An ambitious horizon.
A precise conversation.

01 / The term

What is artificial superintelligence?

ASI generally describes a hypothetical form of AI with intellectual capability beyond human intelligence. IBM’s overview uses that hypothetical framing. It is a proposed level of capability, rather than the name of a particular interface, product architecture or business model. Source: IBM’s ASI overview.

That distinction matters for a company name. ASI can express a research interest or a long-term ambition, but its appearance in a name does not demonstrate a technical achievement. A business should describe its current work in terms that a customer or researcher can examine.

Nor does the label by itself answer questions about consciousness, reliability or authority. A system can produce an impressive result without being dependable in every situation, and a capability claim does not tell a buyer which actions the system is allowed to take.

02 / Better questions

How good, how broad and how independent?

In their 2024 ICML paper, Morris and colleagues propose a framework for discussing AGI that separates depth of performance from breadth of capability and considers autonomy separately. The framework provides a useful vocabulary for examining broad intelligence claims without treating every improvement as the same kind of progress. Source: Levels of AGI.

Consider a proposed assistant that prepares research briefs. A useful evaluation might examine whether statements are supported by their cited sources, how much correction reviewers need to make and which requests the system cannot handle. Those are product questions. A sweeping claim that the assistant “understands everything” would make the product harder to assess.

03 / A research direction

Can intelligence keep finding something new?

A 2024 position paper by Hughes and colleagues argues that open-endedness is essential to artificial superhuman intelligence. They discuss novelty and learnability, and explore systems capable of producing new discoveries meaningful to a human observer. This is the authors’ research position, not a settled definition or proof that such a system has been achieved. Source: Open-Endedness is Essential for Artificial Superhuman Intelligence.

For a founder, this raises an interesting design question: what would count as a useful new discovery in a particular field? Generating more variations is not enough by itself. A team needs a way to recognise whether an output is new, understandable and worth following up.

Our practical interpretation is to begin with a bounded learning loop. State the question, preserve the material used, review the result and decide what evidence would justify the next step. The team should also define when to stop or seek outside review. This workflow suggestion is editorial guidance, not a claim that it produces ASI.

04 / The company layer

Ambition can be broad.
Evidence should be specific.

A company interested in advanced intelligence could begin with research tooling, evaluation infrastructure or an assistant for a well-defined workflow. Its first product need not solve a grand research problem to be useful. It does need a clear account of the work it performs and the conditions under which that work is dependable.

A useful product description identifies its users, its inputs, its outputs and its limits. If a tool reads a document, can it show the passage supporting an answer? If it prepares an action, who approves it? If an answer changes, can a reviewer see why? These questions turn ambition into something a buyer can investigate.

Witasi can serve as the company-level identity above that work. WitASI can explain the possible category association when relevant. Keeping those levels separate lets the name remain expansive while the product story remains concrete.