Why Abundant Answers Demand Human Discretion
Artificial intelligence can generate explanations, examples, arguments, summaries, calculations, hypotheses, images, code, and many other forms of output.
That capability is remarkable.
But generation itself does not settle what an output means.
Nor does it settle whether the output is correct.
Nor whether it is useful.
Nor whether it is worth teaching.
A useful distinction is therefore:
\[
\boxed{
\text{Generated}
\neq
\text{Correct}
\neq
\text{Useful}
\neq
\text{Worth Teaching}
}
\]
These are four different judgments.
And as the ability to generate grows, the responsibility to distinguish among them may grow with it.
1. The First Difference: Generated Is Not Known
A machine can produce an output.
That does not, by itself, make the output knowledge.
This is not a criticism of artificial intelligence. It is simply a distinction between generation and knowing.
A calculator produces a numerical result by applying specified computational rules to an input.
A generative AI system works differently. It can produce language, explanations, examples, hypotheses, code, images, and other forms of output through a probabilistic generative process. The result is not simply retrieved from a fixed catalogue of answers. It is generated from patterns learned across an enormous body of material.
This is part of what makes generative AI so powerful.
A single question can open into many possible responses.
The system can produce alternative explanations, different levels of abstraction, competing formulations, examples, analogies, counterexamples, and entirely different approaches to the same problem.
In that sense, generative AI does not merely produce an answer.
It can produce a space of possibilities.
But possibility is not knowledge.
Something has been generated.
The next question remains:
What should we make of what has been generated?
That question existed long before artificial intelligence.
AI simply makes it much harder to ignore because the space of possible outputs can expand very rapidly.
2. Generation Is an Expansion of Possibility
This is where the earlier idea of Ulta becomes relevant.
Generative technology expands.
A single question can produce:
- one explanation;
- ten explanations;
- different levels of explanation;
- competing interpretations;
- examples;
- counterexamples;
- analogies;
- mathematical formulations;
- historical perspectives;
- and entirely different ways of approaching the same problem.
This is extraordinarily useful.
It can help us explore a knowledge space that would otherwise take much longer to navigate.
But exploration is not the same as conclusion.
A larger field of possibilities does not automatically tell us which possibility deserves to become part of our understanding.
That requires another operation.
Discretion.
3. The Four Distinctions
Consider the four statements separately.
Generated
Something has been produced.
Correct
The produced statement survives appropriate checking.
Useful
The correct statement serves some purpose.
Worth Teaching
The useful knowledge deserves to be incorporated into another person’s learning at a particular time and in a particular context.
These are progressively stronger claims.
We can represent them as:
\[
\boxed{
\text{Generated}
\rightarrow
\text{Correct}
\rightarrow
\text{Useful}
\rightarrow
\text{Worth Teaching}
}
\]
But the arrows should not be understood as automatic.
They represent successive judgments.
A generated statement may be wrong.
A correct statement may be useless for the immediate problem.
A useful fact may still be inappropriate for a particular learner.
And something worth knowing may not yet be worth teaching.
4. A Correct Answer Can Still Be the Wrong Answer
Consider a student learning elementary mechanics.
The student asks:
What happens to the acceleration of a body if the net force is doubled while its mass remains constant?
A correct answer might simply say:
\[
a \rightarrow 2a
\]
That may be exactly what the student needs.
But imagine responding instead with a long discussion of Newtonian mechanics, relativistic limits, non-inertial frames, tensor notation, historical formulations, and experimental considerations.
The information might contain correct statements.
But it could still be the wrong answer for this student, at this moment, for this question.
This gives us another distinction:
\[
\boxed{
\text{Correct} \neq \text{Appropriate}
}
\]
Teaching requires the second judgment as well as the first.
5. A Useful Fact Is Not Necessarily Worth Teaching
There is another step that is even less obvious.
Suppose something is both correct and useful.
Does that mean it should be taught?
Not necessarily.
A medical student could theoretically benefit from knowing thousands of additional facts.
An engineering student could study hundreds of mathematical techniques.
A physics student could learn many historical formulations of a theory.
A mathematics student could encounter thousands of beautiful identities.
All of them may be valuable.
But education cannot simply become an inventory of everything that has value.
There has to be priority.
Some knowledge is foundational.
Some is supporting.
Some is specialized.
Some is useful only in particular circumstances.
Some is interesting but not necessary.
Some becomes important later.
Teaching requires making these distinctions.
6. The Teacher’s Four Judgments
A teacher encountering a new piece of information may therefore ask four different questions:
Is it generated?
If it came from an AI system, this may be obvious.
Then:
Is it correct?
This requires verification appropriate to the subject.
Then:
Is it useful?
Useful for what?
And finally:
Is it worth teaching?
Worth teaching to whom?
At what stage?
For what purpose?
With what consequences?
The fourth question contains the largest amount of judgment.
7. AI Can Help With All Four
There is no reason to assume that AI will remain confined to the first stage.
AI can already assist with verification.
It can compare sources.
It can identify inconsistencies.
It can test calculations.
It can generate counterexamples.
It can evaluate explanations against specified criteria.
It can identify prerequisites.
It can assist teachers in deciding among alternative explanations.
These capabilities may become much stronger.
We should allow that possibility.
The argument does not depend on assuming that machines will remain permanently weak at evaluation.
The more interesting question is what happens even if they become extremely good at it.
8. The One Dangerous Output
Consider an operational edge case.
Suppose an AI system generates one hundred pieces of information.
Ninety-nine are useful.
One is dangerous.
The system has therefore been extraordinarily productive.
But the one dangerous output matters.
In fact, it may require more attention precisely because it is hidden among ninety-nine useful ones.
Now imagine the system generates one million outputs.
If only one is seriously problematic, the numerical proportion may look reassuring.
But someone still needs to recognize that one.
This is not an argument that AI will necessarily produce dangerous information at any particular rate.
It is a structural observation:
The value of abundance depends partly on our ability to distinguish among its contents.
The better the useful outputs become, the easier it may become for an unsuitable output to appear alongside them without immediately attracting attention.
Discretion therefore does not disappear when usefulness increases.
It can become necessary more often.
9. The Larger Problem Is Not Error
The example of one dangerous output among many useful ones illustrates an important operational problem.
But it would be a mistake to make that the central argument.
The deeper issue exists even if the system becomes extraordinarily reliable.
Suppose AI eventually becomes much better at detecting factual errors, identifying contradictions, checking calculations, and comparing competing explanations.
The fundamental question would still remain.
What deserves our attention?
The problem is therefore not merely that AI can sometimes produce something wrong.
The problem is that correctness alone does not determine importance.
A correct answer can be irrelevant.
A correct explanation can be inappropriate for a particular learner.
A useful fact can be unnecessary at a particular stage.
And many things can be simultaneously correct, useful, and yet not worth teaching at that particular moment.
So the question is larger than:
Can we detect the bad output?
It is:
Among the enormous field of possible outputs, what deserves to become part of human understanding?
That is the fundamental pedagogical problem.
10. Everything Is Not Equally Useful
This may seem obvious.
But technological abundance makes it increasingly important.
Imagine two correct explanations.
One is slightly easier to understand.
Another is technically more complete.
One is ideal for a beginner.
Another is ideal for an advanced student.
One establishes a foundational relationship.
Another gives an interesting application.
Neither is necessarily the universally “best” explanation.
Usefulness depends upon purpose.
We therefore need to be careful with the language of optimization.
There may not be a single globally optimal answer.
There may be a right answer for a particular purpose.
This is one reason education cannot be reduced to maximizing information.
11. The Teacher Works With Relationships
A teacher does not encounter information in isolation.
A teacher encounters:
\[
\text{Knowledge}
+
\text{Learner}
+
\text{Purpose}
+
\text{Time}
+
\text{Context}
\]
The same piece of knowledge can have different value when any of these changes.
A theorem may be essential today and unnecessary tomorrow.
An example may be perfect for one student and confusing for another.
A sophisticated explanation may be appropriate at university and inappropriate in an introductory classroom.
A historical detail may be irrelevant to solving a problem but invaluable for understanding how an idea developed.
The information has not changed.
The relationship around it has changed.
That relationship is central to teaching.
12. When More Information Becomes More Responsibility
This brings us back to the larger argument.
If there are only a few available explanations, selection is relatively easy.
If there are thousands, selection becomes more important.
If there are millions, the ability to make good distinctions becomes even more important.
The growth of the knowledge space therefore does not reduce the need for teachers.
It changes the scale of their responsibility.
The teacher’s task is not simply to stand between the student and information.
It is to help establish which relationships within that information are worth carrying forward.
13. The Teacher Is Not the Owner of Knowledge
There is an important clarification here.
No teacher has ever possessed all knowledge.
No institution has possessed all knowledge.
No civilization has possessed all knowledge.
And there is no reason to assume that any future machine will possess a completed inventory called all knowledge.
Knowledge is not a finite warehouse waiting for some sufficiently powerful system to catalogue it completely.
The teacher has therefore never been the person who “had all the answers” and transmitted a selected portion to students.
The teacher’s responsibility has always been different.
It has been the responsibility to exercise discretion.
To determine what is worth learning.
To establish sequence.
To connect ideas.
To distinguish foundational knowledge from peripheral knowledge.
To help a learner make meaning from what is encountered.
AI does not remove that responsibility.
It may increase the amount of material over which that responsibility must operate.
14. The Scale May Change
The future may therefore produce a strange inversion.
Technology can make information increasingly accessible.
It can make explanations increasingly easy to obtain.
It can make alternative approaches increasingly abundant.
It can make the production of educational material increasingly inexpensive.
And yet the need for teachers may not fall in proportion.
It may grow.
Not necessarily because every additional piece of information requires another teacher.
But because every increase in available knowledge increases the importance of deciding what deserves human attention.
The pedagogical function may therefore become increasingly central as the knowledge space expands.
15. The Teacher’s Discretion Is Not a Filter at the End
There is a temptation to imagine a simple pipeline:
\[
\text{AI}
\rightarrow
\text{Information}
\rightarrow
\text{Teacher}
\rightarrow
\text{Student}
\]
But teaching is not merely a final quality-control stage.
The teacher can influence the question itself.
The teacher can change the sequence.
The teacher can recognize that the learner has misunderstood the premise.
The teacher can decide that a different example is needed.
The teacher can introduce a simpler model.
The teacher can deliberately postpone an advanced idea.
The teacher can decide that an apparently minor relationship is actually the foundation of everything that follows.
The teacher is therefore not merely checking the output.
The teacher is structuring the learning environment in which the output acquires meaning.
16. What Happens When AI Becomes Better?
This is where we should resist both pessimism and excessive optimism.
Perhaps AI will become extraordinarily reliable.
Perhaps its ability to verify, compare, rank, explain, and adapt will improve dramatically.
Perhaps many things that currently require human judgment will eventually receive substantial machine assistance.
All of that remains possible.
We do not need to close those possibilities.
But none of it changes the fact that technology exists within human purposes.
Even an extraordinarily capable system would still be operating in a world where people have to decide:
- what they are trying to accomplish;
- what they value;
- what they want to preserve;
- what they want to teach;
- what they want future generations to understand;
- and what should become part of shared human meaning.
The future may change how these decisions are made.
It does not follow that the decisions themselves disappear.
17. The Growing Responsibility of the Teacher
This brings us back to the central idea of this series.
The more capable our technologies become at generating possibilities, the more important it becomes to distinguish among them.
The more information becomes available, the more often discretion is required.
The more useful information becomes, the more difficult it may become to identify what is most worth carrying forward.
And the more capable machines become at assisting with those decisions, the more important it becomes to understand the purpose for which the assistance is being used.
This is not a retreat from technology.
It is an argument for taking technology seriously.
AI is for human beings.
Technology is for human beings.
Its value therefore cannot be measured only by how much it generates.
We must also ask what that generation allows human beings to understand, build, preserve, teach, and share.
Conclusion: What Is Worth Carrying Forward?
Artificial intelligence may become much more capable than it is today.
It may generate better explanations.
It may detect more errors.
It may compare more alternatives.
It may help teachers make better decisions.
It may even perform forms of selection that we currently associate strongly with human expertise.
We do not need to close those possibilities.
But we should keep one distinction clear:
\[
\boxed{
\text{Generated}
\neq
\text{Correct}
\neq
\text{Useful}
\neq
\text{Worth Teaching}
}
\]
The distance between those four terms is where discretion lives.
And discretion is not reduced by the existence of more information.
It becomes more necessary—and necessary more often.
The teacher’s responsibility has therefore never been to carry all knowledge.
It has been to help determine what among the vast field of what can be known is worth knowing, understanding, carrying forward, and making shared meaning from.
As that field expands, that responsibility does not become smaller.
It becomes larger.