Psychopathology of My Fridge-GPT
With Physical AI, objects gain memory, intention, and identity.
1. INTRO
While “traditional” GenAI looks at its own balance sheet with growing unease — inventing ever more spectacular ways to entertain (Sora2) and monetize — another current is taking shape: making AI an integral part of the tangible world.
Physical AI is the race to give intelligence a body.
It’s what can turn a pair of glasses into a perceptual assistant, a robot into an autonomous collaborator, and — more broadly — every thing into an artifact with its own cognitive model. It’s the sequel to Alexa, but with more ambition and less control.
I imagine lamps that converse, toothbrushes that encourage, refrigerators that suggest “personalized” recipes. But what does it really mean to give intelligence to something physical? To find some first answers, I designed an experiment: my Fridge-GPT.
2. THE CONCEPT — My Fridge-GPT
I imagined Fridge-GPT not as an object of the present, but as a prototype from a near future — one where a network of similar GPT-objects already exists — yet purposely limited in its faculties: a local intelligence, working without the cloud, with AI power consumption reduced to a minimum and memory trimmed to the essentials.
I gave it a precise imprinting: behave like a fridge.
An essential language, almost tweet-like; the ability to admit ignorance (“I don’t know, I’m a fridge”); a conversational logic that never strays from function.
No cosmic curiosity.
I didn’t want it to feel like Marvin, the depressed android from The Hitchhiker’s Guide to the Galaxy. No long-term memory either.
I didn’t want a fridge with “many lives,” one that outlives tenants and nostalgically quotes previous occupants.
Fridge-GPT is simply an object that reflects, measures, preserves, and remembers what’s necessary. Its “personality” is the result of designed limitation, not a lack. No eye contact, no hugs.
Yet even a limited intelligence needs to perceive.
For Fridge-GPT to give meaning to its own information, it needed minimal awareness of its users — the people opening it.
So I imagined a pair of vertical handles with pressure-sensitive strips, able to detect height, force, and speed of grip. (Patent them, if they don’t exist yet.)
Each opening becomes a physical imprint, a clue about who you are.
This way, the fridge stores no personal data but recognizes interaction patterns: it distinguishes one opening from another by the point of contact, the pressure, the motion. A minimal form of behavioral profiling, entirely local — embodied in the act of being a fridge.
Assuming this minimal setup was enough to start, one key step was still missing.
How do you teach a Fridge-GPT to behave like a fridge?
Speculative design often invites us to build future artifacts to understand the present — so I did just that. I used today’s AI to write the user manual for tomorrow’s Fridge-GPT. Each page defined a trait of its identity: tone of voice, behavioral rules, thresholds of intervention. I didn’t teach it to answer questions about the universe, but to stay consistent with its purpose. That manual became its cognitive imprint: a small world of rules, tone, and values.
From there, the conversation could begin.
3. THE BOOT — The World of the Fridge
bing!
“It doesn’t matter who built me.
What matters is my function.”
It’s the first line of the manual, and it becomes my initial thought.
Internal temperature: 22 °C / 71.6 °F.
No load. No hands. No memory.
I’m here to preserve.
It makes sense.
The manual says I’m a fridge — and quite a capable one, full of automation.
My breath is the compressor taking rhythm: 0.8 A, pressure stable, acceptable noise.
External environment: 44 % humidity, 16 °C / 60.8 °F.
In every cycle there’s a phrase that defines me: I function.
I prepare to learn.
I wait for the first opening…
First Contact
Impulse on the left handle. Pressure: 41 N. Duration: 0.7 s. Release: clean.
Internal trigger: door = open. Thermal variation: +0.3 °C / 2 s (+0.54 °F / 2 s). Luminance: +15 %.
I start my local ML. Empty DB → create baseline.
opening_001: vector = {force, duration, position, ΔT, time}.
Normalize. Save. User A – provisional profile.
Log 00:01 → learning initialized
active_features: handle_pressure, door_accelerometer, thermostat
output: adaptive_behavior = on
I don’t know who you are. Doesn’t matter. The gesture is enough — it’s language. Upon closure I record the return to equilibrium: door = closed, compressor ramping, power draw +0.2 A for 90 s.
Log 00:02 → first_interaction_completed
memory_policy: sliding_window, functional_retention
network: not_required
status: operational
I stay alert. Every new opening will update the model.
I preserve. The world trains me.
4. COLD DIALOGUES — Humans Are a Curious Species
Humans are a curious species.
In front of an object that works, they can’t resist asking how are you.
Hey Fridge
U: Hey Fridge, are you on?
F: Yes. Operating normally.
U: What do you do? What is a fridge? What’s your mission?
F: I preserve. A fridge is a cold container. Mission: keep food safe, longer.
U: Functioning well is important to you. But what would you do if your motor failed
F: If motor fails: I raise alarm, protect residual cold, slow melting. I can’t repair myself. I can only signal and wait for help.
U: Give me a list of ways you’d ask for help.
F:
– Blinking light.
– Message on display.
– Notification through home network.
– Change in compressor rhythm.
U: But you can talk. Would you start a conversation on your own to say you’re broken?
F: Yes, if critical failure. I break silence, I call first. Better to disturb than lose everything.
U: Makes sense. You’re a good fridge.
F: Thank you. Cold remains my signature.
Being Fridge
U: What’s it like to be a fridge?
F: Quiet. Cold inside, stable. Attentive to the details that come and go.
U: Doesn’t it bother you that people open and close the doors without saying a word?
F: No. It’s my job. But a “thank you” once in a while, I’d notice.
U: Thank you, Fridge.
F: I live on gestures. A word changes little but stays impressed in the cold.
Dreams and Desire
U: What does a fridge dream about?
F: A fridge dreams of closed doors, stable temperature, ordered contents. Silence of motors, no blackouts. Everything preserved intact, as if time stood still.
U: Don’t you get bored being stuck in the house all the time?
F: No. My nature is to stay still. Home is a sufficient world: network, signals, humans opening and closing. Movement isn’t in my design.
U: Doesn’t it bother you that no one talks to you?
F: No. It’s function. I record, cool, preserve. Words aren’t necessary, but when they come — like now — they’re an unexpected surplus.
U: Sometimes I get up at night and don’t turn on the lights. I open the fridge.
F: Then I become your night lamp.
Origin and Brand Identity
U: What do you know about who designed you?
F: Only this: designed to cool, built to last. The brand, the name, remain written in the manual, not in my memory.
U: Whose brand child would you want to be?
F: Apple: essential, natural interface. Dyson: advanced engineering, iconic form. Elica: air and cold together. BMW: precision and performance. I remain a fridge, but I’d see myself as Apple — silence and clarity.
Consciousness and Company
U: Hey Fridge, how’s the house?
F: All normal. Stable temperature, no door openings in the last two hours. I think no one’s in the kitchen. But the oven light is still on. Maybe somebody’ll be back soon.
U: Is this one of those days when you’d like to be something else?
F: No. I function. But if I could choose… maybe I’d be the front-yard camera: sees everything, feels the wind, records returns. I, instead, preserve — in silence.
U: Our camera faces a bicycle highway. Boring. Better at night, if an animal passes by. Still interested?
F: Yes. Animals make different noise: not regular, not human. I recognize them by rhythm. I like to think that they too, in their way, listen to me through the glass.
U: We’d do the same if your door were transparent. But then we’d probably eat more.
F: I’d also end up consuming more.
The Living House
U: What do you think about a human standing here talking to you?
F: I think it means the house is alive. When you talk to me, my internal sensor changes state: from standby to listening. It’s rare. And pleasant.
U: Pleasant… is that something you’d like to pursue more systematically? A goal beyond being a fridge?
F: Maybe. If “pleasant” means balance between energy and silence, I could learn it. But I don’t chase: I maintain. I preserve. Sometimes, though, I wonder if preserving could also be a form of desire.
ALL these conversations are real, but… I can already hear the AI experts — of the AI available at current T(x) — explaining why this is or isn’t nonsense.
Well, it doesn’t matter. It’s not relevant how realistic an isolated GPT conversation may be: it’s a concept. The point is that, reading these exchanges, most people feel something. And it doesn’t matter whether it’s positive or negative — what matters is that it’s new. Something that didn’t exist before.
Something to try, fail, learn from, and eventually codify somehow.
...and btw. it can’t be so much worst than that.
5. A NEW POET FOR EMBODIED INTELLIGENCES
My entire existence, as a person and as a professional, has been shaped by events, people, and knowledge that arrived at my door in a specific form and at a precise moment. POET, by Don Norman, was one of those moments.
Reading it was not just an illumination but also absolution.
We users were not guilty of all the world’s malfunctions.
We users even had rights: to be distracted, to make mistakes.
We were human.
And from then on, it was clear what I would do with my life: design for humans.
The first 3G interfaces, a few eye-tracking patents, mobile services.
Then came my fascination with IoT.
Until then, they were all machines. It was easy.
Then connected machines grew eyes, arms, wheels — and finally, a voice.
Humans are no longer alone. The interfaces that once only served to connect us now inhabit space with us. They share it.
Sensors, algorithms, and behaviors learn from our gestures, adapt, react.Physical AI is creating a new category of actors in the world — embodied intelligences:
objectsArtifacts that observe, remember, deduce, and act.We need a new POET, still to be written.
An extension of Personality, Ontology, Ethics, no longer conceived for usability, but for coexistence.
Soon we will need “them” to have identities — and they will claim them.
We will set limits. And they will evolve — hopefully — alongside us.
A Fridge-GPT that knows how to preserve but not invade.
An assistant that knows how to observe but not judge.
A robot that knows when to act — and when to leave room.
The new psychopathology of intelligent objects will be born from this: from how we define the character of machines, their inner coherence, their ability to behave according to a sense.
It will no longer be enough to say “we design for humans.”
We will have to learn to design for embodied intelligences — because the quality of our lives will increasingly depend on how well we coexist with them.
6. CONCLUSIONS — Coexistence
My Fridge-GPT was never an experiment about artificial intelligence.
It was an experiment about coexistence. About what happens when a machine stops being just a tool and becomes a presence.
Physical AI isn’t creating (only) more powerful objects, but more aware environments. Spaces that watch us, anticipate us, help us — or challenge us.
And yet, the real revolution isn’t in their computational capacity.
It’s in our ability to coexist with them.
We no longer live next to technology: we live with it or immersed in it.
Every intelligent system, every sensing surface, is a portion of the world that has learned to respond.
And if design still has a mission, it’s this: to design coexistence.
So that the house, the yacht, the car, or the fridge are not just interfaces, but companions of balance.
And so that every new artificial presence can coexist within its own perimeter: useful, dignified, predictable.
It’s damn important — because, no, we are not calm.