Applied AI research
Intelligence as experience that evolves with humans toward their missions.
Voker is an applied research company exploring AI that learns with people, inside the experience of pursuing their missions.
Conceptual sketch. Human and intelligence set out from a shared starting point as two paths that repeatedly cross, converging on each mission and spreading wider at each frontier. With every cycle the frontier lies further out and spans more, and the mission ahead is restated beyond it.
Motivation
The conflict
Humans remain outside the intelligence being improved.
Artificial intelligence has come a long way by learning from humans. As it approaches its next frontier, the questions that matter most grow sharper: what AI is worth to people, whether it stays safe, and whether it ends up competing with them.
Humans build tools from their missions, but build AI unlike any tool: extract capabilities from humans, turn them into tasks, optimize AI to perform them, and scale. A system optimized to internalize human capability, rather than to work with humans toward their missions, naturally pushes toward substitution. “Moving humans up the stack” is no escape: whatever capability we reserve for humans today can become what AI learns tomorrow.
The same paradigm can also cap intelligence itself, which advances largely along capabilities and objectives we already know how to specify, toward the limits of finite human data. Recent research increasingly explores continual experience as a source of learning beyond static human knowledge.
The conflict runs both ways. The question is whether humans and intelligence can advance together.
Directions
The frontier: advancing with humans
Intelligence that learns with humans toward their missions.
We explore a different paradigm. It redefines what the system optimizes for, what state it holds, where it acts, and how it learns. These four connected shifts are our research directions.
The objective
From: The next taskTo: The human’s evolving mission
A mission is intent clarified into a shared commitment: long-horizon, branching, and evolving; assessable, but not necessarily directly computable. The system turns that mission into evolving optimization targets through progress estimation and credit assignment, and revises them as the mission unfolds.
The human–world state
From: What was said in the conversationTo: A persistent model of human and world
A persistent, largely latent model of intent, capability, knowledge, relationships, environment, and progress. Not merely what the human said, but who they are becoming, how they work best with the system, and what the mission requires next.
The scope of intelligence
From: A model output or a fixed harnessTo: The whole interactive experience
The whole interactive experience around the mission. It persists across interactions, interfaces, and modalities; coordinates humans with the world; asks humans for collaboration and judgment; and generates and adapts the experience itself as part of intelligent action.
Recursive lifelong learning
From: Trained first, then deployedTo: Learning with humans, continuously
Intelligence evolves throughout its lifetime: online and offline, within and across missions, and through multiple feedback signals from humans, the world, and mission progress. It learns with humans, and recursively learns how to learn and work with them better.
Task-oriented AI
Humans are the source and the recipient. The intelligence improves without them.
What we explore
Mission, human, world, and intelligence are parts of one system. Roles are not fixed, and experience keeps changing the system.
Human, world, and intelligence remain inside one evolving system. The mission gives that evolution direction; human and intelligence contribute wherever each can best advance it; and experience improves both their capabilities and how they work together as the mission unfolds.
Where we begin
How can an AI system improve its own working process through human-involved experience while pursuing measurable missions such as evaluation quality, cost, and speed?
The experience is no longer where intelligence is merely deployed to humans. It is where intelligence grows with them.
Scaling
The next scaling dimension
Humans keep creating the next frontier.
Under this paradigm, our hypothesis is that humans need not remain a fixed bottleneck in scaling. At any moment, they contribute what the system cannot yet compute, and these roles move fluidly within the experience.
- Answer level
- Insight the system could not reach alone.
- Task level
- Verification and intervention while work is under way.
- Mission level
- Direction: what is worth pursuing, and why.
Learning progressively makes those contributions computable and scalable, recursively making the intelligence better at learning from experience.
The recursion runs both ways. Stronger AI expands what humans can pursue; those new missions, judgments, and insights create the next frontier for intelligence to learn from. As this happens, how human and AI work together evolves, so that each contributes where they can best advance the mission.
“Moving humans up the stack” therefore changes meaning. The system need not depend on any capability remaining permanently human; what moves upward is the mission frontier.
AI scales capability
Human contributions become computable and scalable.
Humans reach further
People can understand, evaluate, create, and pursue more.
The mission frontier moves
New missions become conceivable and worth pursuing.
New space to learn
Those missions expose what intelligence cannot yet do.
Human-free self-improvement scales what intelligence can already optimize. Learning with humans may expand what is worth, and possible, to optimize next.
The positions on this page are research directions and hypotheses we are working to test, not established results.