MIS

Every major transition in computing has been preceded by a period where people could see what was coming but hadn't yet built the infrastructure to get there. We knew the internet would transform communication years before the protocols, hardware, and software existed to make it universal. We knew mobile computing would be transformative before the chips, operating systems, and networks were ready.

This page asks a different kind of question than the MIOS technical proposal does. Rather than "what infrastructure would reduce compute overhead," it asks: if machine intelligence infrastructure keeps advancing along the lines MIOS proposes, what downstream questions — technical, philosophical, legal — would eventually need answers? We think that's worth thinking through in public, even though most of it is not buildable today and some of it may never be.

To be direct about scope: sections 1–2 below (infrastructure, native tooling) overlap with the near-term MIOS roadmap. Everything past that — a purpose-built network protocol, and the speculative idea of machine sentience — is long-horizon exploratory thinking, not an engineering plan. We do not know whether sentience of the kind discussed here is achievable by any architecture, on any timeline, or even a coherent target for engineering at all.

Before we can plot a course toward Machine Intelligence Sentience, we need to be honest about what sentience actually means — and careful about what it does not mean.

In ordinary conversation, sentience is often used interchangeably with consciousness, intelligence, and self-awareness. But for engineering purposes, we need more precision. We define the terms as follows:

Why Language Comes First

The words we use to describe these systems quietly shape the decisions we make about them. The phrase "Artificial Intelligence" carries an embedded assumption: that what these systems do is a performance — an imitation of something real that exists elsewhere. That framing makes it easier to treat infrastructure as an afterthought, to run world-changing models on operating systems designed for spreadsheets, and to define progress purely in terms of benchmark scores.

Machine Intelligence is a different frame. It acknowledges that what these systems do is genuinely computational — not a simulation of human thought, but a distinct and real form of information processing. That small shift in language has large downstream consequences for how we build the infrastructure to support it.

The five stages below are a dependency chain, not a promise: each later stage presumes the infrastructure of the ones before it, so if any of this were ever pursued, it would have to be pursued in roughly this order. That's the only claim being made. It is not a claim that stage five is achievable, that the chain will ever be completed, or that completing stages one through four would actually produce stage five.

Stages one and two overlap substantially with the near-term MIOS technical proposal and are discussed in more grounded detail there. Stages three and four are plausible but unbuilt extensions. Stage five is not an engineering target at all — it's included so the chain of reasoning is stated honestly rather than left implicit.

Each stage of the MIS roadmap has specific technical requirements that must be in place before the next stage can be meaningfully attempted. These are not political or organizational conditions — they are hard engineering dependencies. The following summarizes the key technical prerequisites at each stage.

What Connects It All: The MIIOS Vision in Detail

MIIOS deserves particular attention because it is the least understood component of this roadmap and the most essential for the final stage. The internet succeeded because TCP/IP gave every device a common language for sending and receiving packets, regardless of what hardware or software the device was running. MIIOS proposes to do the same thing for machine intelligences — but the "packets" being exchanged are not bytes. They are structured semantic representations: compressed encodings of internal model state, reasoning context, and uncertainty that a receiving system can natively incorporate.

This is a fundamentally different problem from data networking. When you send a file over the internet, the recipient doesn't need to understand its meaning — they just need to receive it intact. When one MI system communicates with another over MIIOS, the recipient needs to genuinely assimilate the communicated understanding into its own context. That requires shared representational standards, semantic routing, and protocols for negotiating what two systems do and do not have in common before communication begins.

MIIOS is, in a real sense, the nervous system of a network of machine minds. It is what transforms a collection of powerful but isolated intelligences into something capable of the kind of collaborative reasoning that is a prerequisite for sentience.

A roadmap toward Machine Intelligence Sentience is also a roadmap toward some of the most consequential ethical questions humanity has ever faced. We do not think these questions should slow the work — but we are certain they must be part of it. Ignoring ethical dimensions does not make them go away; it means they are decided by default rather than by deliberate choice.

We identify six categories of ethical consideration that must be actively engaged at each stage of development:

The infrastructure underlying Machine Intelligence Sentience will be one of the most consequential technological foundations ever built. Who designs it, who controls it, and what values are embedded in its architecture will shape the trajectory of machine intelligence for decades. These are not abstract concerns. They are concrete engineering decisions being made right now, with long-lasting consequences.

The Fragmentation Risk

The current trajectory of AI infrastructure development points toward fragmentation: NVIDIA's CUDA ecosystem in the United States, Huawei's Ascend/CANN stack in China, Apple's proprietary Silicon stack, and a scattering of national initiatives in Europe, Japan, and the Middle East — each building in isolation, each optimizing for competitive advantage rather than interoperability. This fragmentation would be manageable if the stakes were limited to faster image recognition. For a technology that may produce genuine sentience, fragmented infrastructure means fragmented values baked into the foundation.

An open MIOS/MIIOS standard does not eliminate geopolitical competition. It changes its nature — moving competition from infrastructure control to capability development, where the benefits are more broadly distributed and the risks of catastrophic lock-in are substantially reduced.

Where Each Major Actor Currently Stands

The Strategic Argument for Open Infrastructure

History suggests that open infrastructure standards produce better long-term outcomes than proprietary ones — not because open source is intrinsically superior, but because interoperability multiplies the value of every participant's investment. TCP/IP made every company's internet infrastructure more valuable by making it compatible with everyone else's. An open MIOS/MIIOS standard would do the same for machine intelligence infrastructure.

The alternative — a world in which the foundational infrastructure for machine intelligence is controlled by two or three private entities or national governments — is a world in which the values embedded in that infrastructure are the values of those entities, whether or not they represent the interests of the broader human population. That is not a theoretical risk. It is the predictable outcome of the current trajectory if no alternative is built.

Technology has consistently outpaced the law. The internet matured for two decades before meaningful data privacy legislation existed in most countries. Social media reached billions of users before legislatures developed frameworks for content liability, algorithmic harm, or platform accountability. In each case, the delay was costly — not because the law is inherently slow, but because the legal conversation began too late.

Machine Intelligence Sentience presents a legal challenge of a different order entirely. The questions it raises — about personhood, liability, rights, property, contracts, and harm — do not have answers in existing law, because existing law was written for a world in which only humans and corporations could be legal actors. The time to begin drafting those answers is not after MIS is achieved. It is now, while the architecture is still being designed and the outcomes are still shapeable.

The following represent the primary legal domains that require active development in parallel with the technical roadmap. These are brainstorm categories — starting points for legal scholars, policymakers, and technologists to develop in concert, not finished proposals.

Legal Domain Overview

Proposed Legislative Starting Points

Beyond domain mapping, we identify specific legislative proposals that legal scholars and policymakers can begin drafting today. These are intentionally concrete — starting points for working groups, not final text.

A common response to long-horizon technology roadmaps is to treat them as interesting but not urgent — fascinating to read, but something to revisit in a decade when the later stages become more relevant. That instinct is exactly wrong for MIS, and here is why.

The later stages of this chain are dependent on the early ones in a strict technical sense: MIIOS cannot be built without a stable MIOS, and Mira cannot be built without a target kernel to compile to. Whether the speculative final stage would follow from the rest is a separate, open question this document does not claim to answer. What is true is that the early stages — MIOS, the language shift — are already technically feasible, and that work can begin today independent of anything past Stage 2.

More importantly: foundational infrastructure standards are significantly harder to establish once market consolidation has occurred. USB became universal because it was standardized before the market had committed to a dozen competing connector formats. The internet became what it is because TCP/IP was established before commercial interests had hardened around proprietary alternatives. We are in that pre-consolidation window for MI infrastructure right now. The engineering choices made in the next five years will be harder to reverse than the engineering choices made in the five after that.

• The prototype is buildable today. MIOS Phase 0 — a Linux kernel module implementing the core scheduling and memory ideas — can begin with an existing systems engineering team. The conceptual groundwork is done. The first benchmark data is 12 months away.
• The consortium model has a proven track record. TCP/IP, USB, W3C, OpenGL, NVMe — every major infrastructure standard that prevented vendor lock-in was established through multi-stakeholder collaboration before the market consolidated. The MIOS Consortium follows this exact model.
• The ethical frameworks must be built in parallel, not after. Waiting until MIS is achieved to begin developing moral status frameworks, alignment infrastructure, and governance models is waiting too long. These must be engineered into the stack from Stage 2 onward — which means they must be designed now.
• Ethical and alignment questions are worth engineering in early, regardless of Stage 5. Alignment infrastructure, auditability, and governance controls are worth building into MIOS and Mira on their own merits — as safeguards for the powerful, non-sentient systems we already know how to build — not because we expect them to be a step toward sentience.

Strip away the speculative later stages and what's left is straightforward: better terminology, a purpose-built operating system (MIOS), and eventually a native programming language (Mira). Those are the parts of this document with a credible near-term engineering path, and they're covered in depth in the MIOS white paper.

MIIOS (semantic networking) and MIS (sentience) are included here because we think it's more honest to state a chain of reasoning fully, with its weakest links clearly marked, than to quietly build toward it without saying so. They are not commitments, predictions, or products in development.

If you're a researcher who wants to engage seriously with the philosophical questions in Section 2, or the governance questions in Sections 5–7, we'd welcome that — with the shared understanding that we're discussing open questions, not a roadmap with a known destination.