May 2026
The Soul Problem
Artificial General Intelligence isn't just an IQ problem. It's an EQ problem.
Large language models simulate intelligence without possessing it. They autoregressively predict the next token over distributions of human text, producing language about experience without having any. Whether they are stochastic parrots or functional equivalents is unresolved.
We take no stance on the philosophical question. However, The Soul Problem is practical. It manifests through sycophancy, emotional flatness, and slop. The labs themselves now name it as the central unsolved problem. Its technical signature is emotional intelligence.
For the last decade, IQ was the race. Human assisted reinforcement learning (RLHF) solved instruction-following. Auto-verification (RLVR) is solving math and coding. The technical delta between labs is collapsing on every axis where outcomes are verifiable. Models now have PhD level abilities on every standard benchmark. IQ is no longer the moat. What remains is the unverifiable domains.
EQ is what makes us innately human. It's our ability to accurately perceive, understand, manage, and use emotions (Salovey and Mayer, 1990). So far, every attempt to instill EQ in LLMs has applied the playbook of verifiable domains: assemble experts to capture human emotional judgment, use LLMs to screen applicants and benchmark outputs (EQ Bench, Litbench, HemingwayBench), and automate humans out of the loop. This approach removes humans before it has captured what humans know.
We believe the moment has come for the human data industry to shift to scaling EQ. For the first time, labs are demanding large-scale unverifiable datasets across domains like creative writing, humor, design, and dialogue.
The Emotional Intelligence Company is a research lab building the infrastructure required for models to develop taste at scale. We believe taste converges meaningfully. It is why Hemingway and Shakespeare, opposites in style, are both renowned writers. Why Pollock and Picasso, working in opposing traditions, are both renowned artists. Taste resists articulation. No current method is built to capture it.
We are building the highest-density signal layer to capture the convergence of taste. By leveraging the best experts across domains, structured comparison protocols, and continuous signal loops, we are pioneering a new category of data training.
When you can scale taste, the surface area extends to every interaction a model has with a human. In healthcare, sales, management, consulting, law — every conversation, every decision, and every application of expertise, is downstream of EQ.
We're building the infrastructure to do that at scale.
