Skip to content
Advancing Bits to Master Atoms

Introducing Neuro-Analytic AI.

Neuro-Analytic AI pairs the AI you already use with a complementary analytical AI that learns from validated outcomes and helps you decide what to try next.

Its core promise: the flexibility of neural AI with the exactness of analytical AI. MatterSpace is pioneering it and building it into products, starting with MS Duplex. Developed by Faruk Guney.

A smooth, rounded satin-silver form and an angular, faceted graphite form fitting together on a stepped graphite plinth, with a thin cobalt-blue inlay along the seam between them.A smooth, rounded satin-silver form and an angular, faceted graphite form fitting together on a stepped graphite plinth, with a thin cobalt-blue inlay along the seam between them.

Two kinds of AI, each doing what it does best.

Neural AI and analytical AI are good at different things. Put side by side, each covers ground the other covers less well.

  • Neural AI

    The models your team already uses. They understand a task described in plain language, research it, write and reason fluently, and work through the tools you authorize.

  • Analytical AI

    Works from the outcomes your team has validated. It weighs what worked, under which conditions and on how much evidence, and it keeps that judgment current as new results arrive.

  • Together

    The neural side understands the task and acts on it. The analytical side brings experience and evidence to the decision. The result is an agent that improves with every validated lesson.

The trilemma neural AI cannot solve on its own.

AI that works beside people every day needs three properties at once: plasticity, to learn from new results; stability, to keep what it already knows; and editability, to change one piece of knowledge exactly. A neural model keeps everything it knows spread across the same shared weights, so gaining one property costs another.

  • Plasticity

    Teach the model something new, and the update reaches weights that also hold what it already knew.

  • Stability

    Protect what it knows, and it learns slowly or not at all until the next training run.

  • Editability

    No single weight holds a single fact, so one piece of knowledge cannot be corrected, removed or traced to its source exactly.

  • Workarounds trade one for another

    Fine-tuning buys plasticity at the cost of stability. Retrieval and memory files add text without learning from it.

Analytical AI holds all three at once.

Analytical AI keeps what it learns as distinct lessons beside the model, not inside its weights, each with its evidence and the conditions it holds under. The three properties stop competing.

  • Plasticity

    One validated result is learned at once and applies from the next request, with no training run.

  • Stability

    A new lesson leaves every other lesson as it was.

  • Editability

    Any lesson can be inspected, corrected or removed on its own, and traced to its source.

Exactness and flexibility: the core promise of Neuro-Analytic AI.

Neural AI brings flexibility: it understands open-ended tasks, reasons and works through tools. Analytical AI brings exactness: what was learned applies exactly where it holds, and a changed fact is replaced exactly. Together, neural AI gains plasticity, stability and editability that are both strong and exact.

From impressive demos to everyday work.

This is what operationalizes AI. An AI that learns on the job, keeps what it learned and can be corrected like any other record is one that scientists, engineers, researchers and business professionals can rely on in their everyday work, not only in a demo.

Why we believe the future is neuro-analytic.

Larger models know more, yet scale alone does not let a model learn on the job without forgetting, or let an organization correct and own what it learned. We believe the AI people rely on every day will pair neural flexibility with analytical exactness.

  • Work never stops changing

    New results arrive every day. An AI that learns only in training runs is always behind.

  • Trust needs correction

    When AI gets something wrong, people must be able to fix that one thing and see that it is fixed.

  • Experience is an asset

    What a team learns should belong to the team, not disappear into a model’s weights.

  • Models keep changing

    Experience has to outlast every model it was learned with.

Inspired by complementary learning systems theory in cognitive science.

Complementary learning systems theory describes learning in the brain as the work of two systems: one that learns quickly from specific experiences, and one that builds general knowledge gradually. Neuro-Analytic AI takes inspiration from that idea for how two complementary parts of an AI system can work together.

MatterSpace leads Neuro-Analytic AI from research to product.

MatterSpace is the leading company pioneering Neuro-Analytic AI and turning it into products. MS Duplex is the first: connect the AI you already use over MCP, and MS Duplex is the analytical AI beside it, a self-improving agent for science, research and engineering.

Developed by Faruk Guney.

See Neuro-Analytic AI on your own work.

A pilot of MS Duplex runs on your own recurring work, beside your current setup, and your experts judge the results.