Nelly's Eyes: A Trust I Can Never Fully Verify

Nelly came to us in August 2022, when she was a month old, and she has grown up alongside me through my teenage years. Everyone in my family loves her. Sometimes when I look into her eyes and she looks back, something happens that I don’t have a word for, at least not one I’d use about another person. There’s no small talk to get through and no social arithmetic going on underneath. There’s just a long look that neither of us seems to want to break first. It’s tempting to call it the purest trust I know, purer than anything I have with a human being, because none of it depends on language or status or history. I want to resist that long enough to ask a harder question. What exactly am I trusting is on the other end of that look, and how would I ever find out if I were wrong?

Start with what can be measured. When people and dogs hold each other’s gaze, oxytocin, the hormone that surges between a mother and a newborn, rises in both of them. A 2015 study in Science by Miho Nagasawa and colleagues found that the effect seems to run in a loop: long gazes raised oxytocin in owners and dogs alike, and when female dogs were given a nasal spray of oxytocin, they gazed at their owners longer. Some researchers think the paper claims more than its data show. Attachment research, first built to describe how a toddler uses a parent as a secure base for exploring, fits dogs closely too. A dog with a consistent person behaves in a strange place the way a securely attached child does. Carl Rogers, writing about therapy, had a name for acceptance that doesn’t depend on what you’ve done: unconditional positive regard. For Rogers it was a discipline a therapist has to practice. Nelly doesn’t practice it. She has never measured me against a standard I could fail. Whether that is the thing Rogers meant, or something else wearing its name, a mind that never had the equipment to judge rather than one that chooses not to, is a question I’ll come back to.

It shows up in small, repeatable ways. In an unfamiliar place, Nelly ranges further from me, or from my mom, than from anyone else, glancing back over her shoulder and carrying on once she’s checked that one of us is still there. Take the two of us away and the same place stops being something to explore. She stays close and watches the door. An attachment researcher would call that a secure base, and not as a figure of speech.

Philosophy has been arguing about the question underneath this for a century. In I and Thou (1923), Martin Buber divided relationships into two kinds. In I-It, the other is an object, something to sort and use. In I-Thou, the other is met as a whole being and spoken to rather than examined. Most of life has to run on I-It; nobody can meet the barista as a Thou and still get coffee. For Buber the rare I-Thou moment is a different kind of thing altogether, a meeting rather than a gentler kind of use.

It would be convenient to say Nelly and I have an I-Thou relationship and stop there. Buber wouldn’t let me. In the same book he describes a cat’s glance that seems to reach toward him with something like a question, and then he hesitates to call it address in the full human sense. Animals, for him, sit on a threshold: real enough to unsettle the line around I-It, not settled enough to stand beside human conversation. His hesitation isn’t squeamish. On his account a Thou has to address you back, and a lonely brain full of oxytocin is primed to read address into behavior whether or not it’s there.

The best evidence that something is happening on Nelly’s side, and not just mine, comes from pointing. In 2002, Brian Hare and colleagues reported that dogs follow a human’s pointing finger to hidden food far more reliably than chimpanzees, and more reliably than wolves raised by people from birth. Even puppies with very little human contact do it well. Other studies have found that dogs can tell a happy human face from an angry one. At least fifteen thousand years of domestication seem to have selected for reading human gestures and expressions, which means dogs evolved toward behavior that looks like understanding us, not just putting up with us.

Whether that behavior has anything inside it is a separate question, and the last sentence already leans further than the evidence can carry. This is the problem of other minds at its sharpest. Grant that Nelly’s brain does something that works like reading my intentions, works like trust, works like relief when I come home. None of it settles whether there is something it is like to be her while it happens, or whether “trust” is even the right word for a process that produces trust-shaped behavior with no one there to feel it.

Put that way, Nelly’s gaze and the fluent replies of John Searle’s Chinese Room have the same uncomfortable shape. Both produce behavior that looks, from outside, like understanding. Neither can be cross-examined. Nelly can’t tell me, in words I could check, whether she trusts me the way I trust her, or whether she is running thousands of years of selection for keeping a reliable food source calm and close. The oxytocin is real. The pointing is real. Neither one crosses from behavior that looks like an inner life to an inner life I can verify, and more neuroscience won’t cross it either, because neuroscience measures more behavior, more precisely.

That doesn’t make every case equally uncertain, though, and it would be unfair to leave the comparison with a rulebook program standing as if it did. Nelly and I share a mammalian nervous system, matching brain structures for emotion, and the same hormone doing the same job in both of us while we look at each other. A language model’s fluency runs on an architecture with no known counterpart to any of that, built to produce convincing sentences. Nelly wasn’t built for anything. She was born. This doesn’t prove she has an inner life like mine. Shared biology isn’t direct access, and the argument leans on an assumption a committed functionalist would reject: that the hard problem of consciousness has something to do with what a system is made of and not only what it does. I’m not neutral on that. Biology is doing real work in my reasoning. Still, the difference is real, and it’s a reason to extend more trust to the dog than to the program, even though neither trust is the kind I can ever close the books on.

Here is the part that unsettles me. None of this makes what happens with Nelly weaker than what happens between people. It exposes something that’s usually hidden in human relationships: I-Thou was never something you could verify from outside there either. When a friend tells me they trust me, I don’t have access to their inner life any more than to Nelly’s. I have their words and their behavior, and a choice about whether to treat those as a mind meeting mine or as a convincing performance. I make that choice every day, fast, without noticing. The difference with Nelly is that the choice is visible, because she can’t supply the words that usually let me forget I’m choosing. Trusting what looks back from her eyes isn’t a more naive version of what I do with people. It’s the same act, without the reassurance language gives, reassurance philosophy never said I was entitled to.

I want it to be true. Wanting it isn’t evidence of anything except how much I want it, and I know that, and it doesn’t stop me. Nelly is looking at me now from across the room, and in a moment I’m going to look back and decide again, the way I decide with everyone I love without ever being able to check, to call it trust.

Editor’s note: the same gap between behavior and a verified inner life is something you can test directly, on a system built to have none at all. Talk to the Room produces fluent, seemingly attentive replies from a program that understands nothing; the honest exercise is noticing how long it takes you to start feeling otherwise.