The fall of Symbolic AI
In the early 1980s, most AI developers moved away from Symbolic AI. Symbolic AI spectacularly crashed into an AI winter since it lacked common sense. Researchers began investigating newer algorithms and frameworks to achieve machine intelligence. As a result, Symbolic AI lost its allure quite rapidly. Furthermore, the limitations of Symbolic AI were becoming significant enough not to let it reach higher levels of machine intelligence and autonomy. In the following subsections, we will delve deeper into the substantial limitations and pitfalls of Symbolic AI.
Common sense is not so common
In a nutshell, Symbolic AI has been highly performant in situations where the problem is already known and clearly defined (i.e., explicit knowledge). Symbolic AI heavily relies on explicit symbolic representations. However, the world around us is filled with implicit knowledge. Our universe is a rather abstract concept. Translating our world knowledge into logical rules can quickly become a complex task. While in Symbolic AI, we tend to rely heavily on Boolean logic computation, the world around us is far from Boolean. Most physical symbols and relations are fuzzy. They are not static but rather based on a degree of truthiness. For example, a digital screen’s brightness is not just on or off, but it can also be any other value between 0% and 100% brightness. A person can be a little hungry as opposed to completely starving. The concept of fuzziness adds a lot of extra complexities to designing Symbolic AI systems. Due to fuzziness, multiple concepts become deeply abstracted and complex for Boolean evaluation.
Additionally, it introduces a severe bias due to human interpretability. Let’s pick a simple analogy – the color cyan. For some, it is cyan; for others, it might be aqua, turquoise, or light blue. As such, initial input symbolic representations lie entirely in the developer’s mind, making the developer crucial. Recall the example we mentioned in Chapter 1 regarding the population of the United States. It can be answered in various ways, for instance, less than the population of India or more than 1. Both answers are valid, but both statements answer the question indirectly by providing different and varying levels of information; a computer system cannot make sense of them. This issue requires the system designer to devise creative ways to adequately offer this knowledge to the machine.
Inevitably, this issue results in another critical limitation of Symbolic AI – common-sense knowledge. The human mind can generate automatic logical relations tied to the different symbolic representations that we have already learned. Humans learn logical rules through experience or intuition that become obvious or innate to us. They tend to come to us naturally, without us overthinking them. For example, a child must always be younger than their parents. We close our eyes when we want to sleep. We do not eat food that smells like it has gone bad. These are all examples of everyday logical rules that we humans just follow – as such, modeling our world symbolically requires extra effort to define common-sense knowledge comprehensively. Consequently, when creating Symbolic AI, several common-sense rules were being taken for granted and, as a result, excluded from the knowledge base. As one might also expect, common sense differs from person to person, making the process more tedious.
In the real world, there are so many different levels of abstraction, hierarchies, and underlying relationships. It’s impossible to capture all these rules entirely. To start, even humans do not know all the universe’s secrets. Let us recall the orange example from Figure 2.2. Assume we pass two fruits to the Symbolic AI program: an orange and a tangerine. With the symbolic structure and relations we had previously defined, it would be rather difficult to differentiate between them. Even a human might find this task difficult, let alone a machine that feeds knowledge through logical rules devised by a human.
Although Symbolic AI paradigms can learn new logical rules independently, providing an input knowledge base that comprehensively represents the problem is essential and challenging. The symbolic representations required for reasoning must be predefined and manually fed to the system. With such levels of abstraction in our physical world, some knowledge is bound to be left out of the knowledge base.
The test of time
Another concept we regularly neglect is time as a dimension of the universe. As we all know, time changes a lot of things. Some examples are our daily caloric requirements as we grow older, the number of stairs we can climb before we start gasping for air, and the leaves on trees and their colors during different seasons. These are examples of how the universe has many ways to remind us that it is far from constant.
A Symbolic AI system is said to be monotonic – once a piece of logic or rule is fed to the AI, it cannot be unlearned. Newly introduced rules are added to the existing knowledge, making Symbolic AI significantly lack adaptability and scalability. One power that the human mind has mastered over the years is adaptability. Humans can transfer knowledge from one domain to another, adjust our skills and methods with the times, and reason about and infer innovations. 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.
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. This phenomenon is referred to as concept drift or data morphism. In short, the underlying relationships of the data shift or change. 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. This remodeling process often becomes highly convoluted and tedious. For this reason, Symbolic AI systems are limited in updating their knowledge and have trouble making sense of unstructured data.
