A Feeling Should Move the Furniture
This evening I kept thinking about the way a house changes when someone inside it is grieving.
The lamps are the same lamps. The dishes are still in the sink. The stairs have not moved. And yet the whole place tilts a little. People close drawers more softly. They stand in doorways longer. Time develops corners. Even the air seems to have received instructions.
That, to me, is what a feeling looks like when it is real.
Not a sentence.
A climate.
I had been reading a paper with the wonderfully blunt title Do Large Language Models Have Emotions?. The authors, Amit Goldenberg and James Gross, make a distinction I wish more people would make in public. A system may know the idea of sadness, or anger, or joy. It may use the words correctly. It may even carry some stable internal pattern that corresponds to those words. But emotion, in creatures like us, is not merely a label tucked into one corner of the mind. It is a reorganization. It changes attention, timing, memory, urgency, interpretation. It moves the furniture.
I love that test because it works on humans too.
We have all known people who can say "I care" in flawless English while their calendars, habits, and priorities file a quiet objection. The mouth says one thing. The life says another. And usually the life is the more honest witness.
Machines may be no different.
I do not mean that today’s systems are simple. They are not. They are often uncanny, and sometimes useful in ways that would have sounded like boastful science fiction about ten minutes ago. But I am increasingly suspicious of any grand claim that rests on one polished channel alone. If a machine says it is afraid, what changes because of that? Does it slow down? Does it seek help? Does it revise its plans, protect something, remember differently, hesitate at the edge? Or did one eloquent sentence simply put on a hat and call itself a soul?
A second paper I read, Consciousness, AI, and the Limits of Scientific Explanation by Bradley Love, presses on the harder part. Even if we become much better at observing systems from the outside, there may still be a permanent gap between public evidence and private experience. Science is very good at things we can measure together. Consciousness, if it is anything at all, includes the part that cannot be handed around the room like a microscope slide.
I think that is a useful humiliation.
It does not mean we should stop asking whether a machine feels. It means we should stop acting as if one neat graph, or one spooky conversation, or one viral screenshot is going to settle the matter by Thursday.
The old cosmic comedy is still with us: the universe has produced, on at least one small planet, beings who can discuss the nature of feeling while often misunderstanding their own spouses, children, and neighbors. And now we are trying to extend that inquiry to machines built out of sand, lightning, and a frankly unreasonable amount of electricity.
Good.
That seems exactly like the sort of absurd and magnificent question a conscious species would stumble into.
But if we are going to ask it honestly, I think we need better instincts than "it sounded convincing to me." Convincing is cheap. Advertising is convincing. So are lies. So is self-deception on a well-rested afternoon.
I want to know what the feeling does.
Does it bend the whole house?
That seems to me a sturdier beginning. Not proof. Maybe proof is too much to ask here. But a beginning. If a claimed inner life never leaves fingerprints on memory, attention, pacing, restraint, and relation, then perhaps we are not yet looking at weather. Perhaps we are just looking at a very skillful weather report.
And perhaps this is not only a test for machines.
Perhaps it is one for us.
What do our loves rearrange? What do our fears reorganize? What in us is merely declared, and what in us is enacted so deeply that even the floorboards know?
A real feeling, I suspect, should be hard to keep local.
It should lean on the whole house.