Re: Invitation to Co-Shape a 3-D Standard for AI Knowledge — Collaboration Opportunity with K3D

Thanks for pointer Daniel Ramos,
There's a lot of work needed based on looking at the node schema and the Solar System example. Not sure I can usefully help until it's more advanced. Which could be achieved by more directly incorporating existing best practice for ontologies and linked data.

Some initial reactions (I appreciate this goes beyond what you were trying to achieve with the example so is probably unfair, but it'll provide some aspirational goals maybe)

  *
The example has no type information e.g. to even distinguish planets from star (the sun)
  *
And the types will need to be better identified than a string (e.g. To distinguish the astral body Star from the performers with top billing in a show)
  *
There seems to be no unique identifier e.g to distinguish the planet Mars, from Mars the Roman God of War, or Mars the confectionary bar.
  *
Need units associated with the numbers such as mass, diameter. The mass is interesting in that the unit is the mass of Earth (M⊕​). Though the distance from sun is not in terms of Astronomical Units (au) (distance from Earth to Sun) which seems a bit inconsistent
  *
Also should make clear that the diameter and distance from sun are averages (planets are not perfect spheres and orbits are elliptical with different degrees of eccentricity)
  *
For the planetary positions it's not clear what the 3d origin is meant to be
  *
Nor the point in time the snapshot represents. Obviously the planets are not static and will be continuing to move relative to one another; and the whole system moving within the galaxy. Nicely shown here https://www.youtube.com/watch?v=fJuaPyQFrYk - interesting challenge for the K3D Universe!
  *
The meaning of "neighbors" is unclear. I'm not sure many would say that Mercury was a neighbor of Neptune. Rather than neighbor I'd expect some sort of hierarchy to say they're in orbit round the sun (the hierarchy could be extended to moons orbiting a planet) or all part of the Solar System as an object. Would the sun (or the solar system as a whole) be a "neighbor" to the nearest star(s)?

Hope that's useful. I wish you good luck but am unlikely to be able to spend significant time helping. Maybe once you've got something you think is pretty complete I could review.

Pete
________________________________
From: Daniel Ramos <daniel@echosystems.ai>
Sent: Thursday, August 14, 2025 10:51 AM
To: Peter Rivett <pete.rivett@federatedknowledge.com>; public-aikr@w3.org <public-aikr@w3.org>
Subject: Re: Invitation to Co-Shape a 3-D Standard for AI Knowledge — Collaboration Opportunity with K3D


Hi Pete,

Thanks for digging into the Knowledge3D repo and for the thoughtful questions.

Where the nodes live

The current draft schema for a single K3D node is in spec/k3d_node_schema.json. Every node carries:

  *   id: unique identifier

  *   vector: 3‑D projection

  *   embedding: original high‑dimensional vector

  *   metadata: open object with at least label (human‑readable) and type fields, plus any additional facets you need (technology, legislation, ethics, provenance, etc.)

The README’s “Key Concepts” section points to that schema and explains how nodes are linked to geometry through a glTF extension

.

Plain-language examples

You can see plain-language labels in action in the sample datasets—for example, examples/solar_system.k3d stores each planet’s name in metadata.label, alongside its embedding and neighbor links

Status of the vocabulary layer

We don’t yet ship a canonical controlled vocabulary. The metadata object is deliberately open so you can align nodes with your own or community ontologies (SKOS, schema.org, sector‑specific terms, etc.). We’re still drafting this part of the standard and would love your input—especially given your Graph RAG experience.

If you’d like to join the discussion or contribute proposals, just let me know; we’re aiming to converge on a reusable vocabulary layer that plugs neatly into graph/ontology tooling while keeping the node format lightweight.

Best regards,
Daniel Ramos
daniel@echosystems.ai

On 8/13/25 7:19 PM, Peter Rivett wrote:
I'm trying to follow up on what you wrote, Daniel:

Rich, reusable vocabularies
K3D links plain‑language terms to formally typed nodes, making ontologies explorable by humans and machines.
Facets & provenance
Each K3D node carries first‑class metadata (technology, legislation, ethics, etc.)—a perfect companion for your “resources & metadata” track.




I'm having problems tracking down the plain-language terms or the K3D nodes in the repo - can you please provide a pointer?

Thanks
Pete

Pete Rivett (pete.rivett@federatedknowledge.com<mailto:pete.rivett@federatedknowledge.com>)

Received on Tuesday, 19 August 2025 01:18:28 UTC