Human In The Loop
We stand at a watershed moment in technological history—a junction where the AI algorithms we craft today will define the world we have to live with tomorrow. It's high time we think about how we embed humanity into this equation.
In the realm of mathematics, data can be framed in two contrasting narratives: as a ceaseless flow or as a series of discrete units. Fascinatingly, this duality is reflected in state-of-the-art knowledge representation techniques. Large Language Models (LLMs) view knowledge as a fluid continuum, encoded as continuous embedding vectors. Conversely, Knowledge Graphs compartmentalize information into discrete units using Uniform Resource Identifiers (URIs).
Yet, the human intellect employs both forms of reasoning—oscillating between intuitive, rapid-fire cognition and methodical, analytical thought. Emerging technologies like graph-based Retriever-Augmented Generation (RAG) models combine the best of both worlds, marrying Knowledge Graphs and LLMs into a hybrid intelligence system capable of both agile and calculated thinking.
But let's not fool ourselves—this technological marvel isn't a panacea. To truly realize the promise of AI, we need to introduce a third, fundamentally human element into this equation. We need to inject humans at the point where information passes from a continuous to a discrete representation. Here are two ways to do that:
🔴 Human-Engineered Ontologies: These are structured conceptual models that allow humans to impose their cognitive patterns on machines. These ontologies can guide the LLMs and place guardrails around them. In turn, the LLMs can expand the ontologies, but crucially, any such augmentations can undergo human scrutiny before making it into production.
🔴 Human-Guided Data Curation: Knowledge Graphs serve as intricate webs of information where every node is a data point, and every edge is a semantic relationship. Humans can populate these graphs with curated data, offering invaluable context for LLM-driven queries. Moreover, any machine-generated extensions to this data can again be vetted by human experts.
To construct an AI future that satisfies our true needs—not just our whimsical wants—we need a mechanism for embedding human values and concerns. "Likes" and "follows" might be suitable metrics for social media engagement, but they're woefully inadequate for capturing the nuances of what humans really need.
Knowledge Graphs and Ontologies provide the sophisticated tools required to articulate our complex desires and ethical frameworks. They offer a robust mechanism for anchoring humanity at the critical juncture where fluid information crystallizes into discrete units. By weaving thoughtful and considered human insight into this AI feedback loop, we're not just making incremental improvements; we're laying the architectural groundwork for a more enlightened form of artificial intelligence.
