We’ve known the phrase for decades, but it took a chatbot to make us finally question it.
“Garbage in, garbage out” used to be a warning for programmers. Now it’s the phrase we reach for when a lazy prompt returns nonsense - proof, we tell ourselves, that the machine only handed back the mess we typed in.
But spend enough time with these tools and you notice something stranger: most of the time, the garbage works. You type a half-formed thought, a misspelled word, a sentence you’d be embarrassed to send a colleague - and the machine understands you anyway.
Here’s the uncomfortable question: if a machine can fill in our blanks this easily, why are we so quick to blame the people who can’t?
The Mind-Reading Tax We Put on Humans
When we talk to another person, we carry an invisible expectation: they should understand us, even when we’re not being clear.
We send half-formed emails and expect colleagues to decode them. We give vague briefs and blame the designer when the output misses the mark. We say “you know what I mean” as if telepathy is a reasonable job requirement.
Psychologists call this the curse of knowledge - a cognitive bias where knowing something makes it nearly impossible to imagine not knowing it. In a famous 1990 Stanford experiment, psychologist Elizabeth Newton had people tap out a song’s rhythm; the tappers predicted listeners would guess the song correctly 50% of the time. The actual success rate? 2.5%. Once you hear the melody in your head, you can’t fathom that others only hear random taps.
This is how we communicate at work every day. We hear our own melody. We assume everyone else does too.
And when things go wrong? It’s rarely our fault. The other person should have asked clarifying questions. They should have known. They should have read between the lines.
The Machine Reads You Anyway
Then you open ChatGPT, type the same half-formed thought you’d never dare send a colleague, and it just gets it.
It fills in the gaps. It forgives the typos. It reconstructs the sentence you meant from the one you actually wrote. You can even hand it something you don’t fully understand yourself - a vague itch of an idea - and it hands back something coherent.
This is the part we rarely sit with. We expected the machine to be literal and unforgiving: garbage in, garbage out. Instead it turns out to be the most charitable reader most of us will ever meet.
But notice why. The machine doesn’t fill your blanks out of kindness. It fills them because it knows us.
It knows us through training. Your clumsy prompt looks like a million clearer ones it has already absorbed. When it “understands” your fragment, it isn’t reading your mind - it’s pattern-matching against the collective writing of humanity, recovering the sentence you were probably reaching for.
And it knows us through context. Give it your earlier messages, your documents, your history, and it knows you specifically - your shorthand, your habits, what you tend to mean when you trail off.
A colleague has neither. They haven’t read the entire internet, and they certainly haven’t memorized every message you’ve ever sent. So before we resent the people who fail to read between our lines, it’s worth naming what we’re actually asking: we want a single, busy human to do from memory and goodwill alone what the machine only manages because it was handed extraordinary context. That was never a fair fight.
The Generosity We Save for Software
Watch what we instinctively do to make the machine understand us better. We add context. We explain what we actually want. We spell out what success looks like.
In other words, we hand the machine the exact thing we withhold from people: context, offered generously, before anyone has to ask for it.
Sound familiar? That’s also the recipe for a good brief, a good handover, a good request to a teammate. The skills are identical. The difference is that we perform them for software and skip them for humans - then act surprised when the humans come up short.
The truth is our communication was never this generous. We rarely supplied context, because we expected people to supply it for us. The machine quietly reset that expectation, and in doing so it showed us how little of this work we were ever willing to do for each other.
Will We Carry It Beyond the Chat Window?
Here’s where it gets interesting. Millions of people are now practicing this every day - adding context, clarifying intent, giving a listener what it needs to succeed - with machines.
And there’s early evidence it’s changing how they feel about communication itself. In Grammarly’s 2024 State of Business Communication study, run with The Harris Poll, workers who use generative AI report less stress, more productivity, and more satisfaction in how they work and communicate. That’s a remarkable signal for a tool most people have only been using for two years.
The open question is whether the habit transfers. Will we start giving the people in our lives the same context, and the same patience, we hand the machine without a second thought? Or will we reserve our generosity for software and keep expecting humans to read our minds?
The Empathy Paradox
There’s a strange irony here. We’re more forgiving of artificial intelligence than of actual intelligence.
We’ll rewrite a prompt five times to get the output we want, cheerfully, no resentment. But when a human misunderstands us once, we get frustrated. We assume bad faith. We write them off.
Think about that. We give a machine more benefit of the doubt than we give the people we work beside every day - even though the machine only earns that benefit because it was fed everything, and the person was fed almost nothing.
The machine isn’t kinder than us. It’s just better informed, and we’ve quietly let being well-informed stand in for being generous. That’s not a technology problem. That’s a question of who we decide deserves our patience.
What if we flipped it? What if we treated the people around us a little more like we treat the machine - assuming that an unclear message means we should add context, not assign blame?
The Real Lesson
“Garbage in, garbage out” was never really about computers. And it turns out it was never quite true, either.
The machine takes our garbage and finds the meaning buried in it - because it knows us, at a scale and an intimacy no single person ever could. The people around us are working with a fraction of that, and we hold them to the higher standard anyway.
That’s the curse of knowledge doing its quiet damage, and awareness alone won’t save us from it: studies show that even warning people about the bias doesn’t reduce it. But practice might. Every time you add context for the machine instead of assuming it can guess, you’re rehearsing a generosity you could just as easily extend to a person.
Maybe the real gift of the AI era isn’t how well the machine understands us. It’s the mirror it holds up - showing how rarely we do that work for each other, and how easily we could.
The machine fills our blanks because it was given everything it needed to. The people around us are working with far less - and the question is what we choose to give them.