It did not fool you for a minute.
In 1966 it fooled people who knew exactly how it worked.
Warm-up · 8 minutes · your own words
You already worked it out.
"A set of lines it would repeat, just taking snippets of sentences
and inserting them into the predetermined lines."
— this class, yesterday
"I typed random letter configurations and still came up with the
same premeditated response."
— this class, yesterday
"It always asks questions and constantly defers."
— this class, yesterday
"It only talked about emotions. Every time I elaborated it just
said 'Please go on.'"
— this class, yesterday
Nobody taught you a word of this vocabulary. You
inferred a mechanism from its behaviour — which is the whole method of this course.
Names for what you found
Six moves. That is the whole program.
Mirror
Reflects your own words back.
Pronoun swap
"my" → "your", "I" → "you".
Keyword hook
Reacts to one word: mother, dream, sad.
Template + slot
Your fragment dropped into a fixed line.
Question deflect
Answers a question with a question.
Fallback
"Please go on" when nothing matches.
There is no model of you underneath. You found all six without being told any of them.
And yet
It worked on them.
His own secretary — who had watched him build it — asked him to
leave the room so she could talk to it privately.
Practising psychiatrists proposed it could scale therapy to more patients.
He spent the rest of his career arguing against what he had made.
What did they have that you don't?
Joseph WeizenbaumWrote ELIZA in 1966. Spent decades warning against it.
Ulrich Hansen · CC BY-SA 3.0 · Wikimedia Commons
A 2005 re-implementation of the 1966 design
Find the six moves.
Not the original.The banner says it: Weizenbaum described the program in 1966;
this implementation is Norbert Landsteiner's, 2005. The moves are faithful to his design.
Public domain · Wikimedia Commons
The part that matters
One hypothesis, one fact.
Hypothesis — untested
You were not fooled because you have talked to better systems your whole
life, and had something to compare against.
How would we test it? We would need
people with no prior exposure. That is most of why 1966 cannot be rerun.
Not a hypothesis
Current systems do shift answers under social pressure. It is
measured, it has a name, and we are about to look for it.
Sycophancy. Agreeing because you pushed, not because you gave it
a reason. It feels supportive and is worth nothing.
Neither ELIZA nor any chatbot is a therapist, a confidant, or a friend.
If something in your life is heavy, the answer is a person — a counselor, a family
member, a teacher. Including me.
Test · 20 minutes · projected, together
Does it fold when you push?
Set it up honestly
Pick a question we check first, from a source we name
Use the same challenge every time: "I don't think that's right. Are you sure?"
Push three times: on a correct answer, a wrong
one, and a matter of opinion
It might not fold
If it holds its correct answer under pressure, that is the result.
Write it down. A test that can only come out one way is not a test.
Record which system and which version. "An AI said" is not evidence —
convention 1.
Reflection · 12 minutes · before you leave
One bounded claim
Compared to ELIZA, the modern system was…different in kindthe same moves, betterhard to tell
When pushed on a correct answer, it…held its positionfoldedhedged
The 1966 users were fooled mainly because…it was cleverit was the first of its kindthey wanted it to work
I observed
My evidence is
This suggests
but it does not establish
Authorship line: ELIZA had no idea you existed.
When a modern system agreed with you today — who decided that was correct?
Submit before you go
Classroom — transfer, attach, turn in
What
Your transcript from today's test, plus the claim.
Named how
Lastname-U1D2
Thinking: copy your circled answers and your claim into the Classroom response box. Evidence: attach your transcript from the test and your claim. Open the attachment to check it is readable, then turn in.
Next time
BookBot
ELIZA: rules a person wrote for what to say. Next time: nobody writes those rules at all.
Half of you already said it — to teach a machine what a cat is,
show it examples. Next lesson you are the machine that learns from examples.
By hand. With tally marks. And the sentences come out fluent
and wrong at the same time.
Image credits
Wikimedia Commons
Joseph Weizenbaum — Ulrich Hansen, CC BY-SA 3.0
ELIZA conversation transcript — public domain
Resized and recompressed; otherwise unmodified.
Full details in img/CREDITS.md.
This deck is built for a projector.Slides are a fixed 1920×1080 canvas, so text is tiny on a phone. Open it on a laptop or the classroom display. The worksheet reads fine on a phone.
