Remy had a lantern named Beacon.
Beacon was the smartest thing Remy owned. It could light up a whole dark room, find a lost shoe under the bed, and tell a maple leaf from an oak leaf faster than you could blink.
But Beacon could not do one thing.
It could not decide where to look.
“Point me,” Beacon would say, in its warm, flickery voice. “I can’t choose where to shine. That part is yours.”
So Remy pointed. And wherever Remy pointed, Beacon lit it up and told Remy what it saw.
That was the deal. Remy was the hand. Beacon was the light.
One Saturday, the town set up a fair on the green. And in the middle of it all was a brand-new machine, tall and proud, with a hundred little lights blinking across its front.
A sign said: THE SPOTTER. IT SEES EVERYTHING.
“Step right up,” boomed the Spotter. “I am the smartest machine ever built. I can spot a person anywhere. I have never been wrong.”
A man stood beside it, not really watching. “It runs itself,” he said, waving a hand. “I just turned it on. Don’t ask me — ask the machine.”
Beacon dimmed a little, the way it did when something felt off.
“Remy,” it said quietly. “That machine is a light. Like me — the same kind of thing. But nobody’s holding it.” Beacon flickered. “And it says it’s never wrong. I’ve never met anything that’s never wrong. Not even me.”
“Test me!” boomed the Spotter. “Try to walk past without me spotting you. You can’t. I spot every person, every time.”
Kids tried. They tiptoed. They crawled. They walked backward.
“PERSON SPOTTED,” said the Spotter, every time. “PERSON SPOTTED. PERSON SPOTTED.” Its lights flashed. The crowd clapped.
Remy frowned and thought.
“Beacon,” Remy whispered. “How does it know what a person is?”
“The same way I learned a maple leaf,” said Beacon. “Somebody showed it. Over and over. Pictures of people, until it learned the shape.”
“What kind of pictures?”
Beacon flickered, thinking. “People… standing up. Walking. The ordinary way.”
Remy’s eyes went wide. “So it didn’t learn people. It learned people standing up.”
“If it only ever saw people standing up,” Remy said slowly, “then it doesn’t really know what a person is. It knows what standing up looks like.”
“That sounds right,” said Beacon. “And it doesn’t know that it doesn’t know. That’s the tricky part.”
Remy grinned. “Let’s fool it. Not to be mean. To see if we can.”
Remy backed up. Took a breath. And instead of walking past the Spotter—
Remy did a somersault.
Tumble, roll, tumble — right across the front of the great machine.
The Spotter’s hundred lights blinked. And blinked. And… said nothing.
“…NO PERSON DETECTED,” it announced. “THE AREA IS CLEAR.”
The crowd gasped.
“But I’m right here!” laughed Remy, popping up. “You just couldn’t see me, because I wasn’t standing up the way your pictures showed you!”
“…THAT IS NOT POSSIBLE,” said the Spotter. “I AM NEVER WRONG.”
“You’re not lying,” said Remy, kindly. “You just only know what someone showed you.”
So the other kids tried too.
Two of them hid under a big cardboard box and shuffled past like a wobbly brown turtle.
“…NO PERSON DETECTED.”
Three more crept by inside a fir tree they carried between them, giggling.
“…JUST A TREE,” said the Spotter. “A VERY WOBBLY TREE.”
The man who’d turned the Spotter on came running. “It’s broken! The machine is broken!”
“It’s not broken,” said Remy. “It’s doing exactly what it learned. It just never learned that a person could be clever.”
Beacon glowed soft and warm, drifting close to the great machine — the same kind of light it was. “It’s very smart,” Beacon said. “The only difference between us is that someone holds me. No one was ever holding it.” A gentle flicker. “And smart isn’t the same as understanding. Or the same as being right.”
Remy looked up at the great machine, with its hundred proud lights and nobody really watching it.
“The problem isn’t that it’s smart,” Remy said. “The problem is that everybody believed it, and nobody was holding it.”
Beacon tilted toward Remy, waiting, the way it always did.
“It’s smart,” said Remy. “But who’s holding it?”
That night, Remy pointed Beacon at the ceiling, where they’d stuck glow-in-the-dark stars.
“You get things wrong too,” Remy said.
“All the time,” said Beacon cheerfully. “That’s why you’re the one who decides. I just help you look.”
Remy smiled, and aimed the light at the brightest star.
“Brightest isn’t the same as truest,” Beacon added softly.
“I know,” said Remy. “That’s my job.”
The End
Think About It
The Spotter wasn’t lying — it really believed it was never wrong. How can something be sure of itself and still be wrong? Have you ever been really sure about something that turned out wrong?
Remy beat the machine by doing something it had never seen — a somersault. What’s something you can do that a machine might not expect? Why is being surprising a kind of being clever?
Remy said the real problem was that “nobody was holding it.” What do you think that means? Who should be in charge when a smart machine makes a decision — the machine, or a person?
The big idea to carry: A machine can be very smart and still be wrong, because it only knows what someone showed it. That’s why a person always has to stay in charge.
And the question for every smart machine you ever meet: “It’s smart. But who’s holding it?”
For the Grown-Ups
The somersault is a true story.
A real defense-research project trained a system to detect approaching people. A squad of Marines beat it easily — by somersaulting toward it, by shuffling past hidden under a cardboard box, by two of them advancing inside a fir tree. The machine had been trained on images of upright, walking humans, so that is the only thing it had really learned to recognize. It hadn’t learned people. It had learned people standing up. And crucially, it had no idea that there was anything it didn’t know.
That gap has a name worth teaching your kid as they get older: a model learns from the examples it was shown, and it can be confidently, completely wrong about anything outside them. AI doesn’t “understand” in the way a person does — it finds patterns. Patterns are powerful and patterns are narrow. The Spotter’s certainty — “I am never wrong” — is the most dangerous part, because confidence is not the same as truth, and a tool that can’t tell you when it’s unsure will state a wrong answer in exactly the same proud voice as a right one.
This is not a reason to fear these machines. It’s a reason to stay in the driver’s seat. The most useful habit you can give a child growing up surrounded by AI is not fear and not blind trust — it’s the third thing: discernment. Use the tool, question the tool, and never hand it the one job that has to stay human: deciding, and being responsible for the decision.
That’s why the book ends where it does — not with a kid defeating a machine, but with a kid realizing the real problem was that everyone believed it and no one was holding it. Keep asking your kid the series question, about homework helpers and chatbots and whatever comes next:
“It’s smart. But who’s holding it?”
Source note: the detection-defeating Marines anecdote is widely recounted in writing on machine perception and its brittleness (notably in Paul Scharre’s Army of None). The broader point — that machine-learning systems generalize poorly outside their training distribution and signal false confidence — is a core, well-documented feature of how these systems work.
Who We Are
Project Milk Carton is a veteran-founded 501(c)(3) nonprofit (EIN 33-1323547) with one mission: transparency in child welfare — and getting missing kids back home.
We’re not a think tank. We’re not consultants. We build the tools and we do the work:
A research engine tracking 340 million+ public records and $148 billion+ in government grants — following the money through the child-welfare system that nobody else is willing to map.
Missing children awareness — modern “milk cartons” for the digital age, putting kids’ faces back in front of people who can help.
Free, open reporting and tools at projectmilkcarton.org — research we hand to families, advocates, and investigators at no cost.
Original investigations and journalism — the work you read right here, every week.
All of it free. All of it public. That’s the whole point.
The Math That Could Change Everything
There are 21,000 of you reading this.
If each of you gave just a dollar or two a month, you would fund this entire mission — twice over. Even if only half of you did, you’d cover our whole operating budget for the year.
That’s it. That’s the secret. A handful of change from a lot of good people keeps this whole thing alive.
Two Ways to Give — Please Read This Part
We’re going to be completely straight with you, because transparency is literally our mission. Where you give changes how much actually reaches the work.
💚 Best for monthly support → right here on Substack
Substack takes the smallest cut of anyone, so the most of your gift goes straight to the kids and the mission. This is the way that helps us most:
→ Tap “Subscribe” at the top of this page and choose a paid plan.
A few dollars a month — or whatever you can give — on a recurring basis is what lets us actually plan ahead and keep the servers, the research, and the milk-carton campaigns running 24/7. Prefer to give just once? You can do that here too.
💳 Best for larger / one-time gifts → PayPal
Want to make a bigger one-time donation? PayPal is quick and easy:
(Honest note: processing fees take a bigger bite on PayPal, so for small recurring gifts, supporting us on Substack puts more of your money to work. For larger gifts, PayPal’s the simplest route.)
Either way — every cent is tax-deductible, because we’re a registered 501(c)(3).
Where Your Money Goes
We run lean, and we run transparent. Your support covers:
Servers & infrastructure — hosting the research database and keeping the tools online around the clock
Research & development — building the technology that makes child-welfare money traceable in the first place
The website & free public reporting at projectmilkcarton.org
Missing children campaigns — getting faces and names in front of the public
Our all-in operating budget is roughly $120,000 a year — and we account for it openly. No executive salaries chasing headlines. No bloat. Veterans and volunteers, doing the work.
Can’t Give Right Now? You Can Still Help.
Subscribe — it’s free. Never miss a report, and grow the audience that keeps this loud.
Share this post. Forward it. Every new reader is another set of eyes on the mission.
Tell one person why this matters.
The Bottom Line
We’ve given this away from day one because we believe families deserve the truth for free — and we always will. We’re simply asking the people who’ve been reading all along to help us keep going.
A few dollars a month on Substack. A larger gift through PayPal. Whatever you can. All of it tax-deductible. All of it straight into the work.
💚 Support monthly on Substack · 💳 Give a larger gift via PayPal
Thank you for being here. Thank you for caring about these kids. It matters more than you know.
— Andrew Fayal President & Co-Founder, Project Milk Carton A 501(c)(3) nonprofit · EIN 33-1323547 · Donations are tax-deductible to the extent allowed by law.