Teachers used to receive:
An unusually well-written assignment.
The first question might be:
“Was this written by ChatGPT?”
Then they’d start looking for:
AI detectors.
Word choice.
Sentence patterns.
Signs the student suddenly:
Improved their writing skills dramatically.
But now with the development of generative AI,
this cat-and-mouse game
faces a very practical challenge.
Students can:
Switch AI models.
Rewrite.
Manually polish.
Use humanizer tools.
Even have an agent:
Directly complete tasks within systems.
If teachers put all their energy into determining:
“Is this text AI-generated?”
they might still miss something more important:
“Has the student actually learned?”
Today,
some U.S. educators are starting to:
Shift the question.
Washington Post’s Observation on Changing Approaches: From AI Policing to Assessment Redesign
On September 6, The Washington Post interviewed a range of teachers, college professors, and writing instructors across the U.S.
These educators do not have a magical tool
to identify AI-generated work with 100% certainty.
Instead, they are beginning to:
Redesign the assignments themselves.
Rather than only reviewing the
final product submitted,
they are adding components
that reveal the learning process.
In other words,
showing how students arrived at their answers.
This shift goes beyond just ‘Allowing or banning ChatGPT’
Previously, many classes followed a simple assessment flow:
Teacher assigns a prompt.
↓
Student writes an essay.
↓
Teacher grades the final product.
With AI's emergence,
an invisible participant appears in the middle.
Teachers still see only the final product,
but students may have:
Thought for themselves.
Written independently.
Had AI help with revisions.
Or AI might have written the entire work,
with only minor edits by the student.
Some might even use an agent to
fully complete the assignment.
Both types of final submissions can appear polished.
So what fails isn’t just the AI detector,
but the approach of judging learning solely based on the final product.
University of Baltimore’s Writing Program Begins Changing This
The Washington Post reports that Rachael Zeleny, Director of University of Baltimore’s Writing Program, spent several months
redesigning the Advanced Writing Curriculum.
Her approach is not to pretend AI doesn’t exist,
but to give teachers visibility into
the students’ actual thought processes.
Recently, she spoke publicly about revising
assignments that are easily completed by AI.
She calls this direction:
“Artisanal AI.”
Focusing on:
Firsthand observation,
Experiential learning,
and how AI can support original thinking,
rather than replace the thinking students should do.
Assignments Now Require More than Just an Essay
An obvious change is that
students may no longer just submit
the final essay.
They might also submit:
AI conversation logs,
showing the AI interaction.
Handwritten outlines,
showing initial organization of ideas.
Reflections,
explaining their evaluation of the process.
Some classes even require:
Video reflections.
This means teachers see not just:
“Is the final 1,500 words well written?”
but rather:
“How did those 1,500 words come to be?”
Changing from ‘Cheating Detection’ to ‘Learning Evidence’
These are two entirely different logics.
Cheating detection asks:
“Did you use AI?”
Learning evidence asks:
“Can you prove you understand?”
For example, if a student submits
a very good market analysis,
the teacher doesn’t have to first wonder:
“Is this AI-generated?”
Instead, questions might be:
Why did you pick these three pieces of evidence?
Which AI suggestion did you disagree with?
What parts did you revise?
If you remove the third data point,
would the conclusion change?
Can you explain the core logic orally without notes?
These questions are much closer to true learning.
Using AI Does Not Necessarily Mean the Student Did Not Learn
This is important.
If a student truly uses AI to:
Find opposing arguments,
Organize materials,
Pose questions,
Challenge their conclusions,
and then personally:
Judge,
Fact-check,
Rewrite,
Make decisions,
they may still rely heavily on AI.
But they’re also engaging in critical thinking.
So simply counting
“how much AI was used”
hardly equals
“how much was learned.”
Conversely, Assignments Without AI May Not Prove Genuine Understanding Either
Students might:
Memorize answers,
Copy peers,
Follow templates,
Rote memorize.
So education should never merely assess
“whether a tool was used.”
The real focus is whether
students can:
Understand,
Analyze,
Select evidence,
Explain reasoning,
Apply concepts,
and handle new situations.
AI just makes ignoring this question impossible.
Supply Chain Professor Also Pushes ‘Final Answers’ Later in the Process
The Washington Post interviewed Andrew Zeiser,
a professor at John Carroll University.
He identified:
The real worry isn’t just whether students used AI,
but rather:
“Are students still learning?”
Some of his assignments require students to:
First complete a simulation involving real choices,
and then explain why they made those decisions.
This makes it hard for AI to simply:
Generate polished text
to replace the full learning process.
Why ‘Authentic Assessment’ Becomes Even More Important
Authentic assessment
can be understood as:
Assessments that closely mirror real tasks.
Not merely asking:
“Write 1,000 words about supply chain risk.”
But giving students a scenario of supply chain issues:
Costs,
Timelines,
Inventory,
Customer demands,
all in conflict.
Students must:
Make decisions,
Then explain why.
AI may help,
but teachers can see
how students make judgments.
University of Colorado Boulder’s Teaching Resources Follow Similar Paths
University of Colorado Boulder’s
AI Assessment Scale
lists AI-resilient assessment ideas including:
Socratic dialogue,
Debate,
Live demonstrations,
Personal/place-based experiences,
Real-world case studies.
All share the feature:
It’s hard to equate a polished final answer with total learning.
Teachers witness:
Process,
Responses,
Reasons,
Actual performance.
Essay Writing Remains Valuable
Absolutely.
Writing is thinking.
The problem comes if grading
only involves submitting an essay.
AI might then separate
the written product
from genuine thinking.
The new approach is not to stop writing,
but to add checkpoints like:
Topic selection,
Research,
Outlining,
First draft,
AI conversations,
Revisions,
Reflection.
The final essay
is just one step.
This Is Actually a Return to How Writing Always Was
Anyone who’s written research reports,
plans,
papers,
or articles knows:
Writing isn’t:
Opening Word and typing from start to finish.
The real work involves:
Posing questions,
Gathering information,
Cutting wrong directions,
Changing structure,
Revising evidence,
Overturning original ideas,
Writing anew.
AI only makes
“the final product”
too easy.
As a result, education is forced to
see again those crucial prior steps.
Tools like Brisk Also Shift from ‘AI Detection’ to ‘Process Tracking’
The Washington Post highlights Brisk.
Its Inspect Writing feature
doesn’t just give teachers a report like
“87% AI probability.”
Instead, Brisk’s current design
can play back the student’s writing process,
showing when text was added,
deleted,
copied and pasted,
and revised.
Brisk states clearly:
Inspect Writing is not about
an accuracy score,
but providing
a factual record of how the document was formed.
That Difference Is Huge
An AI detector tries to
guess the source.
Writing history looks at
what actually happened.
For example, if an essay suddenly has
1,200 words pasted at once,
that doesn’t mean
cheating for sure.
Maybe the student wrote it elsewhere first.
But at least the teacher can
ask questions.
Conversely,
spending two hours editing paragraph by paragraph
doesn’t prove
no AI was used.
But teachers then have
learning evidence
that’s far richer than a single AI score.
Brisk Even Suggests: Discuss Suspicions with Students
This is a vital boundary.
Inspect Writing isn’t designed for
“Zero the paper the moment you see a number.”
Instead, Brisk advises:
If the Writing Replay raises questions,
use it as a basis to start
an academic integrity conversation.
This is much more mature than
“Detector says 95%, so you cheated.”
Because AI Detection Is Stuck in a Never-Ending Arms Race
Teachers have detectors.
Students have humanizers.
Teachers look at revision history.
New tools can simulate gradual input of AI-generated text.
Next, agents could potentially
operate browsers,
learning management systems,
and assignment pages directly.
If education keeps focusing on:
“How to get one technology to catch another?”
it may never catch up.
Today's News Strongly Contrasts with Developments Just Days Ago
On September 3,
SasaDaily Morning Update reported
that New York City paused generative AI use for about 600,000 students.
Los Angeles Unified
also implemented tighter restrictions.
This shows that at the system level,
some still choose to hit the brakes first.
But today, The Washington Post shows a different path.
Some frontline teachers now ask:
“Since AI will inevitably enter classrooms, how do I redesign evidence of learning?”
These two approaches
are happening simultaneously.
So the Education Sector Has Not Suddenly ‘Accepted AI’
This must be clarified.
Do not interpret the news as:
All U.S. teachers now
welcome ChatGPT.
That is not true.
Different:
schools,
districts,
teachers,
ages,
and subjects
still have
vastly different policies.
Some:
ban it,
some:
restrict its use,
some:
allow but require disclosure,
and some:
integrate AI directly into courses.
Today’s real new signal is:
Some educators no longer believe solely catching AI use can solve the problem.
More Challenges Arise as AI Agents Make Traditional Assignments Easier to Circumvent
Previously, students had to:
Copy,
Paste.
Now, agents can:
Browse,
Research,
Fact-check,
Operate websites.
If in the future students only have to say:
“Log into my course platform and finish today’s quiz,”
then teachers face not just an AI essay,
but possibly
an entire automated assignment workflow.
This is why:
The “final answer”
will increasingly be insufficient.
The Most Valuable Learning Evidence Might Be a ‘Decision Trail’
For example, questions like:
What were your original thoughts?
Why did you change?
Which AI answer did you accept?
Which part did you reject?
What sources did you check?
Which evidence changed your conclusion?
If you did it again,
what would you do differently?
These things
can involve AI assistance,
but without genuine understanding,
students usually struggle to explain them clearly.
For Students, This Could Be a Fairer Approach
Students who use AI well
do not have to pretend
they never touched ChatGPT.
If permitted by rules,
they can say:
“I used AI
to find opposing viewpoints.
I used AI
to improve my English.
I used AI
to detect argument gaps.”
But then:
this is
my own judgment.
I checked this source myself.
I can explain the conclusion on my own.
This is far healthier than:
“Everyone secretly uses AI, and teachers secretly catch it.”
However, This Could Make Students' Work Harder
Previously,
handing in a final essay
was enough.
Now there may also be:
Outlines,
Drafts,
Source records,
AI transcripts,
Reflections,
Oral explanations.
In other words,
AI makes producing answers
easier,
but teachers might require
more
proof of learning.
That’s rather interesting.
AI does not necessarily make courses
simpler.
It could be quite the opposite.
Teachers Also May Not See Less Work
Redesigning assessments
takes time.
They must decide:
Where AI can be used,
Where it cannot,
What students must disclose,
What counts as evidence of learning,
How to grade,
How to protect privacy,
How to handle disparities in students’ access to AI tools.
This is not a case of
“Teachers accept AI and problems disappear.”
Instead,
work shifts from catching cheating to redesigning teaching.
AI Companies Enter This Market Directly
OpenAI,
Anthropic,
Google,
and many education AI companies
now aim to enter
schools,
teachers,
and students’
market.
For example, as previously covered by SasaDaily,
Claude for Teachers
isn’t just a chat box for teachers.
It starts integrating
lesson planning,
checking understanding,
curriculum design,
and teaching skills
into educational workflows.
AI is shifting from
a tool secretly used by students
to an officially adopted teaching tool in schools.
This also forces systems to reconsider
what truly defines
“their own work.”
BoodleBox Turns ‘AI Process’ Into Assessable Content
BoodleBox’s higher education features explicitly support professors in assessing:
The final product,
student-AI chats,
group projects,
and even
the process of using AI.
This approach deserves attention,
because it doesn’t seek to hide AI use,
but rather
turns
how AI is used
into
evidence of learning.
Two Students Using ChatGPT Could Show Vastly Different Learning
Student A:
Inputs the prompt,
copies the answer directly.
Student B:
Writes their own position first,
asks AI for opposing views,
researches and verifies,
finds their own argument flaws,
revises their points,
and finally writes their own version.
If you only ask,
“Did you use AI?”
Both say:
Yes.
But if you look at
the learning process,
the quality of their learning
is completely different.
This Is What Education Assessment Needs to Learn to Differentiate
AI Usage
is not
AI Dependency.
AI Assistance
is not the same as
AI Replacement.
What really matters is
where in the process AI
helps people think,
and where it
starts replacing learning a person should do.
There is no one answer
that fits all classrooms.
Different Subjects Will Handle This Differently
English writing classes may require students to
develop their own core arguments,
with AI only providing
feedback.
Programming courses may allow Copilot,
but students must
explain every line of code.
Language classes may permit AI
for grammar correction,
but oral assessments still
must be done live.
Business classes might allow AI to
conduct research,
but decisions,
trade-offs, and
recommendations
must be explained and defended by students.
So a mature AI policy
won’t look like merely
“AI banned.”
Or
“AI allowed.”
It’ll Most Likely Have Permissions for Each Step
For example:
Research: allowed.
Brainstorm: allowed.
Organizing ideas: allowed.
Writing the final answer directly: limited.
Citing false sources: not allowed.
Replacing implementation: not allowed.
Formal exams: not allowed.
Disclosure after use: required.
This approach is quite similar to how businesses adopt AI.
Not AI on or off,
but
step-by-step permissions.
This Is Also Highly Relevant Beyond Schools
Don’t think this is only
a teacher and student issue.
Businesses face the exact same challenges.
In the past, companies judged employees by
submitted
reports,
presentations,
code,
and plans.
Now AI can make polished
final products.
The real test
is whether the person
understands,
can judge,
spot errors,
take over without AI,
and adapt to new conditions.
So workplaces might shift from
output-based evaluation
to
process-based evidence.
Interviewing May No Longer Rely Solely on Portfolios
Portfolios look good,
but may involve heavy AI involvement.
Companies might:
Give candidates a problem on the spot,
let them
use AI,
then observe how they prompt,
choose,
reject answers,
verify,
and revise.
AI isn’t banned,
but
integrated into the test.
This reveals much more clearly
if the candidate can actually do the job.
This May Be the Most Important Educational Turning Point in the AI Era
For years, the question was:
“Can students use AI?”
That question remains important.
But now,
another question grows in importance:
“If students have access to AI, how do I know they truly learned?”
Once this question changes,
classrooms change too.
Assignments,
assessments,
exams,
homework,
and feedback
all may be redesigned.
The Most Important Takeaway Isn’t That AI Detectors Are Useless
AI detection,
writing history,
and exam security
still have uses.
The real need is to avoid treating any tool
as the sole arbiter of truth.
As SasaDaily wrote on August 30:
A short story that won the Commonwealth Short Story Prize
was once flagged as
100% AI-generated by detectors.
But the organizer settled the issue through:
drafts,
timestamps,
creative notes,
and a thorough investigation.
This story and today’s educational shift
both point to the same idea.
The Most Reliable Evidence Isn’t an ‘AI Percentage’
It’s
more cross-verified process evidence.
How did the assignment start?
How did the content evolve?
What were the sources?
How does the student explain?
Did they really understand?
Mature education in the AI era
may not be better at catching cheating,
but better at recognizing learning.
If you remember one thing, remember this:
When AI makes producing answers easier, education must redesign assessments to prove real learning—not just the answers.
Today, progress a bit with AI.
Learn one AI skill daily.
Save a little time daily.
Grow a little in ability each day.
SasaDaily, growing with you.
Recommended Reading
Why Did an Award-Winning Short Story Get 100% AI by Detectors Even After Organizers Verified It?
