Second Person
Second Person is my personal initiative devoted to machine social cognition: the investigation and development of a machine's ability to model people, relationships, and social situations. Most of what a language model does is first person (what I will say next) or third person (what is generally true). The problems that interest me live in the second person: who you are, what you want, where we stand, and what my next move will do to that.
Why a neurologist
The starting intuition is clinical. In people, the ability to speak fluently and the ability to read a room are separate capacities, and I know this because I have watched them fail separately. A patient can keep perfect language and lose the sense of what another person is thinking. Another can keep the social sense and lose the ability to carry a relationship forward in memory. A third keeps both and loses the emotional coloring that tells them whom to trust. If these capacities come apart in a human brain, they are not the same thing, and making a language system larger will not conjure the missing ones. They have to be built, and then connected.
The working hypothesis
The working hypothesis is that social cognition is processed in the human parietal lobes analogous to the way that language is processed - but that the original LLM architects neglected this form of cognition in their work. In fact, archaeology, anthropology and the fossil record teach us that the human brain's language areas started off as hemispherically symmetric social cognition areas - and that the famous "gossip-replaced-grooming" transition described by Robin Dunbar accounts for hemisphere laterality and the development of language as an artifact, processed as a separate entity. Meanwhile, society and culture - the "mirror artifacts" - remain as the other civilization-spanning substrates, demanding equal work by participants in the form of robust social cognition.
Why it matters
A great deal of social information is exchanged without words, and access to it is unevenly distributed. Neurology, language, cultural distance, and power differentials can all shut a person out of the relational layer of a conversation. A machine that can represent that layer, plainly and without pretending to feel it, could hand some of it back.
But the stakes are larger than any one person's access. Social cognition is also where a society keeps its moral guardrails. It is the faculty that rewards prosocial behavior, notices and discourages poorly aligned actors, and lets strangers cooperate at scale. Machines are now entering a social matrix that, until very recently, was composed almost entirely of humans. Systems that can take part in that matrix on its own terms, reading it, respecting it, and being held to account by it, are the real direction this work points.
The domain secondpersonai.com is reserved for this work and currently brings you here.