AI Cheaters
A deep dive into cheating and why it makes sense right now
I don’t reach out to people as much as I used to. Instead I’m talking to AI… actually several AI’s (Chat, Claude, Gemini).
Its not just me. And this isn’t new.
You’ll see this turning point around 2012-2013. This is the era where Instagram and Netflix took off. Everything become personalized and you could disconnect from the boring stuff around you to see something cool or interesting.
Cold Take: Everyone is talking to AI first, then people because AI is more interesting to them.
Why it matter: Students and workers get blamed for using (or not using) AI. And things are really misaligned right now (given the trend line). We need aligned action now.
To watch the behind the scene of how this post was made, watch me write this live: https://www.loom.com/share/f6fc3bcd0b8744ccb8d1926f231e1286
Tracing The Misalignment
The Pillars of Cheating
Let’s break down why cheating makes sense right now. First up, let’s understand the incentives:
Incentives: Pillar One
In this game we have teachers, learners, practitioners and admins. Each of them has their own incentives for (money and status). And each has an actual goal of what they wanna do.
There’s a big misalignment between administrators and teachers. Practitioners and learners support the goals of the teacher. However, the system is financed by the government which keeps admins focused on the goals of the state and financing authorities.
Administrators are losing students because why would anyone trust the system right now?
Time: Pillar Two
What keeps the current system is that it still owns the schedules of the teachers, learners and admins (and sparingly invite practitioners).
This is radically different in adult learning environments where the class/course does not own the schedule of the learner. The employer does. And this is part of the reason why MOOC completion rates are so low.
There are 3 levers that schools still have on the time of each player in the system that help it stay as the dominant control structure.
Scheduling
Entire calendars are planned around the logistics and resources of physically transporting and caring for minors. This massive coordination in resources like bus systems, school lunch programs, after school programs, travel programs and more. Schools are the bedrock of many communities. Its one of the last pieces of social glue that we have in America (and the world).
As a world, when we see young children in broad daylight during the workweek, we expect to see them in schools… not roaming the neighborhood.
School is so baked into our schedules, that any just-in-time learning can’t match daily compounding and societal conditioning (at least for this generation).
Frequency
School is a daily commitment for ~13 years of your life. This is also one of the big reasons why students struggle post-graduation without a formal continuity structure.
When I cover assessment design, I’ll talk more about why students aren’t prepared for the real world and propose a test and ceremony structure for the new AI world.
Depth
Teachers can keep asking you questions. They can see you in the hallway. They can hold you accountable. A teacher can refuse to accept the work of student if the student fails to cooperate. This is exceptionally important because the teacher-student relationship requires communication, looking the other in the eye and being able to share physical gestures (like a handshake) after crucial conversations.
AI can’t replicate that. And no amount of cognitive behavioral techniques will help AI prepare students for gracefully dealing with members of their community.
Conclusions About The Relationship Between Time & School
Time has memory. Location has energy.
Because school physically puts you into the same location at the same times for years at a time, you develop a temporal memory that school is what you should be doing.
AI, on the other hand, is when you disassociate with those around you, alone and quiet for hours.
For most students, they should be coming to humans first to be seen and guided until they are understand the differences between learning with AI versus learning with humans.
AI is winning the attention of administrators because you can hack rewards for the system (better state test scores). But students are gravely unprepared for what’s to come ahead. But what do I know, only time will tell.
Attention
Now, we understand the players of the game (i.e. teachers, learners, practitioners and admins). And we understand the misalignment between the traditional structure of a school versus an AI school. And why the admins support this, and why the teachers, students are defecting right now.
Now, I want to tell you a personal story about learning.
I want to talk about the conditions of learning, and specifically I want to tell you my story about learning through the life journey of tragedy, grief and acceptance.
WATCH FROM 7:34 - 10:40
My question to you answer you watch this section is:
“Are institutions systematically looking to exploit traumatized children?”
I’m curious to hear first hand accounts of what you’ve experienced and observed. Reply privately or comment.
Games
Now in this last section, I’m going to cover an experiment I’m doing for my AI Safety class. I’m taking a night class in Brooklyn to learn Research Engineering with a cohort of other students, no teachers. We’re following this open source curriculum together.
Cheating Is The Optimal Path Because Administrators Aren’t Aligned
In our game, we identified that the most misaligned players is the administrative coalition.
In the next issue, I’m going to share a series of characters that I made to hack AI tutors. And I track what makes them break. Our lead scorer who can break the system ~30% of the time is THE GASLIGHTER.
Do you recognize him?
Does anyone come to mind when you see him?
What is gaslighting?
A form of psychological manipulation in which someone makes another person doubt their own memories, perceptions, or sense of reality, usually over a period of time and often to gain power or control over them.
Next issue, we will cover why he’s so effective at breaking AI tutors and we’ll talk about how he doesn’t just exist in this simulation. But there are teachers, learners, practitioners and admins who all share similar qualities to him…
About Jai
Jai is a hacker and teacher. His goal for this 10 part series is to track AI Alignment and Safety for STEAM education. He currently taking a Research Engineer night class based on ARENA 3.0.
Meet Jai and other Curious Cats
We’re hosting AI for Science. Jai has some really cool demoes and we have a mix of research engineers, community health practioners, breathworkers, doctors, teachers and everyone who is aligned with learning and making the world a better place.
Tickets are free and we’re looking for sponsors!
Life is so rich,
Jai