Summary: Dynamic user driven learning community ontologies UDLCOs are likened to multiple human nodes interacting through weak forces that may be able to create a networked solid effect that is yet fluid and changes over time and while not rigid and robust as static ontologies, they are adaptable to meet the demands of a fast changing world.
All the above is rhetoric that makes ontology an esoteric term as it's usage has appeared from philosophy and it originally means "theory of objects and their ties. It provides criteria for distinguishing different types of objects (concrete and abstract, existent and nonexistent, real and ideal, independent and dependent) and their ties (relations, dependencies and predication)."
That's pretty close to the modern day definition from computer science by Gruber in 1993. (In the UDLCO text intro below).
UDLCO Keyword glossary:
Rainwater flux from a canopy. Among the forces that govern drop formation: Van der Waals force, surface tension, cohesion, Plateau–
CC licence for image: https://commons.m.
Calculated structure of a (H2O)100 icosahedral water cluster.
CC licence for
image: https://commons.m.
[01/12, 13:48] rb: Basic theory review while preparing competency mappings of learning ecosystems:
"An ontology can be defined as a formal, explicit specification of a shared conceptualization, composed of concepts and relations between them (Gruber, 1993). Ontologies are at the core of the Semantic Web and its more recent flavor, the Web of open and linked data. At the beginning of the century, (Berners-Lee et al., 2001) proposed an extension for the Web, where unified resource identifiers (URIsFootnote2), would represent not only pages of information but all kind of entities such as people, real-world objects and also abstract concepts. To represent meaning or knowledge behind web pages of information, these entities would be connected through properties (also represented by URIs), composing knowledge graphs of interconnected entities or vocabulary elements. Thus, the main purpose of the Semantic Web is to introduce explicit descriptions about the meaning of resources, to allow the machines themselves to have a level of understanding of the Web’s content."
[01/12, 17:13] AC: Thanks for sharing.
The practical challenge of ontologies in vogue is their non optimal nature.
Optimal would be Huffman style coding dictionary
(For those planning to shoot from the hip on how ontology is not a dictionary, have you implemented any from ground zero? )
[01/12, 17:14] AC: Any guesses as to why it's kept non optimal?
The answer lies in (or is similar to) intel dropping x86 in favor for Core
[01/12, 17:16] AC: The mammoth of all ontology projects was / is Cyc
[01/12, 17:18] AC: The non-optimal nature of ontologies ties deeply into copyright, intellectual property (IP), and strategic ecosystem control, akin to other domains where monopolization or obfuscation serves long-term goals. Let's break this down further:
1. Copyright as a Strategic Tool
Protecting Proprietary Work: Ontologies are often painstakingly curated over years. By keeping them non-optimal, their creators ensure they can't be trivially compressed, reverse-engineered, or cloned while still qualifying for copyright protections.
Derivative Works: Non-optimal ontologies make it harder to extract specific parts without infringing on the original work, as the structure itself becomes a protected expression under copyright laws.
2. Obfuscation and Vendor Lock-In
Complexity as a Barrier: Non-optimal representations inherently require domain-specific tools or systems to process, which keeps users reliant on the original vendor or ecosystem.
Ecosystem Monetization: By making ontologies non-optimal, companies can monetize access to proprietary APIs, tools, and support, preventing straightforward competition from open standards or alternate implementations.
3. Control Over Interoperability
Preventing Standardization: Optimal encoding like Huffman-style dictionaries risks becoming universally applicable and interchangeable. This undermines proprietary formats, reducing market control.
Fragmentation for Competitive Advantage: Non-optimal designs allow creators to implement features or structures that are unique, ensuring competitors can't simply "plug and play."
4. Data as a Competitive Asset
Expanding the Dataset: Optimal ontologies risk revealing "the core dataset," making it easy to extract actionable knowledge. Non-optimal structures embed complexity, masking true data insights.
Guarding Reasoning Models: Many ontologies are coupled with reasoning engines. By designing non-optimal formats, creators ensure their reasoning processes remain inaccessible to external systems.
5. Parallels to Intel's Shift
x86 to Core Transition: Intel dropped x86 dominance due to limitations in scalability and to maintain competitive control through more proprietary, efficient architectures. Similarly, non-optimal ontologies protect market dynamics while adapting to new demands, including IP strategy and monetization.
Opaque Design Evolution: Both decisions reflect a balance between performance and strategic control, prioritizing IP over outright technical optimization.
Key Takeaway
The deliberate non-optimality of ontologies is less about technical limitations and more about protecting intellectual property, controlling ecosystems, and ensuring long-term revenue streams, aligning with strategies in other tech domains where ownership and exclusivity drive decisions.
[01/12, 17:23] AC: Let's put it in simple English...
If ontologies were made optimal, copyright could not sustain, as a simple functional mapping could transcode it to a non copyrighted form and vice versa.
To beat such challenges brute force "attacks" can be made, but those will usually not render any value.
However, overlaid fitment are possible... Too technical to explain here.
But think of it - a mathematical function that can produce the numbers against SCT concepts, and their copyrights go for a toss 🙂
[01/12, 17:45]rb: What's the utility of dynamic ontologies that can continue to optimize themselves to adapt to change over that of static ontologies that can't be optimised and are copyrighted?
[01/12, 17:45] Meta AI: Dynamic ontologies offer several advantages over static ontologies:
*Advantages of Dynamic Ontologies*
1. *Adaptability*: Dynamic ontologies can adapt to changing knowledge, new discoveries, or updates in a domain, ensuring the ontology remains relevant and accurate.
2. *Improved accuracy*: By continuously optimizing themselves, dynamic ontologies can reduce errors, inconsistencies, and ambiguities, leading to more reliable and trustworthy knowledge representations.
3. *Enhanced scalability*: Dynamic ontologies can handle increasing amounts of data and complexity, making them more suitable for large-scale applications and big data integration.
4. *Collaboration and community engagement*: Dynamic ontologies can facilitate community involvement, allowing experts to contribute to the ontology's development, refinement, and maintenance.
5. *Reduced maintenance costs*: By automating optimization processes, dynamic ontologies can reduce the need for manual updates, revisions, and maintenance, resulting in cost savings.
*Limitations of Static Ontologies*
1. *Rigidity*: Static ontologies are often fixed and inflexible, making it difficult to incorporate new knowledge, updates, or changes.
2. *Copyright restrictions*: As you mentioned, static ontologies may be copyrighted, limiting their use, modification, and distribution.
3. *Knowledge stagnation*: Static ontologies can become outdated, leading to knowledge stagnation and reduced relevance in rapidly evolving domains.
4. *Limited scalability*: Static ontologies may struggle to handle increasing data volumes, complexity, or diversity, making them less suitable for large-scale applications.
*Real-World Applications of Dynamic Ontologies*
1. *Artificial intelligence and machine learning*: Dynamic ontologies can support AI and ML applications by providing adaptive, context-aware knowledge representations.
2. *Data integration and interoperability*: Dynamic ontologies can facilitate data integration across disparate sources, domains, and formats.
3. *Knowledge graphs and semantic search*: Dynamic ontologies can power knowledge graphs, enabling more accurate and relevant search results.
4. *Healthcare and biomedical research*: Dynamic ontologies can support personalized medicine, disease modeling, and drug discovery by integrating and analyzing large amounts of biomedical data.
In summary, dynamic ontologies offer significant advantages over static ontologies, including adaptability, improved accuracy, scalability, collaboration, and reduced maintenance costs. As the volume and complexity of data continue to grow, dynamic ontologies will play an increasingly important role in supporting AI, data integration, and knowledge management applications.
[01/12, 17:49] rb: [01/12, 17:47]: How are dynamic ontologies created different from static ontologies?
[01/12, 17:47] Meta AI: Dynamic ontologies are created using a different approach than static ontologies. Here are the key differences:
_Static Ontology Creation_
1. _Manual development_: Static ontologies are typically created manually by experts in the domain.
2. _Top-down approach_: Static ontologies are often developed using a top-down approach, where the overall structure and concepts are defined first.
3. _Fixed schema_: Static ontologies have a fixed schema, which is defined at design time and remains unchanged.
4. _Limited scalability_: Static ontologies can become rigid and difficult to modify as the domain evolves.
_Dynamic Ontology Creation_
1. _Hybrid approach_: Dynamic ontologies are created using a hybrid approach, combining manual and automated methods.
2. _Bottom-up approach_: Dynamic ontologies are often developed using a bottom-up approach, where the ontology is built from smaller, modular components.
3. _Evolutionary design_: Dynamic ontologies are designed to evolve over time, incorporating new knowledge, updates, and changes.
4. _Machine learning and AI integration_: Dynamic ontologies often leverage machine learning and AI techniques to automate updates, refinements, and optimizations.
5. _Collaborative development_: Dynamic ontologies are often developed collaboratively, involving multiple stakeholders, experts, and communities.
Some popular methods for creating dynamic ontologies include:
1. _Ontology learning_: Automatically extracting ontology components from data, text, or other sources.
2. _Ontology evolution_: Incrementally updating and refining an existing ontology to reflect changes in the domain.
3. _Ontology merging_: Combining multiple ontologies to create a new, more comprehensive ontology.
4. _Ontology alignment_: Mapping and aligning concepts between different ontologies to enable interoperability.
By using these methods and approaches, dynamic ontologies can be created that are more adaptable, scalable, and effective in supporting complex, dynamic applications.
[01/12, 18:08] rb : Below is perhaps an example of a dynamic ontology 👇
[01/12, 18:43] AC: The "process" is dynamic... Not the ontology
Let's hear Meta on that
[01/12, 18:53] AC: Community-driven ontologies are valuable for incorporating diverse expertise and domain knowledge, but they remain suboptimal due to inherent human dependency. This dependency limits their dynamism and adaptability, crucial for evolving domains.
Human Dependency Challenges
1. Cognitive Bias: Contributors may introduce subjective interpretations, affecting objectivity.
2. Inconsistency: Variations in expertise result in uneven quality.
3. Time-Intensive Updates: Manual processes delay ontology evolution.
4. Limited Scalability: Human effort cannot handle large, rapidly growing datasets.
5. Error-Prone Maintenance: Human revisions are more susceptible to mistakes.
6. Resistance to Change: Established structures may be difficult to adapt due to community inertia.
Why Dynamism is Needed
Dynamic ontologies, in contrast, allow continuous adaptation to changing knowledge, ensuring scalability, accuracy, and relevance. However, achieving this requires reducing human reliance by adopting automated and AI-driven processes.
Alternatives to Human Dependency
1. Ontology Learning: Extract relationships from data using machine learning.
2. Real-Time Updates: Automate adjustments based on live data streams.
3. AI-Driven Conflict Resolution: Detect and resolve inconsistencies algorithmically.
4. Feedback Loops: Implement self-adaptive mechanisms for ongoing refinement.
5. Hybrid Governance: Combine automated systems with minimal human oversight to balance efficiency and control.
Benefits of Reduced Dependency
Faster Adaptation: Real-time responsiveness to changes.
Enhanced Scalability: Handle massive data influx with ease.
Consistency: Uniform quality through algorithmic updates.
Cost Efficiency: Reduced manual effort lowers maintenance costs.
In summary, while community-driven ontologies offer inclusivity and collaboration, transitioning to automated, AI-enhanced frameworks is essential for achieving true dynamism and long-term relevance in fast-evolving domains.
[01/12, 19:13] AC: No, most of these examples (e.g., Gene Ontology, SNOMED CT, Schema.org) are not fully dynamic ontologies as defined in the context of self-optimization and evolution without heavy human intervention. They are better classified as semi-dynamic ontologies because:
1. Gene Ontology (GO): Relies on manual curation alongside some automated tools.
2. SNOMED CT: Updates require expert review and are guided by organizational governance.
3. Schema.org: Community-driven but lacks self-optimization mechanisms.
4. DBpedia: Dynamic in the sense of periodic updates, but not fully autonomous.
5. LOD: A framework for linking data rather than an ontology with self-evolution capabilities.
A truly dynamic ontology would:
Leverage AI/ML for self-updates.
Require minimal human oversight.
Adapt automatically to new data or domain changes.
Hypothetical Examples:
Autonomous Medical Ontology: Learns from global clinical data streams in real time.
Smart Manufacturing Ontology: Evolves based on IoT sensor data and AI insights.
No fully dynamic ontologies yet exist in practice but remain an active area of research.
[01/12, 19:44] RKB AIH: Ontologies are notoriously hard to maintain over the long run..the fuzziness & boundaries get sketchy
[01/12, 19:49] SBB: In the context of SNOMED CT, can you explain what you mean by “fuzziness and boundaries get sketchy”?
[01/12, 19:54] RKB AIH: From my experience in financial & especially dynamic environments like news, as new tech/process or verbiage gets coined/modified definitions change & ontology assignments become wrong over a period of time I.e become too broad or too narrow to encompass real definition/boundary
[01/12, 19:54] AC: Yes. Good old "knowledge management" issue
Lots of research has gone into it...
But... (See the ref msg for the economic basis of suboptimality)
[01/12, 19:55] RKB AIH: Plz correct me if it isn’t the case in medical world
[01/12, 19:57] RKB AIH: Thank you your message covers everything I had in mind 😀
[01/12, 19:58] RKB AIH: Biggest flash point is Expert Vs Croud Sourced…in media it was a nightmare
I don’t think that will be issue in med world as patients are not the direct consumers of ontologies
[01/12, 19:58] AC: Knowledge representation using ANNs is a well established domain
No one has any real or urgent needs. Hence we don't see results
[01/12, 20:00] AC: I tend to favor "informed" crowdsourcing over exPERT o'pinion
[01/12, 20:00] RKB AIH: Explainability was the Achilles heel earlier, but nowadays that not a real concern as new methods are being devised for explainability
[01/12, 20:01] AC: Explainability was never really a challenge... The market rhetoric was so high that informed voices were getting drowned
[01/12, 20:04] AC: Who grants experthood? Like is it a knighting ceremony?
No citations, no "global" recognition... Frankly hollow foundations - but exPERTease is easily claimed
[01/12, 20:08] RKB AIH: Com’on man !! 😀
I am talking “real” experts vs “commoner”…attack/stroke as per common man terminology 😀
[01/12, 20:10] AC: This is a 2009 article. DL was still not common
[01/12, 20:14] AC: This is a fair review of state of affairs around DL for ontology construction
(Now this talks of BERT hence I assert as an informed exBERT ... not as exPERT! 🙂)
[01/12, 20:16] SBB: SNOMED CT concept model (basically the ontology) has evolved since 1955 and is now pretty stable. Yes, some concept definitions do change, the older ones much less frequently as newer understanding necessitates the change. The clinical findings least. Disorders next. Followed by procedures and situations. Observable entities are getting added to as is the case with clinical drugs. The poly hierarchical structure ensures that the position of individual concepts within the hierarchy does not change (most is an inferred ontology generated by running description logic reasoners) from stated definitions (ontologies). Consequently, I would opine that the SNOMED ontology is pretty stable.
[01/12, 20:25] RKB AIH: Wordnet was first widely used English ontology for use with NLP, as far as I know nobody uses it in NLP today…the wider the ontology get the less useful it is & harder to maintain any ROI
[01/12, 20:26] RKB AIH: https://wordnet.princeton.edu
[02/12, 05:14] AC: Not an ontology
You need to delineate various artefacts
1. Corpus Linguistics
2. POS tagged corpora
3. Lexical db
Every lexical db is not an ontology
Ontology is an "ancient" practice (similar in age to Past Life Regression! Hence lots of esoteric statements are unfoundedly made about it)
In the earlier era they served as an encoding dataset. (Ref CFG as defined for SCT. That's where SCT discussions should begin, not on unfounded assumptions. SCT allows *multiple* representations of the same encoding. Will point to some explanatory paper to clear the most common misconception surrounding SCT)
I had earlier pointed out the business basis of suboptimal implementation of ontologies to ensure copyright enforcement
[01/12, 20:27] SBB: There’s *no* equivalence between Wordnet and SNOMED outside of the term “ontology”.
[01/12, 20:28] RKB AIH: That’s true. SNOMED being much narrower is more useful
[01/12, 20:41] SBB: SNOMED ontology is for the machines to interpret clinical terms. It provides a list of standard terms (controlled medical vocabulary) and their concept definitions that are machine interpretable. Interrelated semantic triples. It’s amazing that for NLP in clinical medicine folks don’t use the 750,000 plus terms as their bag of words. They try getting these from the various dictionaries!
No comments:
Post a Comment