The Old and the New - Using Semantic Web Technologies to Build Better AI

Introduction

As artificial intelligence (AI) continues to advance at a breakneck pace, particularly with the rise of large language models (LLMs), it is imperative to ensure that these systems are truthful, trustworthy, accountable, and robust. Without such properties, current LLM-style AI systems will struggle to move beyond their roles as sophisticated search engines and writing assistants, hindering their ability to autonomously perform tasks such as acting as web services or agents.

Enter the Semantic Web, a long-standing vision of a machine-readable web that enables agents to understand and interact with data in a meaningful way (Berners-Lee et al., 2001). To learn about the original architectural vision of the Semantic Web, see this design issue. In this blog post, we'll explore how Semantic Web technologies can be leveraged to build better AI systems that are more reliable, accountable, and aligned with human values.

Grounding LLMs with Knowledge Graphs

One of the key challenges with LLMs is ensuring the quality and truthfulness of their responses. By grounding LLMs with high-quality data represented in a knowledge graph (KG), we can significantly improve the accuracy and reliability of their outputs. Knowledge graphs, a core component of the Semantic Web, provide a structured and interconnected representation of data, allowing LLMs to reason over and draw insights from this information. Industry is rapidly adopting this approach, with companies like Ontotext and Neo4J offering out-of-the-box solutions for performing Retrieval Augmented Generation (RAG) with Knowledge Graphs. The interest in grounding LLMs using information from Knowledge Graphs extends beyond companies run by Semantic Web enthusiasts. Andrew Ng, a prominent figure in the AI community, has co-developed a course on RAG with Knowledge Graphs. Furthermore, during Langchain's Memory Hackathon, 30% of the participating teams sought to implement knowledge graphs in their architectures, highlighting the growing recognition of their potential in enhancing LLM performance.

Protecting User Privacy and Ensuring Ethical Data Usage

As AI systems increasingly rely on user and enterprise data to function, it's crucial to ensure that this data is used ethically and in compliance with legal governance frameworks. Semantic Web, and satellite technologies such as Solid offer strategies for attaching access controls, and usage agreements to data in order to support responsible data usage.

Towards Trustworthy and Accountable Personal AI Agents

The Semantic Web has long envisioned a future where autonomous AI agents work on behalf of users, assisting them with various tasks and decision-making processes. By leveraging Semantic Web technologies, we can build the groundwork protocols that enable conversational agents to negotiate and transact over the web which contain features such as:

  1. Mechanism for describing the origin and provenance of exchanged data
  2. Mechanisms to determine the "trustworthiness" of data by modelling which sources / individuals / organisations are reliable, and which are not.
  3. Unambiguous description of usage restrictions on exchanged data to protect privacy allowing conversational LLM-agents can become more trustworthy, accountable, and compliant with data protection regulations such as GDPR.
  4. Clear definition of dialogue outcomes that require agreement or transaction
  5. The ability to discover AI representitives of individuals & organisations using Web Identities.

More on this topic soon … in the meantime, here is my article on machine dialogues

Conclusion

The Semantic Web, with its rich history and powerful technologies, holds the key to building better AI systems in the age of LLMs. By grounding LLMs with knowledge graphs, ensuring ethical data usage through explicit policies, and laying the foundation for trustworthy personal AI agents, the Semantic Web complements and enhances the capabilities of modern AI. As we continue to push the boundaries of what's possible with AI, it's essential to look to the past and leverage the insights and technologies developed by the Semantic Web community. By combining the old and the new, we can create AI systems that are not only powerful but also accountable, transparent, and aligned with human values.