The evidence
GAiDFLY is built for one thing: students who own
their strategic reasoning. That takes practice until the reasoning holds
up under challenge, and faculty who can see that it does.
The need
A polished paper no longer tells faculty whether the student did the
thinking. AI did not lower the bar for strategic judgment. It raised it,
then hid whether the student cleared it.
The evidence that strategy education has to change keeps mounting,
and it points at one gap. Written work has stopped reliably showing who
did the thinking. AI prose is fluent and technically accurate. The oral
exam, as most programs run it, tests coverage, and AI covers well. The
harder problem is deeper than detection. Strategy programs now have to
produce leaders who work with AI as a matter of course: people who use
AI’s expanding capability to search, simulate, and test more
frames than one person could alone, and who still own the goals, the
frame, and the final judgment. That ownership is built through
friction. The struggle to define a problem before a structure arrives,
the first attempt at an argument that does not hold together, the claim
that fails under challenge in seminar. Those are the events that form
judgment, and AI removes them as readily as it removes the clerical
work around them. The finished product still matters. It just carries
less of the evidence, and something else has to carry the rest.
- A standard oral exam can be passed without judgment.
In February 2026, Army War College faculty gave ChatGPT, Gemini,
Claude, and Grok the college’s real ninety-minute oral
comprehensive exam, with no access to the curriculum, and all four
passed. What separated them from a student was the examiners’
impromptu follow-up: pushed past the prepared ground, the models
retrieved rather than constructed, misjudged their own performance,
and gave way. The exam only measures ownership at the point where the
examiner pushes.
- Faculty have stopped trusting the paper. A senior
war-studies professor at Marine Corps University reports that AI use
on papers cannot be monitored, that an AI taught his rubric matched
his grades on thirty papers, and that he no longer grades the paper
as evidence of the student’s thinking.
- Students mistake fluency for their own thinking.
Researchers studying student AI use describe epistemic confinement:
the student believes they are thinking independently while the frame
doing the work was never theirs.
- Students cannot see where AI stops being reliable.
In a field experiment with 758 consultants, AI raised quality on
tasks inside its competence and made people 19 points less likely to
be right on a task just outside it. The boundary was not visible in
advance.
- Confidence does not track correctness. A review
of 22 studies found that how much people trust an AI does not predict
whether they can tell its right advice from its wrong advice.
- An AI that gives answers weakens the student.
Nearly a thousand high-school students with a plain AI tutor scored
higher in practice and 17 percent lower than students with no AI once
it was removed. A version that refused to give answers avoided most
of the harm.
- Judgment and machine output have to check each
other. When the Marine Corps Command and Staff College gave
planning teams an AI model, the gains faded unless the team was
reorganized so that human judgment and machine output tested each
other.
- Spoken argument shows more. In nearly a thousand
student debates with an AI adversary, students who debated by voice
explored a wider range of ideas than those who typed. Evaluators who
hear a person they disagree with judge them more thoughtful than
evaluators who read the same words.
GAiDFLY uses AI to build the capability the evidence says matters
most: a student who can take a position, defend it under questioning,
know where their own understanding runs out, and revise without handing
the frame away. It puts the AI on the examiner’s side of the
table. The student argues a position out loud and the AI pushes back
the way a good examiner does, with new information, a counter to
something they got right, and a question past where they prepared. This
is AI supplying the friction that the rest of a student’s AI use
removes. It is also the work they will do after they graduate, and most
of it is spoken: briefing a commander, defending a recommendation to a
skeptical colleague.
Faculty get more than a record. The debrief shows them, for each
student and across the course, which positions held under pressure,
where the reasoning gave way, and what changed between the opening
argument and the closing one. That tells them where to spend seminar
time, what to push on in the real oral exam, and which students are
performing fluency rather than owning the frame. They set the question,
they read the evidence, and they keep the verdict.
The bottleneck
The most reliable way to verify that a student owns an argument is
likely oral defense, sustained questioning by someone who has read the
work. It is also the least scalable thing faculty do. One examiner, one
student, one hour. GAiDFLY takes the thousand hours of sparring and
leaves the verdict with the humans. Every student gets the experience of
being tested and challenged, as often as they need it. Faculty get a way
to see, across a whole course, how each student held up under that
pressure, which no seminar schedule allows.
The method
A GAiDFLY session follows a very old shape. The student commits to a
position and argues it first, so the frame stays theirs. GAiDFLY then
argues the other side, one short challenge at a time. Arguing both sides
of a hard question is an old exercise. An anonymous Greek text called
the Dissoi Logoi laid out paired cases for and against, and Roman
rhetorical schooling made arguing in utramque partem, on either
side, a standard drill.
GAiDFLY’s probing moves come from the Socratic elenchus,
the cross-examination that tests whether a speaker’s own
commitments hold together. The short turns are Socrates’ rule too.
In Plato’s Protagoras he refuses to go on unless the
long speeches stop, because only short exchanges can be examined.
The session ends with the student, not the examiner. They say where
they have come out and restate their case as they would argue it
tomorrow. GAiDFLY does not decide who won. In Plato’s terms it
practices dialectic, argument in service of testing, and leaves eristic,
argument in service of winning, alone. The judgment stays with the
student and their instructor.
What the debrief does
After the session GAiDFLY writes a debrief for student and
instructor: the ideas explored, the pressure points and how the student
handled them, and what changed between the opening argument and the
closing restatement. Scoring is not the point. The exercise is, and the
evidence it leaves behind. The debrief reports what happened so the
student can see their own reasoning and the faculty member can judge it.
GAiDFLY assigns no grade. The standard of proof stays human.
