Image by Author | Ideogram
Symbolic AI is one of those areas of Artificial Intelligence (AI) with such an interesting story to be told. Strongly tied to the most remote past in AI history, after a few decades of being largely forgotten due to the emergence of other prominent areas (not to mention emerging applications like ChatGPT and many more), Symbolic AI is now starting to regain momentum. In fact, some experts are hinting at Symbolic AI as one of the future directions in the field. Why is that, what are the advantages and limitations of symbolic AI, and most importantly, what is symbolic AI to begin with? Read on to discover the answers to these questions in this article.
What is Symbolic AI: Origins and History
Symbolic AI refers to the ability of machines to emulate human thinking by manipulating symbols instead of purely numerical data. While names in the middle of last century like Alan Turing and Herbert Simon are considered pioneers in building computers capable of solving problems in a human-like fashion using by modeling and reasoning over symbols, the foundations were laid centuries back, with philosophers like Thomas Hobbes and Rene Descartes hinting at the entire real-world (universe) being “written” by mathematical symbols.
Upon early advances made in the dawn of AI 70 years back, symbolic AI has been largely based on representing knowledge by using rules applied to symbols, with these symbols representing real-world objects or concepts. An example of such rules in the medical diagnosis domain is If fever and rash, then suspect measles.
This representation gave rise to a specific area within symbolic AI called logic-based programming, where based on facts represented by symbols and their truthfulness (facts can be either true or false), inference processes are applied on rules to draw insightful conclusions. Expert systems and decision support systems are clear examples of systems built predicated on these foundations, and they have still largely used until today, in areas like healthcare, finance, autonomous systems, and legal reasoning.
These are some of most salient applications of symbolic AI, through their most notable forms: expert and decision support systems.
Despite the advances and extensions of symbolic AI approaches during subsequent decades throughout the second half of 20th century, one of the most notable being Lofti Zadeh’s fuzzy logic and fuzzy systems, barriers and limitations also started to manifest. Some limitations included the incapacity of symbolic AI systems to learn by themselves, the requirement of an exhaustive knowledge base to fully model the target application domain, and the challenge of dealing with uncertain or ambiguous information (fuzzy systems partly helped navigating this challenge).
The result: little or no progress in this field since the 1990s until today, along with an ever-increasing predominance of other AI areas that began to advance much more rapidly, like machine learning, deep learning, computer vision, and natural language processing. And so eventually came the current boom led by language models like ChatGPT, and generative AI models capable of creating astonishing content.
Why is Symbolic AI Regaining Attention in the Era of ChatGPT?
Some experts say that despite Symbolic AI’s hibernation for the last 30-40 years, this branch of AI may still awake and have something to say.
Unlike modern data-driven AI systems that require vasts amounts of data to be properly trained, symbolic AI approaches do not require such amounts of data, relying instead on knowledge representation and reasoning. In a manner of speaking, we can say that “knowledge representation and reasoning” is another way to call symbolic AI as one of the oldest areas in the field. This property of symbolic AI brings two advantages: interpretability to understand conclusions or decisions, and flexibility to adapt the knowledge base to different domains.
For this reason, researchers are looking into ways to combine latest data-driven AI models (neural AI, due to most modern models being based on neural network architectures) with symbolic AI approaches, yielding as a result hybrid solutions like neuro-symbolic AI: “neural” AI systems like language models or computer vision systems, supported by a symbolic AI that applies logical reasoning process on the analyzed data to reinforce their understanding. These hybrid systems could also pave the way to more interpretable and nearly white-box AI systems, comprehending how the system reaches a conclusion or fine-tunes it with the aid of symbolic reasoning.
Companies recently focusing on hybrid AI solutions that integrate symbolic AI include: IBM, Franz Inc., UMNAI, and startups like Symbolica AI.
Wrapping Up
This article provided a gentle introduction to symbolic AI, a term referring to a partly forgotten area of AI that is regaining importance recently due to its potential to further enhance modern AI applications while trying to overcome some of their challenges. By providing a glimpse of the notion of symbolic AI, its historical milestones and decline, we hope you gained a better understanding of this intriguing but perhaps re-awakening area of AI.
Iván Palomares Carrascosa is a leader, writer, speaker, and adviser in AI, machine learning, deep learning & LLMs. He trains and guides others in harnessing AI in the real world.