Actions

ConferenceCall 2024 03 27: Difference between revisions

Ontolog Forum
 
(5 intermediate revisions by the same user not shown)
Line 21: Line 21:


== Agenda ==
== Agenda ==
* '''[[AmitSheth|Amit Sheth]]''' ''Forging Trust in Tomorrow’s AI: A Roadmap for Reliable, Explainable, and Safe NeuroSymbolic Systems''
* '''Markus J. Buehler''' ''Accelerating Scientific Discovery with Generative Knowledge Extraction, Graph-Based Representation, and Multimodal Intelligent Graph Reasoning''
** In Pedro Dominguez's influential 2012 paper, the phrase "Data alone is not enough" emphasized a crucial point. I've long shared this belief, which is evident in our Semantic Search engine, which was commercialized in 2000 and detailed in a patent. We enhanced machine learning classifiers with a comprehensive WorldModel™, known today as knowledge graphs, to improve named entity, relationship extraction, and semantic search. This early project highlighted the synergy between data-driven statistical learning and knowledge-supported symbolic AI methods, an idea I'll explore further in this talk. <br/> Despite the remarkable success of transformer-based models in numerous NLP tasks, purely data-driven approaches fall short in tasks requiring Natural Language Understanding (NLU). Understanding language - Reasoning over language, generating user-friendly explanations, constraining outputs to prevent unsafe interactions, and enabling decision-centric outcomes necessitates neurosymbolic pipelines that utilize knowledge and data.
** Abstract: For centuries, researchers have sought out ways to connect disparate areas of knowledge. With the advent of Artificial Intelligence (AI), we can now rigorously explore relationships that cut across distinct areas – such as, mechanics and biology, or science and art – to deepen our understanding and to accelerate innovation. To do this we transformed a set of 1,000 scientific papers in the field of biological materials into a detailed ontological knowledge graph. We conduct a detailed analysis of the graph structure and calculate node degrees, communities and community connectivities, as well as clustering coefficients and betweenness centrality of key nodes, and find that the graph has an inherently scale-free nature. Using a large language embedding model we compute deep node representations and use combinatorial node similarity ranking to develop a path sampling strategy that allows us to link dissimilar concepts across the graph that have previously not been related. We apply this method to reveal insights into unprecedented interdisciplinary relationships that can be used to answer queries, identify gaps in knowledge, propose never-before-seen material designs, and predict material behaviors. One comparison revealed detailed structural parallels between biological materials and Beethoven's 9th Symphony, highlighting shared patterns of complexity through isomorphic mapping. In another example, the algorithm proposed an innovative hierarchical mycelium-based composite based on a joint synthesis of path sampling with principles extracted from Kandinsky's `Composition VII' painting, where the resulting composite features balance of chaos and order, adjustable porosity, mechanical strength, and complex patterned chemical functionalization. We uncover other isomorphisms across science, technology and art, revealing a nuanced ontology of immanence that reveal a dynamic, context-dependent heterarchical interplay of entities beyond traditional hierarchical paradigms. Because our method transcends traditional disciplinary boundaries, and because it integrates diverse data modalities (graphs, images, text, numerical data, etc.) we achieve a far higher degree of novelty, explorative capacity, and technical detail, than conventional generative AI. The approach establishes a widely useful framework for innovation, drawing from diverse fields such as materials science, logic, art, and music, by revealing hidden connections that facilitate discovery.  
** Problem: Inadequacy of LLMs for Reasoning<br/>LLMs like GPT-4, while impressive in their abilities to understand and generate human-like text, have limitations in reasoning. They excel at pattern recognition, language processing, and generating coherent text based on input. However, their reasoning capabilities are limited by their need for true understanding or awareness of concepts, contexts, or causal relationships beyond the statistical patterns in the data they were trained on. While they can perform certain types of reasoning tasks, such as simple logical deductions or basic arithmetic, they often need help with more complex forms of reasoning that require deeper understanding, context awareness, or commonsense knowledge. They may produce responses that appear rational on the surface but lack genuine comprehension or logical consistency. Furthermore, their reasoning does not adapt well to the dynamicity of the environment, i.e., the changing environment in which the AI model is operating (e.g., changing data and knowledge).
** Bio: Markus J. Buehler is the McAfee Professor of Engineering at MIT. Professor Buehler pursues new modeling, design and manufacturing approaches for advanced biomaterials that offer greater resilience and a wide range of controllable properties from the nano- to the macroscale. He received many distinguished awards, including the Feynman Prize, the ASME Drucker Medal, the J.R. Rice Medal, and many others. Buehler is a member of the National Academy of Engineering.
** Solution: Neurosymbolic AI combined with Custom and Compact Models:<br/>Compact custom language models can be augmented with neurosymbolic methods and external knowledge sources while maintaining a small size. The intent is to support efficient adaptation to changing data and knowledge. By integrating neurosymbolic approaches, these models acquire a structured understanding of data, enhancing interpretability and reliability (e.g., through verifiability audits using reasoning traces). This structured understanding fosters safer and more consistent behavior and facilitates efficient adaptation to evolving information, ensuring agility in handling dynamic environments. Furthermore, incorporating external knowledge sources enriches the model's understanding and adaptability across diverse domains, bolstering its efficiency in tackling varied tasks. The small size of these models enables rapid deployment and contributes to computational efficiency, better management of constraints, and faster re-training/fine-tuning/inference.  
** About the Speaker: Professor Amit Sheth (Web, LinkedIn) is an Educator, Researcher, and Entrepreneur. As the founding director of the university-wide AI Institute at the University of South Carolina, he grew it to nearly 50 AI researchers. He is a fellow of IEEE, AAAI, AAAS, ACM, and AIAA. He has co-founded four companies, including Taalee/Semangix which pioneered Semantic Search (founded 1999), ezDI, which supported knowledge-infused clinical NLP/NLU, andCognovi Labs, and emotion AI company. He is proud of the success of over 45 Ph.D. advisees and postdocs he hs advised/mentored.


== Conference Call Information ==
== Conference Call Information ==
Line 36: Line 34:
{{:OntologySummit2024/ConferenceCallInformation}}
{{:OntologySummit2024/ConferenceCallInformation}}


== Participants ==
* [[BevCorwin|Bev Corwin]]
* [[BruceBray|Bruce Bray]]
* [[DouglasMiles|Douglas Miles]]
* E. W. Schuster
* [[GaryBergCross|Gary Berg-Cross]]
* [[JanetSinger|Janet Singer]]
* [[JohnSowa|John Sowa]]
* Josefina
* [[KenBaclawski|Ken Baclawski]]
* [[MarkUnderwood|Mark Underwood]]
* [[MarkusBuehler|Markus J. Buehler]]
* [[MichaelDeBellis|Michael DeBellis]]
* Phil Jackson
* [[RaviSharma|Ravi Sharma]]


== Discussion ==
== Discussion ==
=== Precision of Language ===
[12:25] John Sowa: There is a huge difference between logic and mush.
* In a mush system, which combines a huge amount of small lumps with probabilities, no single lump can make more than a tiny change to the total.
* But in logic, a single symbol for NOT can reverse the total.
* Language can be precise or mushy.
* It all depends on the problem, the source of information, and the sensitivity to a single word, such as NOT.
* Visual languages, such as diagrams, can be precise or mushy -- in the same way as linear languages.
* The only difference between linear languages and diagrammatic languages is the dimension.
* Mushy or precise depends on the nature of the knowledge source and the way of combining parts.
* Re difference between logic and language:  It all depends on the subject matter.
* If the source is precise, logic is precise.  Language can be precise, but it's easier to slip into a mushy mode.
=== Logic as a Natural Language ===
[12:31] Douglas Miles: Somewhere I read that Bertrand Russell finally confessed that Logic was merely yet another natural language
[12:32] John Sowa: Re Bertie:  He had some good ideas, but he wasn't as smart as Whitehead.
[12:34] Michael DeBellis: Replying to "Somewhere I read tha..."  Where did Russell say that "logic was merely yet another natural language"? I'm skeptical. If he did say it he was wrong. Natural language and logical languages are fundamentally different. Logic is a context free language. A formula's syntax is independent of any WFFs around it in logic. This is not the case in natural language. In natural language syntax can be context sensitive and the syntax of a sentence can be ambiguous. That's not the case with logic. A WFF is either a WFF or it isn't and if it is the syntax is unambiguous.
[12:36] Douglas Miles: re Berrtie: exactly, I interpreted that to mean logic is designed to be as non-precise as needed to be usable
[12:36] Douglas Miles: non-precise as required to not contradict itself
[12:37] Douglas Miles: or anything else we want to imagine it was supposed to do
[12:39] John Sowa: Replying to "Somewhere I read tha..." I don't know where, whether, or how Russell compared language to logic.  But you can express a huge amount of language in a context-free syntax.
[12:40] Michael DeBellis: Replying to "non-precise as requi..." I don't agree that logic needs to be non-precise to not have contradictions. Logic is precise and you can have very complex logical systems that don't have contradictions. That's why ZFC set theory is the foundation for conventional math. Because it has the minimum number of axioms (as far as we currently know) that can support mathematics but doesn't have contradictions the way Frege's system did and it doesn't need the huge complex rules that Russel and Whitehead's logic did.
[12:40] John Sowa: Replying to "Somewhere I read tha..." My favorite comment is by Whitehead:  "Bertie thinks that I am muddle-headed.  But I think that he is simple-minded."
[12:42] John Sowa: Replying to "Somewhere I read tha..." Whitehead made that comment when he was introducing Russell for a set of lectures at Harvard in the 1930s.
[12:45] Michael DeBellis: Replying to "Somewhere I read tha..." Actually, I would bet if you look at any book there are ambiguous sentences on every page where the syntax is only obvious due to context. We often take these things for granted because we just do it unconsciously. You can also have very simple Natural Language sentences where the syntax is ambiguous: "I saw the man on the hill with the telescope" Who has the telescope? The speaker or the man on the hill? The syntax of that simple sentence has at least two different interpretations. You don't get this kind of ambiguity in logic.
[12:46] John Sowa: Logic cannot be vague, but it is very easy to make mistakes that are horribly bad.
[12:47] John Sowa: It is easy to be vague in ordinary language, and sometimes vagueness is good enough.
[12:48] John Sowa: But in logic, vagueness is not possible.  Instead, it easy to be false.
[12:50] Douglas Miles: Replying to "Somewhere I read t..." It was in what is often called his finally (video taped) Interview
[12:51] Douglas Miles: Replying to "Somewhere I read t..." "Final (video taped) Interview"
[13:06] Michael DeBellis: Replying to "Somewhere I read tha..." One question I have is how his work relates to ideas regarding emergent behavior? E.g, Krakauer?
=== Multi-Agent Design Patterns ===
[12:38] Ravi Sharma: you [Markus Buehler] are practicing a dynamic visual language
[12:41] Bev Corwin: Reacted to "you are practicing a..." with 👍
[12:41] Mark Underwood: Very happy to see multi-agent design patterns returning to practice.  If only it was better standardized . . .
[12:42] Ravi Sharma: for new designs we not only use Visual dynamics but also in reaching new designs based on generative AI
[12:43] Ravi Sharma: but can you integrate AI and Physics together?
[12:48] Ravi Sharma: Simulate to select the actual material development route.
[12:49] Ravi Sharma: Your work can also help Prosthetics and other bio integrated medical devices.
=== What does an AI system know? ===
[12:26] Ravi Sharma: model learns about itself is similar to self healing materials?
[12:27] Douglas Miles: can non "natural language" structures such as logic be enough to teach a model to reason?
[12:48] Phil Jackson: Does the model know what it does not know, and needs to better understand?
[12:51] John Sowa: There are many kinds of ways to use the word 'know' about AI systems.  But some systems, such as LLM-based systems, may create an illusion of knowing.
[12:52] John Sowa: It's better to avoid words like 'know' and 'understand' unless you have a very deep understanding of what your AI system is doing.
[12:53] John Sowa: Today, I would not apply the words 'know' or 'understand' to any current AI system.
[12:56] Phil Jackson: Understood. I still wonder whether Markus' system has any knowledge or understanding of its limitations.
=== Standard Tests ===
[12:33] Gary Berg-Cross: I wonder if there is a standard test set to see how an AI hybrid system provides useful material models integrating the various levels of granularity.
[12:35] Gary Berg-Cross: A related q might be how to demonstrate that the AI system is learning how to reason about these material issues.
=== Summary Comments ===
[12:11] Ravi Sharma: Philosophically Nature is elusive and knowledge is evolutionary
* once we understand local and long range we can understand better how wealth of chemistry and biology -variety can be understood
* graphs glyphs and images are innate to understanding
* language of visual images and graphs needs to be developed but will it be like math or programming language or yet a new paradigm?
* visual language and thought integration will have to be dynamic both in space and time?
* bio development is based on vailability of elements, similarly will other forms evolve differently in exobiology?
* Markus you are really addressing a glaring gap in visual languages.
* therefore bio samples have to be varied as well as subject to determination? How can balance be achieved?
* But you showed only one mode or all modes of protein movement
=== Compliments ===
[12:56] Josefina: Amazing course!
[13:05] Bruce Bray: excellent discussion!
[13:06] Bev Corwin: Thank you
[13:06] janet singer: Great material, great presentation.
[13:06] E. W. Schuster: Thanks Markus.
[13:06] Markus Buehler: Thanks everyone!
[13:07] E. W. Schuster: Reacted to "Thanks everyone!" with 👍


== Resources ==
== Resources ==
* [https://bit.ly/3IYqiyt Video Recording]
* [https://bit.ly/3TBJcQD Video Recording with Closed Captions]
* [[ConferenceCall_2024_03_27/Transcript|Session Transcript]]
* [https://youtu.be/p_cF35YA0rg YouTube Video]


== Previous Meetings ==
== Previous Meetings ==

Latest revision as of 17:42, 1 April 2024

Session Foundations and Architectures
Duration 1 hour
Date/Time 27 Mar 2024 16:00 GMT
9:00am PDT/12:00pm EDT
4:00pm GMT/5:00pm CST
Convener Ravi Sharma

Ontology Summit 2024 Foundations and Architectures

Agenda

  • Markus J. Buehler Accelerating Scientific Discovery with Generative Knowledge Extraction, Graph-Based Representation, and Multimodal Intelligent Graph Reasoning
    • Abstract: For centuries, researchers have sought out ways to connect disparate areas of knowledge. With the advent of Artificial Intelligence (AI), we can now rigorously explore relationships that cut across distinct areas – such as, mechanics and biology, or science and art – to deepen our understanding and to accelerate innovation. To do this we transformed a set of 1,000 scientific papers in the field of biological materials into a detailed ontological knowledge graph. We conduct a detailed analysis of the graph structure and calculate node degrees, communities and community connectivities, as well as clustering coefficients and betweenness centrality of key nodes, and find that the graph has an inherently scale-free nature. Using a large language embedding model we compute deep node representations and use combinatorial node similarity ranking to develop a path sampling strategy that allows us to link dissimilar concepts across the graph that have previously not been related. We apply this method to reveal insights into unprecedented interdisciplinary relationships that can be used to answer queries, identify gaps in knowledge, propose never-before-seen material designs, and predict material behaviors. One comparison revealed detailed structural parallels between biological materials and Beethoven's 9th Symphony, highlighting shared patterns of complexity through isomorphic mapping. In another example, the algorithm proposed an innovative hierarchical mycelium-based composite based on a joint synthesis of path sampling with principles extracted from Kandinsky's `Composition VII' painting, where the resulting composite features balance of chaos and order, adjustable porosity, mechanical strength, and complex patterned chemical functionalization. We uncover other isomorphisms across science, technology and art, revealing a nuanced ontology of immanence that reveal a dynamic, context-dependent heterarchical interplay of entities beyond traditional hierarchical paradigms. Because our method transcends traditional disciplinary boundaries, and because it integrates diverse data modalities (graphs, images, text, numerical data, etc.) we achieve a far higher degree of novelty, explorative capacity, and technical detail, than conventional generative AI. The approach establishes a widely useful framework for innovation, drawing from diverse fields such as materials science, logic, art, and music, by revealing hidden connections that facilitate discovery.
    • Bio: Markus J. Buehler is the McAfee Professor of Engineering at MIT. Professor Buehler pursues new modeling, design and manufacturing approaches for advanced biomaterials that offer greater resilience and a wide range of controllable properties from the nano- to the macroscale. He received many distinguished awards, including the Feynman Prize, the ASME Drucker Medal, the J.R. Rice Medal, and many others. Buehler is a member of the National Academy of Engineering.

Conference Call Information

  • Date: Wednesday, 27 March 2024
  • Start Time: 9:00am PDT / 12:00pm EDT / 5:00pm CET / 4:00pm GMT / 1600 UTC
    • ref: World Clock
    • Note: The US and Canada are on Daylight Saving Time while Europe has not yet changed.
  • Expected Call Duration: 1 hour

The unabbreviated URL is: https://us02web.zoom.us/j/87630453240?pwd=YVYvZHRpelVqSkM5QlJ4aGJrbmZzQT09

Discussion

Precision of Language

[12:25] John Sowa: There is a huge difference between logic and mush.

  • In a mush system, which combines a huge amount of small lumps with probabilities, no single lump can make more than a tiny change to the total.
  • But in logic, a single symbol for NOT can reverse the total.
  • Language can be precise or mushy.
  • It all depends on the problem, the source of information, and the sensitivity to a single word, such as NOT.
  • Visual languages, such as diagrams, can be precise or mushy -- in the same way as linear languages.
  • The only difference between linear languages and diagrammatic languages is the dimension.
  • Mushy or precise depends on the nature of the knowledge source and the way of combining parts.
  • Re difference between logic and language: It all depends on the subject matter.
  • If the source is precise, logic is precise. Language can be precise, but it's easier to slip into a mushy mode.

Logic as a Natural Language

[12:31] Douglas Miles: Somewhere I read that Bertrand Russell finally confessed that Logic was merely yet another natural language

[12:32] John Sowa: Re Bertie: He had some good ideas, but he wasn't as smart as Whitehead.

[12:34] Michael DeBellis: Replying to "Somewhere I read tha..." Where did Russell say that "logic was merely yet another natural language"? I'm skeptical. If he did say it he was wrong. Natural language and logical languages are fundamentally different. Logic is a context free language. A formula's syntax is independent of any WFFs around it in logic. This is not the case in natural language. In natural language syntax can be context sensitive and the syntax of a sentence can be ambiguous. That's not the case with logic. A WFF is either a WFF or it isn't and if it is the syntax is unambiguous.

[12:36] Douglas Miles: re Berrtie: exactly, I interpreted that to mean logic is designed to be as non-precise as needed to be usable

[12:36] Douglas Miles: non-precise as required to not contradict itself

[12:37] Douglas Miles: or anything else we want to imagine it was supposed to do

[12:39] John Sowa: Replying to "Somewhere I read tha..." I don't know where, whether, or how Russell compared language to logic. But you can express a huge amount of language in a context-free syntax.

[12:40] Michael DeBellis: Replying to "non-precise as requi..." I don't agree that logic needs to be non-precise to not have contradictions. Logic is precise and you can have very complex logical systems that don't have contradictions. That's why ZFC set theory is the foundation for conventional math. Because it has the minimum number of axioms (as far as we currently know) that can support mathematics but doesn't have contradictions the way Frege's system did and it doesn't need the huge complex rules that Russel and Whitehead's logic did.

[12:40] John Sowa: Replying to "Somewhere I read tha..." My favorite comment is by Whitehead: "Bertie thinks that I am muddle-headed. But I think that he is simple-minded."

[12:42] John Sowa: Replying to "Somewhere I read tha..." Whitehead made that comment when he was introducing Russell for a set of lectures at Harvard in the 1930s.

[12:45] Michael DeBellis: Replying to "Somewhere I read tha..." Actually, I would bet if you look at any book there are ambiguous sentences on every page where the syntax is only obvious due to context. We often take these things for granted because we just do it unconsciously. You can also have very simple Natural Language sentences where the syntax is ambiguous: "I saw the man on the hill with the telescope" Who has the telescope? The speaker or the man on the hill? The syntax of that simple sentence has at least two different interpretations. You don't get this kind of ambiguity in logic.

[12:46] John Sowa: Logic cannot be vague, but it is very easy to make mistakes that are horribly bad.

[12:47] John Sowa: It is easy to be vague in ordinary language, and sometimes vagueness is good enough.

[12:48] John Sowa: But in logic, vagueness is not possible. Instead, it easy to be false.

[12:50] Douglas Miles: Replying to "Somewhere I read t..." It was in what is often called his finally (video taped) Interview

[12:51] Douglas Miles: Replying to "Somewhere I read t..." "Final (video taped) Interview"

[13:06] Michael DeBellis: Replying to "Somewhere I read tha..." One question I have is how his work relates to ideas regarding emergent behavior? E.g, Krakauer?

Multi-Agent Design Patterns

[12:38] Ravi Sharma: you [Markus Buehler] are practicing a dynamic visual language

[12:41] Bev Corwin: Reacted to "you are practicing a..." with 👍

[12:41] Mark Underwood: Very happy to see multi-agent design patterns returning to practice. If only it was better standardized . . .

[12:42] Ravi Sharma: for new designs we not only use Visual dynamics but also in reaching new designs based on generative AI

[12:43] Ravi Sharma: but can you integrate AI and Physics together?

[12:48] Ravi Sharma: Simulate to select the actual material development route.

[12:49] Ravi Sharma: Your work can also help Prosthetics and other bio integrated medical devices.

What does an AI system know?

[12:26] Ravi Sharma: model learns about itself is similar to self healing materials?

[12:27] Douglas Miles: can non "natural language" structures such as logic be enough to teach a model to reason?

[12:48] Phil Jackson: Does the model know what it does not know, and needs to better understand?

[12:51] John Sowa: There are many kinds of ways to use the word 'know' about AI systems. But some systems, such as LLM-based systems, may create an illusion of knowing.

[12:52] John Sowa: It's better to avoid words like 'know' and 'understand' unless you have a very deep understanding of what your AI system is doing.

[12:53] John Sowa: Today, I would not apply the words 'know' or 'understand' to any current AI system.

[12:56] Phil Jackson: Understood. I still wonder whether Markus' system has any knowledge or understanding of its limitations.

Standard Tests

[12:33] Gary Berg-Cross: I wonder if there is a standard test set to see how an AI hybrid system provides useful material models integrating the various levels of granularity.

[12:35] Gary Berg-Cross: A related q might be how to demonstrate that the AI system is learning how to reason about these material issues.

Summary Comments

[12:11] Ravi Sharma: Philosophically Nature is elusive and knowledge is evolutionary

  • once we understand local and long range we can understand better how wealth of chemistry and biology -variety can be understood
  • graphs glyphs and images are innate to understanding
  • language of visual images and graphs needs to be developed but will it be like math or programming language or yet a new paradigm?
  • visual language and thought integration will have to be dynamic both in space and time?
  • bio development is based on vailability of elements, similarly will other forms evolve differently in exobiology?
  • Markus you are really addressing a glaring gap in visual languages.
  • therefore bio samples have to be varied as well as subject to determination? How can balance be achieved?
  • But you showed only one mode or all modes of protein movement

Compliments

[12:56] Josefina: Amazing course!

[13:05] Bruce Bray: excellent discussion!

[13:06] Bev Corwin: Thank you

[13:06] janet singer: Great material, great presentation.

[13:06] E. W. Schuster: Thanks Markus.

[13:06] Markus Buehler: Thanks everyone!

[13:07] E. W. Schuster: Reacted to "Thanks everyone!" with 👍

Resources

Previous Meetings

 Session
ConferenceCall 2024 03 20Foundations and Architectures
ConferenceCall 2024 03 13LLMs, Ontologies and KGs
ConferenceCall 2024 03 06LLMs, Ontologies and KGs
... further results

Next Meetings

 Session
ConferenceCall 2024 04 03Synthesis
ConferenceCall 2024 04 10Synthesis
ConferenceCall 2024 04 17Applications
... further results
Retrieved from "https://ontologforum.com/index.php?title=ConferenceCall_2024_03_27&oldid=4969"