The Journey of Expert Systems

In the 1970s, skepticism arose among AI researchers regarding the ability of computational systems to gain a comprehensive understanding of various subjects. As a result, they opted to focus on narrowly defined problems. By supplying algorithms with specialized knowledge, they aimed to surpass human experts in decision-making.

These so-called computer-based expert systems comprised three main components:

1. A knowledge base filled with "if-then" rules relevant to the topic (e.g., if an animal has feathers, it is classified as a bird).
2. A working memory system that retains current assertions and facts (e.g., this animal has fur).
3. An inference mechanism that assesses cases using various rules and assigns differing levels of importance (e.g., this animal lacks feathers, hence it is not a bird).

With expert algorithms developed containing tens of thousands of rules, researchers were pleasantly surprised by the outcomes. In certain instances, these systems nearly matched human experts in their evaluations. For instance, a team from Stanford University created an expert system named MYCIN, designed to accurately diagnose blood disorders. This system was informed by the knowledge of seasoned doctors, and as cases grew more intricate, it integrated probability factors to gauge the uncertainty of its conclusions.

Thanks to this advanced database, MYCIN was capable of providing diagnoses comparable to, or even superior to, those of human physicians in similar cases. This marked the onset of a series of other expert systems tackling similarly defined subjects. However, despite the impressive claims, these systems often revealed genuine limitations, particularly when addressing complex topics.

The Allure of a Universal Knowledge Base

Encouraged by the achievements of symbolic-based AI, researchers in the 1980s proposed increasingly ambitious initiatives. One notable project was the Cyc Project, conceptualized by Douglas Lenat, which sought to create a computing system that encapsulated sufficient knowledge to comprehend our world.

To realize such a vision, Lenat and his team aimed to articulate every explicit rule that governs our understanding of the environment (such as the fact that a rock thrown into water sinks). However, before establishing these rules, the researchers first had to teach the system fundamental concepts necessary for reasoning (like defining what constitutes an object and its relations to other objects). This foundational work involved what they termed ontology engineering, essential for fostering a shared understanding.

Over a decade, the team painstakingly developed an extensive knowledge map. Intrigued by their endeavor, Vaughan Pratt, a computer science pioneer, assessed the system's intelligence through cognitive exercises. In one test, which required analyzing an image of a person relaxing, Cyc responded by presenting an image of individuals carrying surfboards, linking the concept of surfing to a beach setting. This demonstrated effective reasoning; however, Pratt also noted that the system relied on some unnecessary logical deductions (e.g., that humans possess two feet).

Cyc exhibited further shortcomings, particularly when addressing more intricate general knowledge inquiries. For example, it struggled to provide a sensible response to whether bread qualifies as a beverage. Similarly, while Cyc was aware of various causes of death, it failed to recognize starvation as one of them. The demonstration concluded with a rather pessimistic note: Cyc appeared to falter due to gaps in its knowledge, undermining its overall coherence.

Nonetheless, Douglas Lenat persisted with the project, continuously exploring innovative methods to construct a knowledge base. As knowledge systems gain traction, they are now discovering exciting new applications.

Utilizing Knowledge Graphs for Enhanced Data Insights

Since the 1980s experiments, intelligent knowledge systems have found substantial commercial applications. One of the most notable examples is how Google has integrated knowledge graphs into its services. In 2012, the company launched a system capable of understanding knowledge structures on the Internet, significantly enhancing its search engine's capabilities. Based on user intent, Google's algorithm can connect a word to a concept, deciphering the various meanings behind a single term.

For instance, a user searching for "Taj Mahal" could be looking for information about the Indian monument, updates on the artist with the same name, or directions to the Indian restaurant nearby. The algorithm's response hinges on the user's intent.

To provide accurate answers, Google is developing a knowledge system capable of comprehending information more profoundly than just matching keywords. It can extract overarching principles from text and logically deduce facts. For example, when searching for "the green monster in Star Wars," Google can infer that the user is referring to Jabba the Hutt based on links that describe it as "a large, slug-like alien." Thus, it can effectively associate the concept of "alien" with the query "green monster," yielding a precise result.

In addition to search models, knowledge graphs are increasingly employed to enhance other AI models. In the healthcare sector, research organizations amass vast quantities of unstructured data, often missing opportunities for valuable insights into drug development and treatment strategies. By automating the annotation process and leveraging human insights, natural language processing models can extract meaningful information from research data, drawing connections between diseases and treatments. Built upon meticulously organized knowledge, these models can accurately link health conditions to scientific studies, offering well-founded and transparent reasoning.

Such models are also gaining traction in investment and business development decisions. Large corporations seek market intelligence that transcends mere data scraping; they require insights relevant to their specific business challenges. Moreover, they aim to monitor news and commentary regarding their companies to evaluate their reputations. Through knowledge graphs, advanced data processing software can unveil new insights and provide comprehensive feedback about the organization.

In essence, the synergy between human intelligence and data analytics has never been as pronounced, marking the beginning of a promising dialogue between humans and machines.

Knowledge Graphs Created through Basic Machine Learning

Knowledge graphs play a crucial role in understanding and processing information. Learn more about their creation and impact on AI in this insightful video.

Knowledge Graphs: Unlocking the Power of AI

Discover how knowledge graphs unlock the true potential of AI, enabling more nuanced understanding and applications.