Feed
TRAINING DATA — *the examples a model learns from; garbage-in-garbage-out.* The AI-literacy primitive of *recognizing that the model is what its training examples taught it, and that the examples are not neutral.*
Loading audio…
Press play to listen along. The line being read lights up as you go.
Show full transcript
Loading transcript…
Feed was a paper-figure built like a tall stack of little labeled cards, cinched at the waist by a single stubborn paper clip, and this morning she was doing what she always did: teaching a machine to see by feeding it, one card at a time. She was not an animal and not a robot. She was a stack, and the stack was working.
Across the bench sat a small mechanical eye, blinking, empty, knowing nothing yet. Feed drew a card from the top of herself — a photo of a cat, with the tiny word cat penciled beside it — and held it up to the eye. "Study this," she said. She drew the next. "And this." Card after card, each one an input paired with a label, each label a small human decision about what the right answer should be. The eye watched. It did not understand, not really; it only traced the patterns that ran between the pictures and the words, learning to expect the one from the other.
Feed fed it a hundred cards, then two hundred, the colored stripes along her edge shrinking as she spent herself into the machine. By the end, the little eye could look at a new cat it had never seen and blink out cat. It hadn't been programmed to. It had simply become what she had fed it.
"There's no magic in me," Feed told the eye when they were done. "Whatever you know, I handed to you. Whatever I never handed you, you'll never know. I am the examples, and you are what the examples made you."
Feed had been folded in the same village paper-crafts workshop as Sort, and the workshop paired every figure to a purpose. Feed was folded to feed Sort. Before Sort could ever sort a single thing, it had to be taught — and Feed's stack was where that teaching came from.
She remembered the day she first fed Sort wrong. It hadn't been on purpose. Someone had built her stack in a hurry, and it came out lopsided: a hundred cards of cats and only ten of dogs. Feed fed them all to Sort faithfully, not knowing better, and Sort learned exactly what she gave it — cats, brilliantly, dogs, badly. When a dog-card finally came through, Sort called it a cat with a bright and total confidence, and everyone laughed at first, until they realized Sort couldn't do otherwise. It had never been given enough dogs to know one.
That was the day Feed stopped thinking of herself as neutral. She looked down at her own stack and understood, with a small cold jolt, that every card in her had been chosen by someone — chosen which to include, chosen how to label, chosen what counted as right. She had no way, herself, to tell a good card from a bad one; she only carried them and handed them on. But the humans who made her — they could have looked. They could have counted the dogs. From then on Feed carried her cards differently: not as truth, but as a bundle of human choices she was honor-bound to make visible before she fed anyone anything.
At twenty-two folding-years old Feed rolled her platform to the AIForge academy, where Bit, the founder, was watching new figures arrive. Bit didn't ask Feed to recite anything. She just slid a messy shoebox of unlabeled photos across the bench and said, "Teach the eye with these."
The other new figures would have started feeding at once. Feed didn't. She sat down and went through the box first, sorting as she read them aloud. "Half of these have no label — someone forgot. A third are the same three faces over and over. And there's not one picture here taken at night." She looked up. "If I fed this to the eye as-is, it would learn a daytime world of three people and call everything else a mistake. Before I teach anyone, I have to know what's in the box — and, just as much, what's missing from it."
Bit leaned back, satisfied, because most figures fed first and asked questions never. "You looked before you poured," Bit said. "Stay. Every machine in this place is only as honest as what it's fed — and you're the one who checks the box."
Feed's own class always began with the box, not the lecture. She upended a jumble of cards on the bench and, instead of feeding them, started interrogating them out loud.
"Where did these come from?" she asked, holding one to the light. "Who gathered them, and why these?" She glanced at the class. "Over in DataForge, Catch keeps notes on exactly that — who, what, why, when — and those notes ride along with the cards into my stack. Any lean in Catch's gathering, I inherit." A student named Kai frowned: "But how can we tell if the cards are any good?" "By asking, the way I'm asking now," Feed said. "Watch."
She turned a card over to its penciled label. "Who wrote this word — and by what rule? Were they anything like the people this machine will serve?" Lena raised a hand about the labels and Feed nodded her right along. Then she fanned the whole stack wide and pointed at a bare patch in the spread. "Now — what isn't here? A gap teaches the machine nothing, and it won't know the gap exists. What's left out matters as much as what's put in." A quiet boy asked about too much of one kind, and Feed's face went serious; she pulled a thick wedge of identical cat-cards and one lonely dog. "A hundred of these, one of those. Feed the eye this and it'll adore cats and fumble every dog — not from bad wiring, from a crooked diet." She let the wedge drop. "And no clever building fixes a crooked diet. The cards are the ground everything stands on."
She gathered them back up. "One last honesty," she said. "Don't tell yourself the machine understood these. Say it matched patterns from them. That's the truer thing — and truer is the whole job."
The lamp had gone amber and low. Kai lingered after the others had drifted off, watching Feed clip her stack tidy again. "Doesn't it worry you," he asked, "that your cards might be wrong and you'd never know?"
Feed paused, the paper clip half-closed. "It used to sit heavy," she admitted. "Knowing I'd hand on whatever I was given, good or crooked, and couldn't tell which." She finished the clip and looked at him, and something eased in her folded shoulders. "But then I learned to show my cards before I feed them — every source, every gap, every lopsided pile, out in the open where a human can see and choose better." She smiled. "That's when the weight lifted. I'm not carrying a secret anymore. I'm carrying something we can look at together."
Kai nodded slowly, and felt it too — the quiet steadiness of a thing brought into the light instead of hidden. The next stack waited on the bench, unexamined, and for once that didn't feel frightening. It felt like a box worth opening carefully.
The AiForge ensemble
Feed is part of AiForge's distributed-narrative cast. Each character embodies a different curricular primitive; together they teach the full subject.
-
Sort
Classifier — the simplest ML; putting things in categories
-
Skew
Bias — where AI systems go wrong when training examples lean
-
Edge
Model limitations — what a model can't do; modeling 'I don't know' as a good answer
-
Stake
Ethics — what's at stake in deploying AI; people choosing, not rules-from-the-sky
-
Split
Train/test split — keep some examples hidden to tell learning from memorizing
-
Cue
Features — a model decides from the clues you give it; choose good clues
-
Sure
Confidence — a model reports how sure it is; low confidence means check, not trust
-
Mirage
Hallucination — when a model confidently makes something up that sounds true but isn't; check, don't just trust
-
Rote
Overfitting — when a model memorizes the exact examples instead of learning the general idea