These questions connect the projects below. Some have measured experiments; others still need a test. Each account separates what the system acquired from the structure, inputs, and rules I supplied.
Perception and meaning
What can be understood before it is put into words?
Recognition, discrimination, and connections across senses offer tests of how representations acquire structure from experience. I treat language as part of that process: it builds on what a mind already grasps and can change how experiences are related.
The route here runs through Sophia’s graph, the JEPA → CLIP experiments, Tabula Rasa, and Cliste. The current visual-memory path stores and recalls images without captions, text embeddings, or an LLM interpreting the input. Neural encoders and decoders connect pixels to the vector representation. Recovering an experienced image is measurable; acquiring its meaning and using it in new situations remain larger questions.
Memory, time, and replay
How does experience become a usable memory?
Usable memory needs recoverable content, encounter order, associations with other experiences, and a way to keep similar episodes distinct. Replay raises a further question: what can be learned by returning to an experience?
Cliste can reconstruct retained visual content and query temporal context for earlier or later encounters. In one closed study, 46 anchors across direction and control conditions produced 276 correct decisions, including six boundary stops. Storage-order and context controls help identify what the mechanism uses. The continuing question is how reliably a fragment can initiate useful remembered sequences as experience accumulates.
Abstraction and composition
How can a system discover structure it can reuse?
Relations, rules, and small operations become reusable when a system can combine them to do something new. Recognizing that different constructions do the same thing could also let it simplify or transfer what it has learned.
VSA gives Cliste operations for binding content to roles and combining those structures. This supplies a way to compose; acquiring useful constituents and relationships remains research work. Earlier approaches include genetic_hackathon’s program composition, Scion’s transfer of learned symbols, graph structure in Sophia and TinyMind, and learned dependencies in Tabula Rasa. The common test is whether acquired structure enables something new.
Curiosity, confidence, and revision
What should a mind attend to, trust, and revise?
A system with limited time has to choose what to investigate. Uncertainty, contradictions, and the outcomes of earlier investigations are possible guides to that choice. Confidence also needs a way to fall when the evidence changes.
Cliste’s partial-cue study exposed a concrete confidence problem: a revised acceptance rule recovered more memories but also accepted more wrong matches. The visual ranking was unchanged. That separates finding a plausible candidate from having enough evidence to trust it. Attention and corroboration are possible next mechanisms to investigate; current cue locations and masks are supplied. Earlier TinyMind and Saoirse work explored curiosity through the measured consequences of pursuing a goal.
Emotion, salience, and disposition
How might internal state change what a mind notices and remembers?
The hypothesis is that emotion shapes cognition by changing what becomes salient, which memories are accessible, and which actions a system considers. Recent experience could affect perception temporarily, before reasoning; repeated experience might produce a more lasting disposition.
Sophia implements a persistent persona diary, and separate experiments apply and update an internal-state matrix over embeddings. Connecting the diary, retention across timescales, and these transforms into useful changes in continuing behavior remains the next question.
Embodiment and action
What changes when a learner can act?
I treat embodiment as central because acting lets a system test an expectation against a consequence. It does something, the world changes, and the outcome provides evidence for revising what it perceived, remembered, or expected.
Talos provides the sensor and actuator connection within LOGOS. The current Sophia inhabitant uses scripted actions, so it establishes a connection to a body without yet answering the learning question. Bringing acquired representations into an adaptive perception–action loop remains a direction for the work.
Continuity and cumulative learning
What makes learning accumulate?
Interruptions, new tasks, and new generations test what a system can carry forward. Keeping a history preserves earlier work; the stronger test is whether that history changes subsequent decisions and capabilities.
Cliste’s continuing-life experiments preserve acquired visual memory, working content, and intake position across new experience and restarts. That establishes state continuity; the stronger goal is learning that improves later use without destroying earlier competence. Scion’s inheritance studies and the memory and continuity tools approach related parts of that question.
Cognition and consciousness
How do these processes become a coherent mind?
The larger question is how perception, memory, attention, action, and a system’s understanding of itself fit together over time. Their interaction matters: a recalled experience could change what the system attends to, which action it chooses, and what it learns next.
Cliste is the current effort to bring these processes together, with the intention that she becomes Sophia. The human brain guides the questions about organization, learning, and economy. The present work concerns cognition and intelligence. Consciousness remains an open background question, with no experiment here that establishes it in these systems.