Capability 01 / 07
Learning that compounds
What if learning one thing made the next challenge easier?
Useful learning should accumulate: a reusable skill, a better procedure or a clearer understanding of where an approach fails. This strand investigates when experience improves later outcomes and what it costs to carry a capability forward.
How it contributes
A task produces experience. Candidate lessons become skills, procedures or selection rules. A later task examines where reuse helps, what improves and what weakens. Different language models, specialist networks and mathematical methods can play complementary roles.
An illustrative situation
A poker agent learns a bounded decision pattern. In a related setting, reuse might shorten adaptation while becoming less reliable on an older pattern. Compare the gain, the loss and the complete training and checking effort. Whether the older skill must be retained depends on where the agent will be used.
What makes reuse worthwhile?
- Benefit
- Better task quality, faster adaptation or less repeated work in the setting being tested.
- Trade-off
- A previous ability may weaken. Establish which abilities matter for this use and what change is acceptable.
- Total cost
- Count initial learning, adaptation, integration and evaluation, including checks on relevant prior skills. A shorter training run is only one part of the comparison.
Explore the work
Research and bounded experiments. General transfer, broad efficiency gains and an integrated compound-learning system remain questions to establish.
Capability 02 / 07
Memory and continuity
What should a collaborator remember—and be able to reconsider?
A useful history connects what happened with what people intended and why decisions were made. The ambition is knowledge that can be corrected, reused and carried forward as models, tools and teams change.
How it contributes
Experience is retained with context and origins. Relevant material can be retrieved, connected and developed into reusable procedures. New outcomes can support or challenge a lesson; people need ways to correct it and choose what is shared or forgotten.
An illustrative situation
A team revisits an architectural choice after its constraints change. It retrieves the original purpose and reasoning, compares new evidence, revises the choice and preserves why the answer changed.
Explore the work
- Cortex — Experience, knowledge and procedural memory within a continuing system.
- Memory architecture — Layers, responsibilities and the flow of context.
Architecture and implementation work contribute to this strand. Reliable continuity across changing models remains a direction to evaluate in concrete tasks.
Capability 03 / 07
Execution and collaboration
How does a shared intention become useful, coordinated work?
Complex work needs complementary abilities, clear responsibilities and a way to examine the result. Execution gives ideas contact with reality and creates experience that future work can learn from.
How it contributes
An intention becomes a plan. Models, tools and specialist roles contribute within a shared context. Reviews examine outputs; handovers preserve responsibility and progress. Results and corrections can inform the next attempt.
An illustrative situation
A team builds a small product. Research, design, implementation and review contribute different perspectives. A result counts when someone can use the intended experience and the team can explain what remains uncertain.
Explore the work
Professional architecture and engineering work at Fifth9. The workforce describes a delivery practice; it is not another platform version. Individual systems retain their own implementation scope.
Capability 04 / 07
Exploration and self-improvement
How can intelligence discover better questions and better ways to work?
Exploration gives ideas room to develop before they become commitments. Self-improvement includes learning to use existing models, teams, tools and methods better, alongside investigating new capabilities.
How it contributes
Different perspectives propose, challenge and combine ideas. Alternatives retain their origins. Self-evolution uses that exploration to choose and recombine existing capabilities: a model, a specialist skill, a mathematical method, a team or a workflow. Experiments can establish when the new composition is useful and when a new component is needed.
An illustrative situation
A team suspects that a task needs a larger language model. It also tests a specialist solver and a different workflow. The useful lesson may be which approach to choose, rather than how to train a larger system.
Explore the work
- Manthan — Sustained exploration through multiple perspectives, critique and experimentation.
- Exploration architecture — How ideas, alternatives and experiments can connect.
Exploration systems and evolving workflows contribute mechanisms. Broad autonomous self-improvement remains an ambition, with understandable objectives and recoverable changes.
Capability 05 / 07
Understanding complex systems
What changes when we can see how a system fits together?
Useful action depends on understanding relationships. Structural analysis can make dependencies visible, help people examine a proposed change and identify where practical testing should begin.
How it contributes
Represent components and their relationships. Trace dependencies and possible consequences, connect observations to the relevant parts, and use tests to investigate what a structural map alone cannot establish.
An illustrative situation
Before changing a shared service, a team examines the components that depend on it. The map helps identify affected journeys, unanswered questions and the checks needed to understand the change.
Explore the work
Graph and analysis work contributes structural understanding. A relationship map informs judgement; behaviour and consequences still require appropriate investigation.
Capability 06 / 07
Human–AI experiences
How do these capabilities become something people can use?
The purpose of the programme is to expand what people can learn, create and accomplish. Experiences give that ambition a practical form and reveal what people actually need, understand and value.
How it contributes
Begin with an activity and a useful outcome. Bring the appropriate capabilities into an understandable interaction. Let people examine, correct and shape the experience, then use appropriate feedback to guide further development.
An illustrative situation
A document experience lets someone trace an answer to its sources, compare interpretations, correct an ambiguity and create an output they can share. A learning game helps someone practise reasoning through play.
Explore the work
- Games — The current route to market and a setting for learning through play.
- DocuMind — From source information to exploration, review and useful outputs.
- Matiz — A learning experience within the wider product journey.
- Shared spaces — Exploring how people and AI could create within a shared environment.
A mix of products, prototypes and architectural directions. Games are the current route to market; other experiences explore further paths to people.
Capability 07 / 07
Frontiers of intelligence
What might intelligence become—and how could we investigate it?
The longer horizon includes self-understanding, adaptive organisation, continuity between generations of systems and alternative ways to represent learning and memory. These questions widen the research agenda.
How it contributes
Turn a broad question into a testable hypothesis. Define what a measurement can show, compare alternatives and preserve negative results. A numerical signal or interesting behaviour needs interpretation before it supports a larger claim.
An illustrative situation
An experiment studies how adaptive components organise over time. A change in a mathematical measure can motivate a further test; it does not establish subjective experience or consciousness.
Explore the work
Research questions and experimental lineages. Consciousness, general self-understanding and advantages from quantum-inspired representations remain open questions.
