilities.
They are not identical.
25. AI Can Lower the Floor
This gives us a powerful way to think about AI.
AI can sometimes raise the performance of people who have less experience.
The beginner writer gets a structure.
The junior programmer gets debugging assistance.
The new manager gets questions to consider.
The student gets another explanation.
The entrepreneur gets a first market-analysis framework.
AI can therefore sometimes:
LOWER THE ENTRY BARRIER TO CAPABLE PERFORMANCE
That can democratize access to useful cognitive assistance.
But there is another side.
If beginners outsource every difficult cognitive step, they may lose opportunities to develop the expertise required for independent judgment.
So the goal cannot simply be:
Make beginners perform like experts.
It should also be:
Help beginners become more expert.
26. AI Can Raise the Ceiling Too
Now consider the expert.
The experienced scientist already understands the field.
The programmer already knows the architecture.
The teacher already understands pedagogy.
The manager already possesses contextual knowledge.
AI may still help by:
expanding alternatives,
accelerating routine work,
searching larger spaces,
generating simulations,
or challenging assumptions.
The expert brings something the novice cannot:
a strong judgment filter.
This can make AI especially powerful.
The better your mental models, the better you may be able to:
ask,
interpret,
correct,
and redirect AI.
So AI does not necessarily make expertise obsolete.
In many situations:
Expertise makes AI more useful.
27. AI Is Exceptionally Consistent at Repetition
Imagine checking 5,000 items against the same defined criteria.
Humans become tired.
Attention fluctuates.
Boredom enters.
Machines do not experience fatigue in the human sense.
AI systems can apply the same computational process repeatedly at scale.
This can be valuable for:
screening,
classification,
format conversion,
routine comparison,
and first-pass review.
But consistency of process does not guarantee correctness.
A system can apply the same mistake consistently.
Therefore:
Consistency is valuable only when the underlying process is valid enough for the task.
28. AI Can Work Across Modalities
Modern AI increasingly operates across:
text,
images,
audio,
video,
code,
and combinations of them.
Stanford's 2026 AI Index documents rapid progress in multimodal understanding and reasoning alongside language, mathematics, coding and video capabilities. (Stanford HAI)
This creates powerful possibilities.
Upload a graph and ask for interpretation.
Provide a photograph and discuss visible features.
Give code and request explanation.
Provide text and generate a visual concept.
Combine a document with questions.
AI increasingly acts not merely as a text processor but as an interface across multiple information forms.
29. AI Can Traverse Domains Quickly
A human career usually develops depth in a limited number of areas.
AI can move conversationally from:
statistics
to
marketing
to
Python
to
history
to
biology
within seconds.
This breadth is extraordinary.
But breadth has a danger.
A confident answer in an unfamiliar domain can sound as polished as a reliable answer in a familiar one.
NIST explicitly identifies confabulation as a generative-AI risk: systems can confidently produce erroneous or false content because generation follows learned statistical distributions rather than guaranteeing factual truth. (NIST)
Therefore:
AI's breadth increases the importance of verification.
30. Fluency Is One of AI's Greatest Strengths—and Greatest Traps
AI can produce language that is:
clear,
structured,
confident,
polished,
and persuasive.
This makes it extraordinarily useful for communication.
But polished language triggers a dangerous human inference:
“This sounds right, therefore it probably is right.”
NIST warns that confabulated content can be especially risky precisely because it may be presented confidently and can even include fabricated reasoning or citations. (NIST Publications)
So we need a permanent distinction:
QUALITY OF EXPRESSION ≠ QUALITY OF EVIDENCE
Beautiful prose can contain a false claim.
Excellent formatting can contain bad reasoning.
Confidence can accompany uncertainty.
That is why JUDGE must follow ASK.
31. AI Is Not Exceptional at Everything
Understanding strengths requires understanding boundaries.
Current AI can still:
hallucinate,
misinterpret context,
fail on unusual cases,
produce inconsistent reasoning,
inherit biases,
misuse weak evidence,
or confidently answer when uncertainty would be more appropriate.
Stanford's 2026 AI Index documents striking unevenness: AI systems can reach extraordinary benchmark performance while still failing a substantial fraction of structured agent tasks and showing large hallucination rates on some evaluations. (Stanford HAI)
So:
Exceptional capability is not universal reliability.
This distinction may be one of the most important ideas in the entire book.
32. Think of AI as a Capability Portfolio
Instead of saying:
AI is intelligent.
Think:
AI has a portfolio of capabilities.
Strong in some contexts.
Weak in others.
A useful simplified portfolio might look like this:
AI IS OFTEN STRONG AT
Scale
Speed
Generation
Variation
Transformation
Comparison
Summarization
Pattern discovery
Simulation
Iteration
Repetition
Cross-domain exploration
AI STILL NEEDS CAREFUL HUMAN OVERSIGHT FOR
Truth
Context
Values
Consequences
Accountability
Novel edge cases
High-stakes judgment
This is a much healthier mental model than either:
“AI knows everything.”
or:
“AI is just autocomplete.”
Both miss the reality.
33. Match the Task to the Strength
Suppose you need to decide whether to close a factory.
Do not begin:
“AI, should we close the factory?”
Instead, decompose the problem.
Ask AI to help:
summarize operational data,
identify trends,
generate scenarios,
compare alternatives,
surface assumptions,
list stakeholders,
construct arguments for and against,
and identify questions requiring evidence.
Now AI is doing what it does well.
Humans remain responsible for what humans must own:
context,
values,
trade-offs,
people,
and consequences.
This is:
CAPABILITY MATCHING
Use the right intelligence for the right part of the problem.
34. The Human-AI Capability Map
Imagine a task containing four stages.
1. EXPLORE
AI strength:
generate possibilities.
2. STRUCTURE
AI strength:
organize information.
3. JUDGE
Human leadership:
evaluate evidence, context and consequences.
4. ACT
Human responsibility:
choose and own the decision.
So:
AI EXPANDS → AI STRUCTURES → HUMAN JUDGES → HUMAN ACTS
Not every task follows this exact sequence.
But it is a powerful default.
35. AI Can Expand Before We Narrow
Human beings often jump too quickly to the first plausible answer.
AI allows another strategy.
Before deciding:
Expand the space.
Ask:
What alternatives exist?
What assumptions am I making?
What perspectives am I missing?
What could go wrong?
What evidence would change my mind?
Then narrow.
This produces:
EXPAND → COMPARE → JUDGE → DECIDE
AI can make the first two stages extraordinarily cheap.
That gives human intelligence more material to work with.
36. AI Can Make Thinking Visible
Suppose you have a vague idea.
You explain it badly.
AI returns a structured interpretation.
You say:
“No—that isn't what I mean.”
Interesting.
The incorrect response has helped you clarify your own thought.
You revise.
AI responds again.
You revise again.
Eventually your own idea becomes clearer.
AI has functioned as a cognitive mirror.
Sometimes the value of AI is not that its answer is right.
It is that reacting to the answer helps us discover what we actually think.
37. AI Can Become a Pattern Amplifier
Chapter 6 gave us:
AI IS A PATTERN TECHNOLOGY
Chapter 7 gave us:
PATTERNS ARE EVERYWHERE
Now Chapter 8 adds:
AI CAN AMPLIFY OUR ACCESS TO PATTERNS
It can search wider.
Compare faster.
Generate alternatives.
Transform representations.
Simulate possibilities.
Surface candidate relationships.
But amplification works both ways.
AI can amplify:
good reasoning
and
bad assumptions.
Useful patterns
and
false patterns.
Good information
and
misinformation.
So amplification increases the need for human judgment.
38. The Multiplication Principle
Imagine:
AI capability = 10
and
Human judgment = 0
What happens?
Potentially:
10 × 0 = 0.
Now imagine:
AI capability = 10
and
Human judgment = 8
Potentially:
10 × 8 = 80.
This is not a mathematical model.
It is a metaphor.
But it captures something important:
Powerful AI combined with weak judgment can create powerful mistakes.
Whereas:
Powerful AI combined with strong judgment can create extraordinary leverage.
Therefore, as AI capability rises, human judgment should rise with it.
39. Think → Ask → Judge Becomes Capability Matching
Our core framework now becomes:
THINK
What am I trying to accomplish?
Which part requires:
generation?
analysis?
comparison?
judgment?
context?
responsibility?
ASK
Give AI the work that matches its strengths.
Generate alternatives.
Transform.
Compare.
Summarize.
Simulate.
Challenge.
Find candidate patterns.
JUDGE
Now apply human intelligence.
Is it true?
Is it relevant?
Is it supported?
What is missing?
What are the consequences?
What should we do?
Then:
ACCOMPLISH
Combine the best of both.
So:
THINK → ASK → JUDGE → ACCOMPLISH
The objective is not maximum AI use.
It is the right division of cognitive labour.
40. Don't Compete With AI Where AI Is Strong
If AI can generate 100 headline alternatives in seconds, why should a human spend an hour proving they can generate 100 manually?
Perhaps human effort is better spent deciding:
Which headline is truthful?
Which fits the audience?
Which reflects our values?
Which supports the objective?
Likewise:
If AI can summarize a routine document quickly, perhaps human attention can move toward implications.
If AI can produce code boilerplate, perhaps the developer can focus more on architecture and testing.
If AI can generate practice questions, perhaps the teacher can focus on diagnosing learning.
This is not surrender.
It is intelligent allocation.
41. Don't Surrender What Humans Need to Grow
But there is an equally important principle.
Suppose the student delegates every essay.
The programmer delegates every difficult bug.
The manager delegates every decision.
The researcher delegates every interpretation.
Performance may rise temporarily.
But what happens to capability?
The goal is therefore not:
Give AI everything it can do.
It is:
Give AI what it can do exceptionally well while preserving the cognitive work humans need for judgment, expertise and growth.
That distinction is central to Think and Grow Intelligent.
42. The Best Use of AI May Change With Your Goal
Suppose your goal is:
PRODUCTIVITY
Ask AI:
“Draft this.”
Suppose your goal is:
LEARNING
Ask:
“Don't draft it. Help me build it.”
Suppose your goal is:
JUDGMENT
Ask:
“Challenge my reasoning.”
Suppose your goal is:
CREATIVITY
Ask:
“Give me radically different possibilities.”
Suppose your goal is:
DECISION
Ask:
“Structure the alternatives, but don't decide for me.”
Same AI.
Different relationship.
The quality of the relationship depends partly on knowing which AI strength you want to activate.
43. From AI User to AI Director
A beginner asks:
“What can AI do?”
A more capable user asks:
“Which part of this task should AI do?”
An advanced user asks:
“Which AI capability should I invoke, what should remain human, and how will I verify the result?”
That is a significant progression.
USER → COLLABORATOR → DIRECTOR
The director does not need to perform every operation.
But the director needs enough understanding to coordinate the whole.
That may become a defining human capability in the AI age.
44. The Human Advantage Becomes Clearer
The more clearly we understand AI's strengths, the more clearly we can see what humans must cultivate.
If AI is exceptionally good at:
generation,
scale,
speed,
variation,
and pattern processing,
then humans need to become increasingly good at:
purpose,
framing,
context,
judgment,
values,
verification,
responsibility,
and meaning.
This is not:
HUMAN vs AI
It is:
HUMAN × AI
The question becomes:
What can this combination accomplish that neither could accomplish as effectively alone?
45. AI's Greatest Strength May Be Leverage
Perhaps the best word for AI's exceptional capability is:
LEVERAGE
One human can explore more alternatives.
One teacher can generate more practice.
One programmer can inspect more possibilities.
One manager can compare more scenarios.
One learner can ask more questions.
One entrepreneur can test more ideas.
One writer can iterate more drafts.
AI can multiply the intellectual territory a human can explore.
But leverage magnifies direction.
Push in the right direction:
more capability.
Push in the wrong direction:
more error.
Therefore:
AI leverage without human direction is incomplete.
46. From Artificial Intelligence to Augmented Intelligence
This gives us a more useful picture of the relationship.
Artificial intelligence can:
GENERATE
EXPAND
COMPARE
TRANSFORM
SIMULATE
RECOGNIZE
ITERATE
Human intelligence can:
PURPOSE
CONTEXTUALIZE
QUESTION
VERIFY
VALUE
JUDGE
TAKE RESPONSIBILITY
Together:
AI CAPABILITY + HUMAN JUDGMENT → AUGMENTED INTELLIGENCE
Not automatically.
Not in every task.
But as a design objective.
The Eighth Principle
The eighth principle of Think and Grow Intelligent is:
AI is exceptionally good at expanding the scale, speed and range of cognitive work—but its power becomes most valuable when paired with human purpose, context and judgment.
AI can process more.
Generate more.
Compare more.
Transform faster.
Simulate repeatedly.
Create endless variations.
Surface candidate patterns.
Help distribute useful knowledge.
And make sophisticated cognitive assistance available to far more people.
Evidence from real workplaces already shows that this can translate into measurable performance gains; in one large customer-support study, generative-AI assistance increased productivity by about 14% on average, with much larger gains among less-experienced workers. (National Bureau of Economic Research)
At the same time, contemporary evaluations show that extraordinary performance can coexist with surprising failures, and NIST explicitly warns that generative systems can confidently produce false or inconsistent content. (Stanford HAI)
So the lesson is not:
Trust AI because it is powerful.
Nor:
Avoid AI because it is imperfect.
It is:
Know what AI does exceptionally well—and design the human role around what still matters.
One Idea to Take With You
The next time you face an important task, do not immediately ask:
“Can AI do this?”
Instead, divide the task into two columns:
AI — WHAT CAN YOU DO EXCEPTIONALLY WELL HERE?
Generate.
Compare.
Summarize.
Transform.
Search for patterns.
Create alternatives.
Simulate.
Challenge.
Iterate.
HUMAN — WHAT MUST I OWN?
Purpose.
Context.
Evidence.
Judgment.
Values.
Consequences.
Decision.
Responsibility.
Then connect the two.
Do not ask AI to replace your intelligence. Design the task so that AI's strengths create more room for yours.
That is not simply using AI.
That is learning how to combine two different forms of capability.
And that is another step toward Thinking and Growing Intelligent.
Source-verification note
The strongest current evidence supports a capability-specific, rather than blanket, account of AI. Stanford's 2026 AI Index documents dramatic gains in coding, mathematics, multimodal reasoning and agentic computer use, while simultaneously documenting a jagged capability frontier and continuing reliability problems. (Stanford HAI)
The workplace-productivity example is grounded in the NBER study of 5,179 customer-support agents: AI assistance increased issues resolved per hour by about 14% on average, with much larger benefits among novice and lower-skilled workers. (National Bureau of Economic Research) NIST's Generative AI Profile provides the important counterweight: generative systems can produce confident falsehoods or inconsistencies, particularly in open-ended and context-dependent settings. (NIST)
Therefore, the chapter deliberately avoids saying that AI is “better than humans.” The evidence supports a more useful conclusion: AI is exceptionally strong at particular kinds of cognitive work, and the opportunity lies in matching those strengths with human judgment.
No comments:
Post a Comment