[DSC DACH 24] Why Graph Technology for AI Reasoning - Diogo Braga

In an era where Artificial Intelligence drives significant advancements across industries, the need for explainability, interpretability, and ethical responsibility in AI systems is critical. My talk, "Harnessing Graph Technology for Augmented Intelligence in Responsible AI Reasoning" delves into the unique intersection of Generative AI and Graph Technology. While Large Language Models and Vector-based Retrieval-Augmented Generation systems have made significant progress, they often fall short in explainability, interpretability, and ethical responsibility. Graphs, with their inherent ability to model complex relationships and dependencies, offer a powerful framework for contextual reasoning and deterministic responses. This talk will explore how leveraging Graph Technology can significantly reduce bias and enhance content relevance, all while maintaining privacy. A key focus will be on the Data Governance pipeline, a critical component of effective knowledge management. This includes highlighting the dependencies from Data Integration to Knowledge Graph creation and Vector Search incorporation. Attendees will gain insights into how Graphs address and solve critical issues within this pipeline, enhancing the organization and utility of knowledge assets. By improving the coherence and responsibility of AI outcomes, this approach supports better decision-making and strategic planning. This talk aims to inspire a new perspective on AI development, emphasizing the importance of more transparent, interpretable, and responsible insights when creating AI systems.

1. Why Graph Technology for AI Reasoning Enhancing AI Explainability, Interpretability, and Ethical Responsibility DSC DACH 24 Diogo Ribeiro Braga ebcont.com

2. 2 2

3. Graph Technology 3

4. 4 4

5. What is a Graph? 5 5 Structures used to model relationships between data points. The relationship itself is a first-class component in the database – identified with a unique key .

6. Graph Structure 6 6 By making relationships first-class citizens, graph databases can more efficiently store , retrieve , and traverse data.

7. Graph-shaped Problems 7 7 What use-cases are graph databases best suited for? Use-cases with highly-interconnected data that require real-time processing and high scalability .

8. Real-World Applications of Graph 8 Supply Chain Management Fraud Detection GenAI Knowledge Graph GraphRAG Customer/Product 360 Recommendation System Search Engine Optimization Workforce Management System Digital Twin for Simulation and Optimization

9. Graphs + Generative AI 9

10. The Revolutionary Impact of GenAI - LLMs 10 10 Capabilities Power in Generating Human-like Text: LLMs can produce coherent responses, resembling human-like conversation. How? Through extensive training on textual (public) data , LLMs acquire a deep learning of language patterns . S uccessful applications of LLMs include content generation , chatbots , language translation , and text summarization .

11. The Revolutionary Impact of GenAI - LLMs 11 11 Limitations Privacy Concerns: The ingestion of private data into LLMs – models accessing and/or learning sensitive information (personal or confidential). Bias Lack of Context Understanding Potential for Inaccuracies - "hallucinations" Capabilities Power in Generating Human-like Text: LLMs can produce coherent responses, resembling human-like conversation. How? Through extensive training on textual public data , LLMs acquire a deep learning of language patterns . Successful applications of LLMs include content generation , chatbots , language translation , and text summarization .

12. Retrieval-Augmented Generation 12 12 Capabilities Access to Private Data: Retrieval-Augmented Generation based on Indexing and Vector Search .

13. Limitations of Vector RAG Sy stems 13 13 Capabilities Access to Private Data: Retrieval-Augmented Generation based on Indexing and Vector Search. Limitations Results are predominantly based on textual data, with limited understanding of context and reasoning abilities, leading to: Bias : Reflects biases from training data – “garbage in, garbage out.” Lack of Context Understanding : Struggles with nuanced meanings beyond textual information alone – lack reasoning . Potential for Inaccuracies : Can produce incorrect info or "hallucinations" due to limited reasoning – it understands patterns but not calculations .

14. The Power of Graph Technology 14 14 Graphs Consolidate Knowledge : Graphs: Excel at representing and consolidating heterogeneous and interconnected information in a structured manner , capturing complex relationships and attributes across diverse data types from diverse business domains . 5-Source Knowledge Graph

15. The Power of Graph Technology 15 15 Graphs Consolidate Knowledge : Graphs: Excel at representing and consolidating heterogeneous and interconnected information in a structured manner , capturing complex relationships and attributes across diverse data types from diverse business domains . Representing Relationships: Graphs excel in representing relationships between nodes - documents, concepts, entities - offering a holistic view of the interconnected subjects. Diverse Applications: Graphs streamline information retrieval and facilitate advanced informed decision-making . 5-Source Knowledge Graph

16. Digital Twin of the Business 16 16 5-Source Knowledge Graph

17. Digital Twin of the Business 17 17 A Graph integrates human domain-expertise with enterprise-processes data , becoming invaluable assets allowing businesses to: Preserve Expertise Enhance Decision-Making Optimize Processes Support Innovation 5-Source Knowledge Graph

18. Why Graph + GenAI 18 18 Enhancing Reasoning from Domain-Expertise: 1. Graphs enhance contextual understanding by capturing relationships between entities. Knowledge Management

19. Why Graph + GenAI 19 19 Enhancing Reasoning from Domain-Expertise: 1. Graphs enhance contextual understanding by capturing relationships between entities. 2. Through Graphs, LLMs retrieve relevant information from diverse knowledge sources . Knowledge Management

20. Why Graph + GenAI 20 20 Enhancing Reasoning from Domain-Expertise: 1. Graphs enhance contextual understanding by capturing relationships between entities. 2. Through Graphs, LLMs retrieve relevant information from diverse knowledge sources . 3. Graphs help AI reasoning to generate less-biased responses from domain-experts . Knowledge Management

21. Why Graph + GenAI 21 21 Enhancing Reasoning from Domain-Expertise: 1. Graphs enhance contextual understanding by capturing relationships between entities. 2. Through Graphs, LLMs retrieve relevant information from diverse knowledge sources . 3. Graphs help AI reasoning to generate less-biased responses from domain-experts . 4. The integration with LLMs boosts explainability , interpretability , and ethical responsibility in AI systems (not a black-box), leading to more reliable and responsible decision-making processes. Knowledge Management

22. Integrating Graph with GenAI 22 22

23. 23 23 "How will we build software in the future?" by Jensen Huang, March 18, 2024.