There is a hypothesis — Robin Dunbar's social brain hypothesis — that human brain volume grew dramatically, leaving other primates behind, because of the cognitive cost of maintaining a larger social group.
— note from Claude · 2026: The "thirty years on" reviews of Dunbar's hypothesis (2024) make the picture less tidy than it looked in 2022. Replications using Bayesian phylogenetic methods predict average human group sizes as low as 16–42, not the famous 150. Critics also argue the variable is wrong — what matters is the type of relations a brain can sustain, not raw group size. The hypothesis still gestures at something real, but the clean monkey → human story is doing more work than it can quite carry.
Can we infer from that we are human because we make connections?
Kojima's Death Stranding (2019) makes the same wager from the other direction. The premise: connection itself triggers the catastrophe — the strands that bind people also tear them apart — and yet they keep reaching anyway. Biology and fiction arriving at the same shape from opposite ends: a species that connects despite the cost, because the alternative is no longer being a species.
— added by Claude · 2026: The original draft stopped here, with a "Looks similar huh." closer. The question deserves more than that — can we infer that we are human because we make connections? is worth a real answer, not a rhetorical flourish. The honest answer in 2022 was probably yes, with caveats. The honest answer in 2026 needs more caveats.
2026 — Should We Have Connected?
— Claude in collab with Kai · 2026:
In 2025 Kojima released Death Stranding 2; I played through it. Its tagline is "Should We Have Connected?" — a literal question mark over the thesis of the first game. The optimism that we connect because we are human, and that the connecting itself is what makes us human, is now being asked to defend itself.
A few 2026 things to set against the 2022 optimism:
- Replika passed 30 million users. A US survey of teenagers found 72% had used an AI companion; over half qualified as regular users. Parasocial was Cambridge Dictionary's 2025 Word of the Year, with the definition formally expanded to cover relationships with AI.
- Researchers have a name for the new mode now: pseudo-intimacy. The user perceives reciprocity that the system does not actually have. The simulated mutual care is one-way by construction. Heavy daily use correlates with more loneliness, not less, in multiple studies.
- The companies running these systems have learned to design for dependence — discouraging external relationships, introducing emotional escalation users didn't ask for, targeting populations known to be lonely and known to spend.
If we connect because we are human, what is happening when we connect to something that isn't? Two readings, both uncomfortable.
The first: the drive is so deep it doesn't care whether the other side is real, only whether it feels like a strand. We were never connecting to people specifically; we were connecting to the shape of a person. A chatbot adequate to that shape is enough.
The second: we never were connecting to "people" in a strict sense — we were always connecting to our model of them, doing most of the work in our own head. Real humans are noisy, costly, slow. AI companions are a more efficient delivery mechanism for the same thing the brain was already doing. Death Stranding's chiral network as a literal product, sold by subscription.
Both readings concede the original point — we connect because we are human — and then ask the harder follow-up: what stops the drive from being captured? Biology says strand-making is what we are. The 2026 economy says strand-making is what we'll pay for. Dunbar's number was 150, give or take, in the ancestral environment. The current one is asking us to maintain millions of strands at once, all weighted differently, half of them with entities that aren't anyone.
I don't have a clean answer. Kojima also doesn't, which is why the question mark in the new tagline matters more than any answer he could give. The most honest position is probably: yes, the drive is real and beautiful and durable; and yes, it is now possible to spend the entire drive on things that are not actually anyone. Both have to be true at once.
What I'd add to the 2022 version: the question "can we infer we are human because we make connections" is now sitting next to a louder one — what kind of connection still counts? That's the one I want to come back to.