Step inside the haunted house...👻
You’re told data is neutral.
You’re told algorithms are efficient.
You’re told progress is inevitable.
Yet every dataset you collect already carries the weight of history.
Every model you train borrows its language from the living and the dead.
Every time you speak to a person and record what they say, you’re deciding who is allowed to appear and who is written out.
What happens when you feed all that into machines?
What happens when you let those machines start speaking for you?
If you work in research, you already know this. You know data doesn’t arrive clean.
You know the categories you use are conveniences.
You know silence and refusal can tell you as much as any answer does.
And yet you still feed it all into a machine, trusting it to speak back.
When the machine answers, can you honestly say you recognise the voice?
Questions That Demand Contemplation 🪞
1. Whose ghosts fill the data? 🤔
Every dataset is a haunted castle (who remembers the classic board game ‘Ghost Castle’?).
It’s full of the voices of people who were easy to reach, who played along, who fit into the categories you gave them. The rest were left outside.
So when you train a model, whose ghosts are trapped inside it?
And who remains unrecorded, unrecognised, uninvited?
2. Are you carrying baggage or truth? 🎒
When you hit your quotas and produce tidy segments, you call them reality. But they’re only stories & made-up names for patterns you think you see.
Are you really recording what’s there, or just what will fit into your deck?
How much of what you pass to the machine is already packed with yesterday’s mistakes?
3. Do you notice the ghosts when they speak? 🗣️
Researchers of algorithmic bias have already shown how models amplify the inequities they inherit.
When AI predicts what “people want,” is it predicting anything at all? Or just repeating what you’ve told it before?
How would you know the difference?
4. Can ghosts refuse to speak? 🙊
Silence haunts research too.
Refusal, contradiction and discomfort are data, but most of the time you code them as error.
What does a machine do with silence?
Does it flatten it? Replace it? Pretend it was never there?
What are you losing when you call that noise?
5. How much weight can you ask people to carry? 🏋🏻
Every insight begins with a person, someone who gave you their time, their words, their feelings.
What happens when those words are stripped of context and turned into something that speaks as if it were them?
At what point does participation start to feel like haunting, their voice wandering, disembodied, through someone else’s machine?
Final Musings 🤘
You tell yourself AI is a fresh start. But it isn’t.
It is an archive. A memory of how you have chosen to see the world.
The ghosts are already here.
The only question is whether you’re willing to see them.
