MoeX

What is the mOeX project and what are its core research objectives?

The mOeX project at INRIA addresses the evolution of knowledge representations in individuals and populations. It investigates three fundamental research questions:

• How agent populations adapt their knowledge representation to their environment and other populations

• How knowledge must evolve when the environment changes and new populations are encountered

• How agents can preserve knowledge diversity and whether this diversity is beneficial

The project combines knowledge representation and cultural evolution methods to study these questions in a controlled computer science context. Knowledge representation provides formal models, while cultural evolution offers a framework for studying situated evolution .

How does mOeX conceptualize knowledge in its research framework?

mOeX considers knowledge as a culture and studies the properties of adaptation operators applied by populations of agents through two complementary approaches:

This dual approach allows researchers to acquire a precise understanding of knowledge evolution across a wide range of situations, representations, and adaptation operators . By treating knowledge as cultural, the research acknowledges the social, dynamic, and contextual nature of knowledge development.

How does the mOeX project address the methodological challenges in modeling knowledge evolution?

The mOeX project has identified three key differences between adaptive agents and logical agents that present methodological challenges:

• Adaptive agents reason locally while logical agents reason globally

• Logical agents share a fixed vocabulary, preventing heterogeneous knowledge representations

• Adaptive agents focus on general knowledge (unable to remember individual cases), whereas logical agents cannot discard specific evidence

To address these challenges, the project introduced Partial Dynamic Epistemic Logic (ParDEL), based on partial valuation functions and weakly reflexive relations. ParDEL allows:

• Dropping the assumption of shared, fixed vocabularies

• Modeling propositions as true, false, or undefined

• Preserving semantic heterogeneity between agents

• Enabling successful communication through raising awareness modalities

• Allowing agents to discard evidence via forgetting modalities

This methodological innovation has enabled more accurate modeling of cultural knowledge evolution that better reflects real-world knowledge dynamics.

What formal properties have been established for adaptation operators in the mOeX framework?

Through rigorous logical analysis, researchers have established that adaptation operators within the mOeX framework have the following formal properties:

• Correctness: Adaptation operators correctly translate between ontologies

• Completeness: For ARG states consisting of two agents, adaptation operators are complete

• Non-redundancy: Adaptation operators are no longer partially redundant

These properties were proven using the notion of awareness, confirming that Dynamic Epistemic Ontology Logic (DEOL) is insufficient to model cultural knowledge evolution. The formal analysis demonstrates that agents do not need full awareness of vocabularies used by other agents to communicate successfully in the Alignment Repair Game (ARG) . This finding has significant implications for designing knowledge systems that can operate effectively with incomplete information.

What is the Alignment Repair Game (ARG) and how does it contribute to knowledge evolution research?

The Alignment Repair Game (ARG) is an experimental framework developed to apply cultural evolution methodology to knowledge representation. Key characteristics include:

• Agents with different ontologies communicate randomly through interaction games

• The outcome of each interaction determines whether agents adapt their alignments

• Agents evolve the translations between their ontologies through local corrective actions when communication fails

• Learning occurs "on the fly" with immediate repair of mistakes

This methodology enables researchers to study knowledge transmission in populations while monitoring system states (e.g., success rates) to establish convergence to stable states. The ARG offers a controlled environment to observe how knowledge adapts and evolves through actual use rather than through pre-programmed rules .

How does the Class game implement mOeX concepts for broader audiences?

The Class game was developed to teach the concepts of cultural knowledge evolution to broader audiences. This implementation:

• Allows players to experience firsthand how knowledge representations evolve through interaction

• Demonstrates the challenges of communication across different knowledge frameworks

• Illustrates how agents can develop translations between ontologies through experience

• Shows how heterogeneity can be preserved while achieving successful communication

The game serves as both an educational tool and a data collection mechanism, as player interactions help researchers understand how humans naturally approach knowledge alignment problems. The game can be downloaded from https://moex.inria.fr/mediation/class/index.html .

What experimental designs have been implemented to evaluate adaptation operators in mOeX?

The mOeX project employs sophisticated experimental designs to evaluate adaptation operators:

Researchers monitor metrics such as interaction success rates, knowledge representation properties, and convergence stability to establish when and how populations reach stable states . Unlike traditional machine learning approaches, these experiments focus on properties of the resulting knowledge rather than just convergence, examining social optimality, extensibility, adaptability, robustness, privacy-conformity trade-offs, and computational resource requirements .

How do mOeX experiments address the tension between knowledge diversity and successful communication?

The mOeX project explores this fundamental tension through experiments that analyze:

• Translation development: How agents develop translations between ontologies without enforcing a single common language

• Heterogeneity preservation: How cultural and representational differences can be maintained while enabling effective communication

• Autonomy balancing: How agents balance preserving their unique knowledge structures with the need to communicate successfully

Experimental results demonstrate that enforcing a single ontology is neither feasible nor desirable, as it may result in cultural loss and reduced autonomy . Instead, the research shows how agents can preserve their heterogeneity through developing mutual translations, similar to how humans navigate communication across different languages and cultural contexts . This approach provides empirical evidence for designing knowledge systems that respect diversity while enabling effective information exchange.

How was moexipril identified as a potential PDE4 inhibitor?

Moexipril, a well-tolerated ACE inhibitor, was identified as a potential PDE4 inhibitor through a sophisticated computational approach:

• Researchers applied the Similarity Ensemble Approach (SEA) to the 2010 MDL Drug Data Report

• SEA compared moexipril to sets of ligands for each PDE4 subtype according to ChEMBL

• Moexipril was identified as a potential PDE4A, B, C, and D inhibitor with an E-value of 1.71^-11

• The maximum Tanimoto coefficient in ECFP4 fingerprints was 0.35

• Dynamic light scattering confirmed moexipril did not form colloidal aggregates

• Moexipril did not inhibit beta-lactamase at 10 or 100 μM

This represents a methodological advancement in drug repurposing, using in silico methods to identify off-target activity in existing approved drugs by measuring topological similarity between bait molecules and annotated ligand sets .

What experimental methods validated moexipril's PDE4 inhibitory activity?

Researchers employed multiple experimental approaches to validate the computational prediction:

• Biochemical assays: Tested moexipril against three widely expressed PDE4 isoforms (PDE4A4, PDE4B2, and PDE4D5), confirming inhibition of cAMP hydrolysis in the micromolar range

• Isoform specificity analysis: Determined moexipril was most potent against PDE4B2 (IC50 38 μM), with PDE4A4 and PDE4D5 showing 4-fold and 6-fold lower sensitivity respectively

• Molecular docking: Used DOCK3.6 against the co-crystal structure (PDB: 1MKD) of the PDE4D catalytic domain with bound zardaverine

• FRET-based biosensor: Constructed from the nucleotide binding domain of EPAC1 to demonstrate moexipril's potentiation of forskolin's ability to increase intracellular cAMP

• Functional validation: Demonstrated moexipril's ability to induce phosphorylation of Hsp20 by cAMP-dependent protein kinase A

These methodologies collectively provided robust evidence that moexipril is a bona fide PDE4 inhibitor that could serve as a starting point for developing novel PDE4 inhibitors with improved therapeutic profiles .

What structure-activity relationships were identified between moexipril and PDE4 inhibition?

Molecular modeling and docking studies revealed key structure-activity relationships:

These structural insights led researchers to identify two additional compounds (compounds 7 and 8) with the tetrahydroisoquinoline core of moexipril but simplified N-acyl extensions, allowing further exploration of structure-activity relationships . The docking studies utilized the (S)-configured structures to match the absolute configuration at the tetrahydroisoquinoline 3-position of moexipril, providing detailed molecular insights into the binding mechanism .

How does moexipril's dual pharmacological activity inform research strategies for developing improved PDE4 inhibitors?

Moexipril's identification as both an ACE inhibitor and PDE4 inhibitor has significant implications for research strategies:

• Polypharmacology exploitation: The dual activity suggests potential for developing multi-target therapeutic agents that address related pathological processes

• Safety profile leverage: As a well-tolerated drug, moexipril provides a safer starting scaffold than other PDE4 inhibitors that have been hampered by mechanism-associated side effects

• Structural modification guidance: The identified binding mode helps rationalize which portions of the molecule could be modified to enhance PDE4 selectivity while maintaining safety

• Therapeutic window improvement: The research suggests pathways to develop PDE4 inhibitors with reduced side effects, addressing a key challenge that has limited clinical use of existing PDE4 inhibitors like roflumilast

This research exemplifies how computational drug repurposing combined with experimental validation can accelerate drug discovery by identifying new activities in existing approved medications, providing researchers with pre-validated scaffolds that have already demonstrated acceptable safety profiles in humans .

What theoretical hypotheses underpin the mOeX approach to knowledge evolution?

The mOeX approach is founded on two specific theoretical hypotheses:

• Knowledge is socially and progressively built through use, not just formally derived or learned once

• Knowledge must continuously adapt to changing environments and constraints

Based on these hypotheses, mOeX studies populations of agents sharing knowledge through interaction, with precisely specified modalities that may involve direct knowledge exchange, talking, or acting together. When constraints change, agents do not relearn knowledge from scratch but instead apply adaptation operators that consider current knowledge and constraints to evolve knowledge appropriately .

How does mOeX define successful knowledge evolution?

The mOeX project defines successful knowledge evolution not simply by convergence to a common state, but by the properties satisfied by the resulting knowledge. These properties include:

This approach represents a fundamental shift from conventional knowledge representation approaches by focusing on the dynamic properties of knowledge rather than static representation structures .

How does the notion of awareness in ParDEL address the limitations of traditional logical frameworks?

Traditional logical frameworks like Dynamic Epistemic Logic (DEL) have three critical limitations when modeling cultural knowledge evolution:

• They assume agents reason globally rather than locally

• They require agents to share a fixed vocabulary

• They don't allow agents to discard specific evidence in favor of general knowledge

ParDEL addresses these limitations through:

• Partial valuation functions: Allows propositions to be true, false, or undefined

• Weakly reflexive relations: Models awareness without requiring complete knowledge

• Raising awareness modalities: Enables agents to incorporate new vocabulary elements

• Forgetting modalities: Allows agents to discard evidence in favor of general knowledge

This theoretical innovation has enabled more accurate modeling of how agents with heterogeneous knowledge can successfully communicate without sharing fixed vocabularies, providing formal validation of the adaptation operators used in the ARG .

What are the theoretical implications of mOeX's computational cultural knowledge evolution approach?

The mOeX approach to computational cultural knowledge evolution has several profound theoretical implications:

• Knowledge as cultural: Redefines knowledge as culturally situated rather than abstract and universal

• Evolution through use: Establishes that knowledge representations evolve serendipitously through their usage

• Heterogeneity preservation: Demonstrates theoretically that successful communication does not require homogeneous knowledge representations

• Formalized adaptation: Provides formal models for adaptation operators that preserve important knowledge properties

• Beyond convergence: Shifts focus from convergence to a common state to properties of the resulting knowledge

These theoretical implications challenge traditional assumptions in knowledge representation, artificial intelligence, and multi-agent systems. The approach recognizes that real-world knowledge systems are heterogeneous, dynamic, and culturally situated, requiring formal models that can account for these properties while enabling successful communication and knowledge evolution .

What are the primary open challenges in computational cultural knowledge evolution?

Several significant challenges remain in the field of computational cultural knowledge evolution:

• Scaling complexity: Extending models to handle increasingly complex knowledge structures beyond current experimental settings

• Multi-modality integration: Incorporating different interaction modalities (linguistic, visual, action-based) into knowledge evolution models

• Empirical validation: Connecting theoretical models with empirical observations of human knowledge evolution

• Evaluation metrics: Developing standardized ways to measure knowledge quality beyond simple interaction success rates

• Alignment with human cognition: Making computational models more closely reflect how humans actually adapt and evolve knowledge

Addressing these challenges requires interdisciplinary collaboration between computer scientists, cognitive scientists, linguists, and philosophers to develop comprehensive models of knowledge evolution that capture the complexity of real-world knowledge systems .

How might mOeX methodologies be applied to practical knowledge management problems?

The methodologies developed in mOeX research have potential applications in several practical knowledge management domains:

These applications represent bridges between theoretical research and practical knowledge management challenges, demonstrating how mOeX's computational cultural knowledge evolution approach could contribute to solving real-world problems .

How can formal properties of adaptation operators inform the design of more robust knowledge systems?

The formal analysis of adaptation operators provides insights for designing robust knowledge systems:

• Correctness guarantees: Systems can be designed with provable correctness properties to ensure reliable knowledge translation

• Completeness boundaries: Understanding the conditions under which completeness holds helps define operational boundaries for knowledge systems

• Non-redundancy optimization: Eliminating redundancy improves computational efficiency without sacrificing functionality

• Awareness modeling: Incorporating notion of partial awareness allows systems to function with incomplete information

• Forgetting mechanisms: Implementing principled forgetting enables systems to discard irrelevant details while preserving essential knowledge

These formal properties provide theoretical foundations for developing knowledge systems that can adapt to changing environments and interact with other systems without requiring complete redesign or relearning .

What methodological innovations are needed to bridge experimental cultural evolution and formal knowledge representation?

Bridging these fields requires several methodological innovations:

• Enriched logical frameworks: Developing formal systems that can better represent the dynamic, contextual nature of cultural knowledge

• Empirically-grounded formal models: Creating formal models whose parameters and structures are derived from experimental observations

• Scaling methodologies: Techniques to scale from simple laboratory experiments to complex real-world knowledge ecosystems

• Cross-validation approaches: Methods to validate formal models against experimental data and vice versa

• Integrated simulation environments: Platforms that combine formal reasoning with cultural evolution dynamics

The mOeX project has begun this bridging work through ParDEL and the Alignment Repair Game, but further methodological innovations are needed to fully integrate the complementary strengths of experimental cultural evolution and formal knowledge representation . These innovations would enable more comprehensive understanding of how knowledge evolves in both human and artificial systems.