- From: 陳信屹 <tyson@slashlife.ai>
- Date: Wed, 3 Dec 2025 14:56:42 +0800
- To: Milton Ponson <rwiciamsd@gmail.com>
- Cc: Paola Di Maio <paoladimaio10@gmail.com>, internal-aikr@w3.org, W3C AIKR CG <public-aikr@w3.org>, public-s-agent-comm@w3.org
- Message-ID: <CAGRPLTNJ8FeZK_dHVXHC9DaTFkn1+Dm1u0Yz-23wwdMDFowc5w@mail.gmail.com>
Noted the two references — I’ll cross-map AgentIDL’s KR interface layer
accordingly.
Below is a plain explanation of the currenttestbed I am using for AgentIDL.
1. Purpose of the Testbed
The testbed explores how an intelligent agent can move from
natural-language instructions to structured, verifiable actions.
The core question is:
How can an agent establish and follow ontological commitments during
execution?
In practice, this means checking whether the agent can:
-
interpret a user task,
-
align it with a shared knowledge structure (TTL/OWL or other KR assets),
-
produce a stable, machine-readable representation,
-
and then execute the task in a deterministic way.
2. What AgentIDL Does in This Setup
AgentIDL acts as an intermediate language between natural language and
actual execution.
The testbed only uses a small subset:
-
simple actions
-
parameters
-
assertions
-
references to vocabularies or concepts
The goal is not completeness but observing where ontological alignment is
required, and where current models fail without KR support.
3. How TTL / KR Interfaces Are Used
In the testbed, TTL is used to:
1.
Declare the concepts the agent should commit to.
2.
Define the allowed actions or roles associated with those concepts.
3.
Provide a stable vocabulary so that different agents/models interpret
instructions consistently.
Right now, this is minimal—just enough to observe how the presence or
absence of KR grounding changes the quality of the agent’s output.
4. What the Testbed Actually Runs
A typical run looks like this:
1.
The user gives a natural-language task.
2.
The model generates a provisional AgentIDL program.
3.
The program is checked against a simple KR layer (TTL).
4.
The agent executes the steps and records:
-
what it understood
-
where alignment succeeded
-
where ontological gaps appear
-
what was ambiguous or unstable
This produces a clear picture of where structured knowledge is needed and
what kind of KR interface is missing today.
5. What We Are Learning
The early results show:
-
agents perform better when they have a stable set of concepts to commit
to;
-
KR artefacts (even minimal ones) reduce ambiguity;
-
models generate more consistent AgentIDL programs when guided by simple
TTL structures;
-
gaps in KR directly correspond to execution errors or unstable reasoning.
This supports the idea that we need a formal KR interface category for
agent systems.
6. Why This Matters
The testbed is not about building a full system.
Its purpose is to expose the boundary between LLM behaviour and KR
commitments, and to understand:
-
where ontology is required,
-
how agents negotiate meaning,
-
how to prevent divergence or hallucinated actions,
-
how to design interfaces that help agents behave predictably.
This provides a foundation for future artefacts in the CG, and helps
clarify where novel contributions may emerge.
Milton Ponson <rwiciamsd@gmail.com>於 2025年12月2日 週二，下午11:09寫道：
> Paola and all,
>
> Sadly this issue of plagiarism has haunted me from the very first day I
> decided to create a W3C user account with the intent to participate in CGs
> for the purpose of helping to develop technologies and standards that could
> benefit the mixed bag of tools commonly known as ICT4D. And as Paola
> indicated in her post on how her emergency management articles, dealing
> with one category of ICT4D were plagiarized, nowhere is the plagiarism more
> rampant than in the open source community.
> Attempts to create regulations for intellectual property rights protection
> and fair use are ignored in the US arms race worth trillions based
> primarily on generative AI using LLMs.
>
> I have noticed as many of you will have undoubtedly too, that scouring
> openly accessible archives of public W3C CG mailing lists and even inside
> mailing lists is impossible to avoid.
>
> If you follow RetractionWatch and are familiar with the organisations like
> Clarivate, you must be aware that AI orchestrated scouring and AI driven
> article production through so-called paper mills have become a veritable
> large scale problem, and peer review processes often fail to capture this,
> and even remedial measures when infringement has been detected, are not
> effectuated.
>
> In light of recent attempts by the US Congress and the White House to
> overturn all US individual state regulations of AI and a number of widely
> followed copyright infringement cases between large players in AI and
> groups of artists, news media and even publishing houses, it is becoming
> evident that stealing intellectual property under the guise of fair use of
> material is now the de facto norm.
>
> W3C should review its current policies and in particular create effective
> mechanisms for sanctions against IP "scouts", masquerading as individuals
> using individual CLA commitments.
>
> This review should include explicit mentions of how publicly archived
> threads from mailing lists of W3C CGs should be used.
>
> Paola, Tyson, Daniel and myself all sense that our efforts are being
> monitored, and we have to be careful in how much of our work should be made
> publicly known in threads.
>
> The AIKR CG deals with one of the most important issues in current AI
> research and consequently is most prone to be victimized by plagiarism.
>
> It would be an idea to determine if AI driven tools can be developed to
> trace how potentially IP sensitive material gets accessed in open archives,
> repositories and fora and if it ends up regurgitated in someone else's work
> without due accreditation of sourced materials.
> And use semantic technology tools in this process.
>
> Consider this as a suggestion for a practical use case for KR standards
> for AI to tackle a problem for which collaboration with other actors will
> not be any problem.
>
> Milton Ponson
> Rainbow Warriors Core Foundation
> CIAMSD Institute-ICT4D Program
> +2977459312
> PO Box 1154, Oranjestad
> Aruba, Dutch Caribbean
>
> On Tue, Dec 2, 2025, 04:03 Paola Di Maio <paoladimaio10@gmail.com> wrote:
>
>>
>> We must conduct due diligence and check out the literature/state of the
>> art
>> to make sure we do not overlook and plagiarize, even unintentionally,
>> other people's work in this space
>>
>> A genuine contribution, worthy of funding, is novel and advances the
>> state of the art and ackowledgeges prior art
>> Much of research unfortunately is bluntly the opposite
>>
>> Unfortunately, funding institutions lack the capacity to evaluate the
>> state of the art because..... entire ecosystems are built on
>> malpractice. Evaluators who object to large consortia getting massive
>> funds for fake projects are simply ejected from the
>> evaluation system
>>
>> Below two pointers, and note that Peter PS is in this list, There is a
>> vid from him in the link
>> More due diligence follows
>>
>> Paola
>>
>> Patel-Schneider has contributed to discussions on semantic web standards
>> and usability that are crucial for reliable knowledge interfaces supporting
>> intelligent agents (Patel-Schneider, 2024).
>>
>> If including foundational theory and semantic web research is desired,
>> Patel-Schneider’s work on OWL and semantics bridges KR theory and
>> agent-oriented applications.
>>
>> Patel-Schneider, P.F. (2024). "Does the Knowledge Graph community care
>> about semantics?" COST DKG Talk Series.
>> https://research.cs.umbc.edu/kqml/papers/desiderata-acl/section3.3.html
>>
>> INTERNATIONAL JOURNAL OF COMPUTERS COMMUNICATIONS & CONTROL
>> ISSN 1841-9836, 13(1), 117-128, February 2018.
>>
>> *Modern Interfaces for Knowledge Representationand Processing Systems
>> Based on Markup Technologies*
>> A. A. Mohammed Saeed, D. Dănciulescu
>>
>> https://scispace.com/pdf/modern-interfaces-for-knowledge-representation-and-3i2ld58n1u.pdf
>>
>>
>>
>> On Tue, Dec 2, 2025 at 2:43 PM Paola Di Maio <paoladimaio10@gmail.com>
>> wrote:
>>
>>> Tyson, thank you for looking at the artefacts in development
>>> It is great that you are finding it useful
>>>
>>> I am working on a bunch of artefacts and experimenting with ways this
>>> can be built into agents!!
>>> I am a bit wary of sharing because I would like to look for funding for
>>> the deserving portions of this work
>>>
>>> You ask when am I going to publish - asap, as soon as I can free my
>>> hands, but publishing does not
>>> mean that it will not get plagiarized, pardon the sour rant.
>>>
>>> SOUR RANT BEGINS
>>> *many academics regularly scour mailing lists to plagiarise other
>>> people's work, including when it is already published, acknowledged and in
>>> use
>>> There are a lot of academics who get funding for their own project and
>>> do not even give credit
>>>
>>> Sad, and very frequent. My own original work in ontology, engineering
>>> and emergency management has been plagiarized and submitted for funding
>>> *with public funding agencies Including my PhD thesis. There are scores
>>> of funded projects and published papers as deliverables of such funded
>>> that cite other plagiarised papers that are copied straight out from other
>>> published works. There are enemies among us. Nobody can afford to pursue
>>> expensive legal battle to demonstrate that the academics so and so, in the
>>> public unversities so and so that receive public funding from all
>>> major public councils world wide are actually criminal organisations
>>> END OF SOUR RANT
>>>
>>> I suppose posting to the internal mailing list is a good idea when
>>> sharing novel artefacts. See internal mailing list address in cc
>>> But all list members, including serial plagiarisers may also be on the
>>> internal list.
>>> I ll keep the roadmap reserved for the moment
>>>
>>> In the meantime, I have given it a preliminary DOI
>>> 10.6084/m9.figshare.30760739
>>>
>>> It would be great to continue to bounce in this direction
>>> what I got from this exchange so far is
>>>
>>> 1. We need a new category *bubble in our concept map for KR interfaces
>>> *I have already done some literature review and apparently Pater Patel S on
>>> this mailing list did some work on this some 25 years ago, I have not had
>>> time to summarize the state of the art. There are KR interfaces out there
>>> to be gathered
>>>
>>> 2. In this CG we are creating some KR artefacts that can be used by
>>> agents/models to establish ontological commitment *to refine and improve
>>> the quality of the queries and possibly improve overall performance? *I
>>> think so
>>>
>>> Do let us know how your test bed works but please explain it in plan
>>> words?!
>>>
>>> I do have a couple more demos myself I can share..... er...
>>>
>>> PDM
>>>
>>> On Tue, Dec 2, 2025 at 11:20 AM 陳信屹 <tyson@slashlife.ai> wrote:
>>>
>>>> To further explore this alignment, I’d like to propose using it as the
>>>> first testbed use case within our CG. a small demo video that
>>>> showcases how the Unified Ontic Ontology can ground the Agent Ontology
>>>> in an executable environment.
>>>>
>>>> I don’t yet have a concrete idea for what the best showcase would be.
>>>> perhaps something minimal that demonstrates
>>>> Intent–Capability–Delegation flow in a verifiable way.
>>>> I’d be happy to hear thoughts or suggestions from everyone on what
>>>> kind of example could best illustrate the value of this alignment.
>>>>
>>>> 陳信屹 <tyson@slashlife.ai> 於 2025年12月2日週二 上午11:02寫道：
>>>> >
>>>> > Hi Paola
>>>> >
>>>> > I’ve been reviewing your Unified Ontic Ontology work. the
>>>> > cross-mapping across BFO, DOLCE, SUMO, GFO, UFO and YAMATO is an
>>>> > excellent unification of top-level categories.
>>>> >
>>>> > To explore its practical impact inside our Semantic Agent
>>>> > Communication CG, I drafted a minimal alignment block that connects
>>>> > the Agent Ontology (core modules: Agent, Intent, Capability,
>>>> > Delegation …) to your Unified Ontic framework. This gives the Agent
>>>> > Ontology a stable ontic frame.
>>>> >
>>>> > From the engineering side, this alignment allows the AgentIDL compiler
>>>> > to automatically bind each interface (Intent, Capability, Delegation)
>>>> > to its ontic category in Unified Ontic Ontology. This makes the
>>>> > generated code semantically interoperable across domains. The same
>>>> > AgentIDL interface can compile into different ontological environments
>>>> > (BFO, DOLCE, SUMO) while preserving meaning.
>>>> >
>>>> > Concretely, we could use it not by importing the whole thing, but by
>>>> > referencing key classes.
>>>> >
>>>> > Here’s the snippet (for review / discussion):
>>>> >
>>>> > ################################################################
>>>> > # Ontic Alignment — Unified Ontic Ontology
>>>> > # (Light-weight mappings to BFO/DOLCE/SUMO/GFO/UFO/YAMATO families)
>>>> > ################################################################
>>>> >
>>>> > @prefix exu: <http://example.org/unified-ontic#> .
>>>> > @prefix agent: <
>>>> https://s-agent-comm.github.io/ontology/ontologies/agent#> .
>>>> > @prefix intent: <
>>>> https://s-agent-comm.github.io/ontology/ontologies/intent#> .
>>>> > @prefix capability:
>>>> > <https://s-agent-comm.github.io/ontology/ontologies/capability#> .
>>>> > @prefix contract:
>>>> > <https://s-agent-comm.github.io/ontology/ontologies/contract#> .
>>>> > @prefix ledger: <
>>>> https://s-agent-comm.github.io/ontology/ontologies/ledger#> .
>>>> > @prefix delegation:
>>>> > <https://s-agent-comm.github.io/ontology/ontologies/delegation#> .
>>>> > @prefix security:
>>>> > <https://s-agent-comm.github.io/ontology/ontologies/security-binding#
>>>> >
>>>> > .
>>>> > @prefix rdfs: <http://www.w3.org/2000/01/rdf-schema#> .
>>>> > @prefix owl: <http://www.w3.org/2002/07/owl#> .
>>>> >
>>>> > ################################################################
>>>> > # Core class alignments
>>>> > ################################################################
>>>> >
>>>> > agent:Agent a owl:Class ;
>>>> > rdfs:label "Agent" ;
>>>> > rdfs:comment "An autonomous computational entity with identity
>>>> and intent,
>>>> > modeled as a social artifact capable of delegation
>>>> > and accountability." ;
>>>> > rdfs:subClassOf exu:ArtifactSocial ;
>>>> > exu:mapsTo "BFO:MaterialEntity->Artifact; DOLCE:SocialObject;
>>>> > SUMO:Agent; UFO:SocialAgent" ;
>>>> > rdfs:seeAlso <https://basic-formal-ontology.org/> ,
>>>> > <https://www.loa-cnr.it/DOLCE.html> .
>>>> >
>>>> > intent:Intent a owl:Class ;
>>>> > rdfs:label "Intent" ;
>>>> > rdfs:comment "A deliberative mental state or communicative
>>>> > commitment of an agent." ;
>>>> > rdfs:subClassOf exu:AbstractEntity ;
>>>> > exu:mapsTo "BFO:GenericallyDependentContinuant;
>>>> > DOLCE:Description/Proposition; SUMO:Intent; UFO:Intention" .
>>>> >
>>>> > capability:Capability a owl:Class ;
>>>> > rdfs:label "Capability" ;
>>>> > rdfs:comment "A realizable disposition or function that an agent
>>>> > can exercise." ;
>>>> > rdfs:subClassOf exu:FunctionDisposition ;
>>>> > exu:mapsTo "BFO:RealizableEntity/Function; DOLCE:Capability;
>>>> > SUMO:Capability; GFO:Disposition; UFO:Disposition" .
>>>> >
>>>> > contract:Contract a owl:Class ;
>>>> > rdfs:label "Contract" ;
>>>> > rdfs:comment "A social artifact representing binding obligations
>>>> > between agents." ;
>>>> > rdfs:subClassOf exu:ArtifactSocial ;
>>>> > exu:mapsTo "DOLCE:SocialObject; SUMO:Agreement;
>>>> UFO:SocialRelator" .
>>>> >
>>>> > ledger:Ledger a owl:Class ;
>>>> > rdfs:label "Ledger" ;
>>>> > rdfs:comment "A record artifact capturing accountable agent
>>>> transactions." ;
>>>> > rdfs:subClassOf exu:ArtifactSocial ;
>>>> > exu:mapsTo "BFO:InformationArtifact; DOLCE:InformationObject;
>>>> > SUMO:Document" .
>>>> >
>>>> > delegation:Delegation a owl:Class ;
>>>> > rdfs:label "Delegation" ;
>>>> > rdfs:comment "A social relation between agents defining authority
>>>> > transfer or responsibility." ;
>>>> > rdfs:subClassOf exu:RelationRelator ;
>>>> > exu:mapsTo "BFO:RelationalRole; DOLCE:Relation; SUMO:Relation;
>>>> > UFO:Relator" .
>>>> >
>>>> > security:ProofBinding a owl:Class ;
>>>> > rdfs:label "ProofBinding" ;
>>>> > rdfs:comment "A verifiable binding between identity, proof, and
>>>> > execution context." ;
>>>> > rdfs:subClassOf exu:DependentContinuant ;
>>>> > exu:mapsTo "BFO:Quality; DOLCE:Quality; SUMO:Attribute" .
>>>> >
>>>> > ################################################################
>>>> > # Optional note
>>>> > ################################################################
>>>> >
>>>> > <https://s-agent-comm.github.io/ontology/ontologies/core#>
>>>> > rdfs:comment "This ontology is ontically grounded via Unified
>>>> > Ontic Ontology alignment,
>>>> > enabling cross-compatibility with major upper
>>>> > ontologies (BFO, DOLCE, SUMO, UFO).
>>>> > The mapping ensures that computational agents are
>>>> > modeled as social artifacts
>>>> > with realizable capabilities, intentional states,
>>>> > and relational accountability." .
>>>> >
>>>> >
>>>> > By the way, do you have an estimated timeline for publishing or
>>>> > finalizing the Unified Ontic Ontology (TTL version)?
>>>> > I’d like to align the AgentIDL compiler’s ontic bindings with your
>>>> > official release once it’s available.
>>>> >
>>>> > Paola Di Maio <paola.dimaio@gmail.com> 於 2025年11月28日週五 下午6:49寫道：
>>>> > >
>>>> > > Thinking of where would Tyson's agentic execution interface
>>>> specification would sit in the diagram, *presumably in the KR
>>>> Language/Formalism bubble?
>>>> > > I have extracted captured in a table and a list of terms most
>>>> ontic categories, which is going to be our first deliverable,
>>>> > > please give feedback, help to improve/expand.
>>>> > >
>>>> > > Please request access to view
>>>> > >
>>>> https://docs.google.com/document/d/1OyZGVDCMozbAGPaKqYpUnWX75raGoCNpd6KJN3bMoFQ/edit?usp=sharing
>>>> > >
>>>> > >
>>>> > >
>>>> > >
>>>> > >
>>>> > > Please request access if you would like to give feedback
>>>>
>>>
Received on Wednesday, 3 December 2025 06:56:58 UTC