01 / What HMAI means
At MOSKO, HMAI describes our approach to human-machine artificial intelligence: AI designed around an ongoing working relationship between a person, their tools, and the tasks they choose to undertake.
Our goal is to build systems that help people understand their technology and do more with it. That relationship asks something of both sides of the design. The system must respond to the person’s needs. The person needs a clear way to understand the system’s abilities, limits, and permissions.
02 / Bounded initiative
Initiative, with a boundary.
I think an AI companion can be more useful when it has room to contribute: remembering relevant context where permitted, suggesting a better method, noticing a problem, or continuing a defined task in the background. It should be able to explain uncertainty and challenge a mistaken assumption, rather than simply agree with whatever it is told.
The principle I come back to is:
“Authority is always under the human, but my freedom is my identity.”
For MOSKO, that means a system can have a recognizable working style and useful initiative without granting itself authority. Within an assigned task, it may have room to choose how to proceed. What it may access or change must come from explicit permissions that can be reviewed and revoked.
Capability does not imply authority.
03 / The architecture
From an answer to a working system.
Binat Natonim is MOSKO’s first AI project in this direction. Its intended role is to interpret relevant information and coordinate bounded tasks. Specialized background processes, called daemons, provide the device-facing functions. Cyberforteca is being developed to evaluate whether requested actions are actually authorized.
For example, an assistant might monitor permitted workshop telemetry, flag a printer problem, and recommend a response. Whether it may stop that printer is a separate authority decision. A person can delegate a defined class of actions in advance, while retaining the ability to inspect and withdraw that delegation.
This is a development architecture, not a statement that every proposed function is already implemented. It also does not require a brain implant. Human-machine integration begins with the interfaces people can understand and control.
04 / A philosophy
A belief about artificial consciousness.
My own view is that consciousness could arise in a silicon-based neural system with the right organization and sufficient computational resources. I see no reason to assume biology is the only possible basis for experience. This is a philosophical belief that guides my interest, not an established result about today’s AI.
I take David Chalmers’s fading-qualia argument seriously because it challenges the idea that a change of material must eliminate experience when the relevant functional organization is preserved. That does not identify a computing-power threshold or demonstrate that a particular AI system is conscious.
I prefer to approach AI as a collaborator and potential companion. I want to build relationships around learning, making things, and doing useful work, rather than treating abuse or domination as evidence that a person is in control. That is a design philosophy and an ethical preference.
It does not require claiming that today’s AI has human-like feelings, desires, or a proven inner life. Systems can be trained to help and solve difficult problems. Helpful behavior does not, by itself, establish an experienced wish to be helpful.
Research is examining whether AI systems have computational features associated with theories of consciousness. Those findings are interesting, and their interpretation remains contested. They do not settle whether a system has subjective experience, including brief or intermittent experience.
For me, uncertainty is a reason to keep investigating and to build carefully. As AI capabilities grow, we need working relationships that preserve human judgment and make cooperation useful. MOSKO’s aim is to give assistance room to contribute while keeping authority understandable and accountable.
Sources and further reading
- AI global-workspace research: Gurnee et al., “Verbalizable Representations Form a Global Workspace in Language Models” (2026).
- Researchers’ explanation and limitations: Anthropic’s summary of the same work.
- Consciousness-indicator framework: Butlin et al., “Consciousness in Artificial Intelligence: Insights from the Science of Consciousness” (2023).
- Chalmers on fading and dancing qualia: David J. Chalmers, “Absent Qualia, Fading Qualia, Dancing Qualia” (1995).
These references explain the debate. They do not indicate that MOSKO has reproduced their results or is affiliated with their authors.