There is a broad consensus that both learning and reasoning are essential to achieve true artificial intelligence. This has put the quest for neural-symbolic artificial intelligence (NeSy) high on the research agenda. In the past decade, neural networks have caused great advances in the field of machine learning.
On a high level, Aristotle’s theory of motion states that all things come to a rest, heavy things on the ground and lighter things on the sky, and force is required to move objects. It was only when a more fundamental understanding of objects outside of Earth became available through the metadialog.com observations of Kepler and Galileo that this theory on motion no longer yielded useful results. We believe that our results are the first step to direct learning representations in the neural networks towards symbol-like entities that can be manipulated by high-dimensional computing.
The practice showed a lot of promise in the early decades of AI research. But in recent years, as neural networks, also known as connectionist AI, gained traction, symbolic AI has fallen by the wayside. Henry Kautz, Francesca Rossi, and Bart Selman have also argued for a synthesis. Their arguments are based on a need to address the two kinds of thinking discussed in Daniel Kahneman’s book, Thinking, Fast and Slow.
We might teach the program rules that might eventually become irrelevant or even invalid, especially in highly volatile applications such as human behavior, where past behavior is not necessarily guaranteed. Even if the AI can learn these new logical rules, the new rules would sit on top of the older (potentially invalid) rules due to their monotonic nature. As a result, most Symbolic AI paradigms would require completely remodeling their knowledge base to eliminate outdated knowledge.
For Symbolic AI to remain relevant, it requires continuous interventions where the developers teach it new rules, resulting in a considerably manual-intensive process. Surprisingly, however, researchers found that its performance degraded with more rules fed to the machine. Symbolic AI is more concerned with representing the problem in symbols and logical rules (our knowledge base) and then searching for potential solutions using logic.
For instance, if you take a picture of your cat from a somewhat different angle, the program will fail. Similarly, Allen’s temporal interval algebra is a simplification of reasoning about time and Region Connection Calculus is a simplification of reasoning about spatial relationships. A more flexible kind of problem-solving occurs when reasoning about what to do next occurs, rather than simply choosing one of the available actions. This kind of meta-level reasoning is used in Soar and in the BB1 blackboard architecture. Japan championed Prolog for its Fifth Generation Project, intending to build special hardware for high performance. Similarly, LISP machines were built to run LISP, but as the second AI boom turned to bust these companies could not compete with new workstations that could now run LISP or Prolog natively at comparable speeds.
Another benefit of combining the techniques lies in making the AI model easier to understand. Humans reason about the world in symbols, whereas neural networks encode their models using pattern activations. Symbolic AI’s strength lies in its knowledge representation and reasoning through logic, making it more akin to Kahneman’s „System 2” mode of thinking, which is slow, takes work and demands attention.
His team has been exploring different ways to bridge the gap between the two AI approaches. The more knowledge you have, the less searching you need to do for an answer you need. This trade-off between knowledge and search is unavoidable in any AI system.
While the early rush into LLM is new hype, it can provide impressive results, and follows the huge diffusion of AI in corporations, with the size of investment in AI went up 5 times from 12 US billion in 2015 to close to 70 USD billion 5 years later. Besides being high profile corporations, they are been experimenting aggressively with foundational models linked to LLMs such as #OpenAI’s chatGPT. The botmaster then needs to review those responses and has to manually tell the engine which answers were correct and which ones were not.
As one might also expect, common sense differs from person to person, making the process more tedious. In a nutshell, Symbolic AI has been highly performant in situations where the problem is already known and clearly defined (i.e., explicit knowledge). Translating our world knowledge into logical rules can quickly become a complex task.
In those cases, rules derived from domain knowledge can help generate training data. A different type of knowledge that falls in the domain of Data Science is the knowledge encoded in natural language texts. While natural language processing has made leaps forward in past decade, several challenges still remain in which methods relying on the combination of symbolic AI and Data Science can contribute.
Symbolic systems acknowledge this and give their algorithms a large amount of knowledge to process. They have been widely applicable to games, as they can model various aspects of game logic, such as blackboard architectures, pathfinding, decision trees, state machines, and more. Intelligent machines should support and aid scientists during the whole research life cycle and assist in recognizing inconsistencies, proposing ways to resolve the inconsistencies, and generate new hypotheses. Don’t get me wrong, machine learning is an amazing tool that enables us to unlock great potential and AI disciplines such as image recognition or voice recognition, but when it comes to NLP, I’m firmly convinced that machine learning is not the best technology to be used. Symbolic AI is built around a rule-based model that enables greater visibility into its operations and decision-making processes.
Symbolic planning investigates how robots can choose the best route based on the task and the constraint on accomplishing that task (such as least travelling time or shortest travelling distance). Formal verification has been applied to this area, and can provide a better solution than other methods.