Symbolic vs Subsymbolic AI Paradigms for AI Explainability by Orhan G. Yalçın

For example, in the first world we saw above, we can see that Bess likes Cody, even though we are not told this fact explicitly. Given two sentences, we know the world must be in the intersection of the set of worlds in which the first sentence is true and the set of worlds in which the second sentence is true. Logical sentences like the ones above constrain the possible ways the world could be. Each sentence divides the set of possible worlds into two subsets, those in which the sentence is true and those in which the sentence is false, as suggested by the following figure. Believing a sentence is tantamount to believing that the world is in the first set.

All operations are executed in an input-driven fashion, thus sparsity and dynamic computation per sample are naturally supported, complementing recent popular ideas of dynamic networks and may enable new types of hardware accelerations. We experimentally show on CIFAR-10 that it can perform flexible visual processing, rivaling the performance of ConvNet, but without using any convolution. Furthermore, it can generalize to novel rotations of images that it was not trained for. We propose the Neuro-Symbolic Concept Learner (NS-CL), a model that learns visual concepts, words, and semantic parsing of sentences without explicit supervision on any of them; instead, our model learns by simply looking at images and reading paired questions and answers.

CNNs are good at processing information in parallel, such as the meaning of pixels in an image. RNNs better interpret information in a series, such as text or speech. New GenAI techniques often use transformer-based neural networks that automate data prep work in training AI systems such as ChatGPT and Google Gemini. We start with a look at the essential elements of logic – logical sentences, logical entailment, and logical proofs.

Note the similarity to the propositional and relational machine learning we discussed in the last article. Perhaps surprisingly, the correspondence between the neural and logical calculus has been well established throughout history, due to the discussed dominance of symbolic AI in the early days. Neural networks and other statistical techniques excel when there is a lot of pre-labeled data, such as whether a cat is in a video. However, they struggle with long-tail knowledge around edge cases or step-by-step reasoning.

Centers, Labs, & Programs

Wherever possible, we try to be clear about this distinction, but the potential for confusion remains. One of the first individuals to give voice to this idea was Leibnitz. He conceived of “a universal algebra by which all knowledge, including moral and metaphysical truths, can some day be brought within a single deductive system”.

The researchers found that NLEPs even exhibited 30 percent greater accuracy than task-specific prompting methods. First, the model calls the necessary packages, or functions, it will need to solve the task. Step two involves importing natural language representations of the knowledge the task requires (like a list of U.S. presidents’ birthdays). For step three, the model implements a function that calculates the answer.

Unfortunately, in general, there are many, many possible worlds; and, in some cases, the number of possible worlds is infinite, in which case model checking is impossible. We say that a set of premises logically entails a conclusion if and only if every world that satisfies the premises also satisfies the conclusion. We use it in our professional lives – in proving mathematical theorems, in debugging computer programs, in medical diagnosis, and in legal reasoning. And we use it in our personal lives – in solving puzzles, in playing games, and in doing school assignments, not just in Math but also in History and English and other subjects. With our NSQA approach , it is possible to design a KBQA system with very little or no end-to-end training data. Currently popular end-to-end trained systems, on the other hand, require thousands of question-answer or question-query pairs – which is unrealistic in most enterprise scenarios.

In fact, some other literary devices, like metaphor and allegory, are often considered to be types of symbolism. Literary devices are the techniques writers use to communicate ideas and themes beyond what they can express literally. “Usually, when people do this kind of few-shot prompting, they still have to design prompts for every task. We found that we can have one prompt for many tasks because it is not a prompt that teaches LLMs to solve one problem, but a prompt that teaches LLMs to solve many problems by writing a program,” says Luo. We note that this was the state at the time and the situation has changed quite considerably in the recent years, with a number of modern NSI approaches dealing with the problem quite properly now. However, to be fair, such is the case with any standard learning model, such as SVMs or tree ensembles, which are essentially propositional, too.

They have created a revolution in computer vision applications such as facial recognition and cancer detection. Neural networks are almost as old as symbolic AI, but they were largely dismissed because they were inefficient and required compute resources that weren’t available at the time. In the past decade, thanks to the large availability of data and processing power, deep learning has gained popularity and has pushed past symbolic AI systems. Symbolic AI involves the explicit embedding of human knowledge and behavior rules into computer programs. The practice showed a lot of promise in the early decades of AI research.

In essence, the concept evolved into a very generic methodology of using gradient descent to optimize parameters of almost arbitrary nested functions, for which many like to rebrand the field yet again as differentiable programming. This view then made even more space for all sorts of new algorithms, tricks, and tweaks that have been introduced under various catchy names for the underlying functional blocks (still consisting mostly of various combinations of basic linear algebra operations). Some proponents have suggested that if we set up big enough neural networks and features, we might develop AI that meets or exceeds human intelligence. However, others, such as anesthesiologist Stuart Hameroff and physicist Roger Penrose, note that these models don’t necessarily capture the complexity of intelligence that might result from quantum effects in biological neurons.

Fuzzy and Annotated Logic for Neuro Symbolic Artificial Intelligence

With this paradigm shift, many variants of the neural networks from the ’80s and ’90s have been rediscovered or newly introduced. Benefiting from the substantial increase in the parallel processing power of modern GPUs, and the ever-increasing amount of available data, deep learning has been steadily paving its way to completely dominate the Chat GPT (perceptual) ML. The true resurgence of neural networks then started by their rapid empirical success in increasing accuracy on speech recognition tasks in 2010 [2], launching what is now mostly recognized as the modern deep learning era. Shortly afterward, neural networks started to demonstrate the same success in computer vision, too.

Moreover, they lack the ability to reason on an abstract level, which makes it difficult to implement high-level cognitive functions such as transfer learning, analogical reasoning, and hypothesis-based reasoning. Finally, their operation is largely opaque to humans, rendering them unsuitable for domains in which verifiability is important. In this paper, we propose an end-to-end reinforcement learning architecture comprising a neural back end and a symbolic front end with the potential to overcome each of these shortcomings.

Computer Science

Additionally, application areas such a visual question answering and natural language processing are discussed as well as topics such as verification of neural networks and symbol grounding. Detailed algorithmic descriptions, example logic programs, and an online supplement that includes instructional videos and slides provide thorough but concise coverage of this important area of AI. Deep reinforcement learning (DRL) brings the power of deep neural networks to bear on the generic task of trial-and-error learning, and its effectiveness has been convincingly demonstrated on tasks such as Atari video games and the game of Go. However, contemporary DRL systems inherit a number of shortcomings from the current generation of deep learning techniques. For example, they require very large datasets to work effectively, entailing that they are slow to learn even when such datasets are available.

Such formal representations and methods are useful for us to use ourselves. Moreover, they allow us to automate the process of deduction, though the computability of such implementations varies with the complexity of the sentences involved. In this vein, since many forms of advanced mathematical reasoning rely on graphical representations and geometric principles, it would be surprising to find that perceptual and sensorimotor processes are not involved in a constitutive way. Therefore, by accounting for symbolic reasoning—perhaps the most abstract of all forms of mathematical reasoning—in perceptual and sensorimotor terms, we have attempted to lay the groundwork for an account of mathematical and logical reasoning more generally. New deep learning approaches based on Transformer models have now eclipsed these earlier symbolic AI approaches and attained state-of-the-art performance in natural language processing. However, Transformer models are opaque and do not yet produce human-interpretable semantic representations for sentences and documents.

While natural language works well in many circumstances, it is not without its problems. Natural language sentences can be complex; they can be ambiguous; and failing to understand the meaning of a sentence can lead to errors in reasoning. One of Aristotle’s great contributions to philosophy was the identification of syntactic operations that . By applying rules of inference to premises, we produce conclusions that are entailed by those premises.

As an author, you can’t prevent readers from making assumptions about what you meant to communicate in your work. You can certainly discuss your work and communicate what you wanted to depict through symbolism, but once your work is published and available to the public, you lose some control over how it’s interpreted. One famous example of false symbolism is the assumption that the Lord of the Rings books are an allegory for World War II.

Each approach—symbolic, connectionist, and behavior-based—has advantages, but has been criticized by the other approaches. Symbolic AI has been criticized as disembodied, liable to the qualification problem, and poor in handling the perceptual problems where deep learning excels. In turn, connectionist AI has been criticized as poorly suited for deliberative step-by-step problem solving, incorporating knowledge, and handling planning. Finally, Nouvelle AI excels in reactive and real-world robotics domains but has been criticized for difficulties in incorporating learning and knowledge. Henry Kautz,[17] Francesca Rossi,[79] and Bart Selman[80] have also argued for a synthesis.

Using symbolic AI, everything is visible, understandable and explainable, leading to what is called a ‘transparent box’ as opposed to the ‘black box’ created by machine learning. In a nutshell, symbolic AI involves the explicit embedding of human knowledge and behavior rules into computer programs. What can we conclude from the bits of information in our sample logical sentences? Since there are different values in different worlds, we cannot say yes and we cannot say no. However, we do not have enough information to say which case is correct.

That is, a symbol offers a level of abstraction above the concrete and granular details of our sensory experience, an abstraction that allows us to transfer what we’ve learned in one place to a problem we may encounter somewhere else. In a certain sense, every abstract category, like chair, asserts an analogy between all the disparate objects called chairs, and we transfer our knowledge about one chair to another with the help of the symbol. Many popular large language models work by predicting the next word, or token, given some natural language input. While models like GPT-4 can be used to write programs, they embed those programs within natural language, which can lead to errors in the program reasoning or results.

A set of premises logically entails a conclusion if and only if every possible world that satisfies the premises also satisfies the conclusion. A sentence is provable from a set of premises if and only if there is a finite sequence of sentences in which every element is either a premise or the result of applying a deductive rule of inference to earlier members in the sequence. Model checking is the process of examining the set of all worlds to determine logical entailment.

A similar problem, called the Qualification Problem, occurs in trying to enumerate the preconditions for an action to succeed. An infinite number of pathological conditions can be imagined, e.g., a banana in a tailpipe could prevent a car from operating correctly. The General Problem Solver (GPS) cast planning as problem-solving used means-ends analysis to create plans. STRIPS took a different approach, viewing planning as theorem proving.

With more linguistic stimuli received in the course of psychological development, children then adopt specific syntactic rules that conform to Universal grammar. The automated theorem provers discussed below can prove theorems in first-order logic. Horn clause logic is more restricted than first-order logic and is used in logic programming languages such as Prolog. Extensions to first-order logic include temporal logic, to handle time; epistemic logic, to reason about agent knowledge; modal logic, to handle possibility and necessity; and probabilistic logics to handle logic and probability together. Expert systems can operate in either a forward chaining – from evidence to conclusions – or backward chaining – from goals to needed data and prerequisites – manner. More advanced knowledge-based systems, such as Soar can also perform meta-level reasoning, that is reasoning about their own reasoning in terms of deciding how to solve problems and monitoring the success of problem-solving strategies.

Since its foundation as an academic discipline in 1955, Artificial Intelligence (AI) research field has been divided into different camps, of which symbolic AI and machine learning. While symbolic AI used to dominate in the first decades, machine learning has been very trendy lately, so let’s try to understand each of these approaches and their main differences when applied to Natural Language Processing (NLP). Read more about our work in neuro-symbolic AI from the MIT-IBM Watson AI Lab. Our researchers are working to usher in a new era of AI where machines can learn more like the way humans do, by connecting words with images and mastering abstract concepts.

You can foun additiona information about ai customer service and artificial intelligence and NLP. But in recent years, as neural networks, also known as connectionist AI, gained traction, symbolic AI has fallen by the wayside. Similar to the problems in handling dynamic domains, common-sense reasoning is also difficult to capture in formal reasoning. Examples of common-sense reasoning include implicit reasoning about how people think or general knowledge of day-to-day events, objects, and living creatures.

At the heart of Symbolic AI lie key concepts such as Logic Programming, Knowledge Representation, and Rule-Based AI. These elements work together to form the building blocks of Symbolic AI systems. We’re just basically enjoying our time together and, you know, hopeful for what the future has,” she continued.

Critiques from outside of the field were primarily from philosophers, on intellectual grounds, but also from funding agencies, especially during the two AI winters. Marvin Minsky first proposed frames as a way of interpreting common visual situations, such as an office, and Roger Schank extended this idea to scripts for common routines, such as dining out. Cyc has attempted to capture useful common-sense knowledge and has “micro-theories” to handle particular kinds of domain-specific reasoning. Forward chaining inference engines are the most common, and are seen in CLIPS and OPS5.

LNNs are a modification of today’s neural networks so that they become equivalent to a set of logic statements — yet they also retain the original learning capability of a neural network. Standard neurons are modified so that they precisely model operations in With real-valued logic, variables can take on values in a continuous range between 0 and 1, rather than just binary values of ‘true’ or ‘false.’real-valued logic. LNNs are able to model formal logical reasoning by applying a recursive neural computation of truth values that moves both forward and backward (whereas a standard neural network only moves forward). As a result, LNNs are capable of greater understandability, tolerance to incomplete knowledge, and full logical expressivity.

• The two biggest flaws of deep learning are its lack of model interpretability (i.e. why did my model make that prediction?) and the amount of data that deep neural networks require in order to learn.
• Originally, researchers favored the discrete, symbolic approaches towards AI, targeting problems ranging from knowledge representation, reasoning, and planning to automated theorem proving.
• P.J.B. performed the research, contributed new analytical tools and analyzed data.
• And if you take into account that a knowledge base usually holds on average 300 intents, you now see how repetitive maintaining a knowledge base can be when using machine learning.
• Given four girls, there are sixteen possible instances of the likes relation – Abby likes Abby, Abby likes Bess, Abby likes Cody, Abby likes Dana, Bess likes Abby, and so forth.

The advantage of neural networks is that they can deal with messy and unstructured data. Instead of manually laboring through the rules of detecting cat pixels, you can train a deep learning algorithm on many pictures of cats. The neural network then develops a statistical model for cat images.

Situated robotics: the world as a model

These dynamic models finally enable to skip the preprocessing step of turning the relational representations, such as interpretations of a relational logic program, into the fixed-size vector (tensor) format. They do so by effectively reflecting the variations in the input data structures into variations in the structure of the neural model itself, constrained by some shared parameterization (symmetry) scheme reflecting the respective model prior. It has now been argued by many that a combination of deep learning with the high-level reasoning capabilities present in the symbolic, logic-based approaches is necessary to progress towards more general AI systems [9,11,12].

The rule-based nature of Symbolic AI aligns with the increasing focus on ethical AI and compliance, essential in AI Research and AI Applications. Symbolic AI-driven chatbots exemplify the application of AI algorithms in customer service, showcasing the integration of AI Research findings into real-world AI Applications. Contrasting Symbolic AI with Neural Networks offers insights into the diverse approaches within AI.

Boost LLM application development with many-shot learning

To test this hypothesis, they investigated the way manipulations of visual groups affect participants’ application of operator precedence rules. Maruyama et al. (2012) argue on the basis of fMRI and MEG evidence that mathematical expressions like these are parsed quickly by visual cortex, using mechanisms that are shared with non-mathematical spatial perception tasks. In fact, rule-based AI systems are still very important in today’s applications.

Similar axioms would be required for other domain actions to specify what did not change. Qualitative simulation, such as Benjamin Kuipers’s QSIM,[88] approximates human reasoning about naive physics, such as what happens when we heat a liquid in a pot on the stove. We expect it to heat and possibly boil over, even though we may not know its temperature, its boiling point, or other details, such as atmospheric pressure. Cognitive architectures such as ACT-R may have additional capabilities, such as the ability to compile frequently used knowledge into higher-level chunks.

The program improved as it played more and more games and ultimately defeated its own creator. In 1959, it defeated the best player, This created a fear of AI dominating AI. This lead towards the connectionist paradigm of AI, also called non-symbolic AI which gave rise to learning and neural network-based approaches what is symbolic reasoning to solve AI. Although deep learning has historical roots going back decades, neither the term “deep learning” nor the approach was popular just over five years ago, when the field was reignited by papers such as Krizhevsky, Sutskever and Hinton’s now classic (2012) deep network model of Imagenet.

They prompt the model to generate a step-by-step program entirely in Python code, and then embed the necessary natural language inside the program. These old-school parallels between individual neurons and logical connectives might seem outlandish in the modern context of deep learning. Meanwhile, with the progress in computing power and amounts of available data, another approach to AI has begun to gain momentum.

Using the methods of algebra, we can then manipulate these expressions to solve the problem. In this regard, there is a strong analogy between the methods of Formal Logic and those of high school algebra. To illustrate this analogy, consider the following algebra problem. The form of the argument is the same as in the previous example, but the conclusion is somewhat less believable.

Automated reasoning programs can be used to check proofs and, in some cases, to produce proofs or portions of proofs. We can use this operation to solve the problem of Mary’s love life. Looking at the two premises above, we notice that p occurs on the left-hand side of one sentence and the right-hand side of the other. Consequently, we can cancel the p and thereby derive the conclusion that, if is Monday and raining, then Mary loves Quincy or Mary loves Quincy. (1) If a proposition on the left hand side of one sentence is the same as a proposition on the right hand side of the other sentence, it is okay to drop the two symbols, with the proviso that only one such pair may be dropped. (2) If a constant is repeated on the same side of a single sentence, all but one of the occurrences can be deleted.

Symbolic AI has numerous applications, from Cognitive Computing in healthcare to AI Research in academia. Its ability to process complex rules and logic makes it ideal for fields requiring precision and explainability, such as legal and financial domains. Along with boosting the accuracy of large language models, NLEPs could also improve data privacy. Since NLEP programs are run locally, sensitive user data do not need to be sent to a company like OpenAI or Google to be processed by a model. They found that NLEPs enabled large language models to achieve higher accuracy on a wide range of reasoning tasks. The approach is also generalizable, which means one NLEP prompt can be reused for multiple tasks.

• As opposed to pure neural network–based models, the hybrid AI can learn new tasks with less data and is explainable.
• Their arguments are based on a need to address the two kinds of thinking discussed in Daniel Kahneman’s book, Thinking, Fast and Slow.
• You can use symbolism in the allusions you make, like alluding to “going down the rabbit hole” in a personal essay by suggesting that you’re late for a very important date.
• The problem in this case is that the use of nothing here is syntactically similar to the use of beer in the preceding example, but in English it means something entirely different.

However, in the meantime, a new stream of neural architectures based on dynamic computational graphs became popular in modern deep learning to tackle structured data in the (non-propositional) form of various sequences, sets, and trees. Most recently, an extension to arbitrary (irregular) graphs then became extremely popular as Graph Neural Networks (GNNs). This is easy to think of as a boolean circuit (neural network) sitting on top of a propositional interpretation (feature vector). However, the relational program input interpretations can no longer be thought of as independent values over a fixed (finite) number of propositions, but an unbound set of related facts that are true in the given world (a “least Herbrand model”).

Also, some tasks can’t be translated to direct rules, including speech recognition and natural language processing. OOP languages allow you to define classes, specify their properties, and organize them https://chat.openai.com/ in hierarchies. You can create instances of these classes (called objects) and manipulate their properties. Class instances can also perform actions, also known as functions, methods, or procedures.

Symbols can be organized into hierarchies (a car is made of doors, windows, tires, seats, etc.). They can also be used to describe other symbols (a cat with fluffy ears, a red carpet, etc.). Controversies arose from early on in symbolic AI, both within the field—e.g., between logicists (the pro-logic “neats”) and non-logicists (the anti-logic “scruffies”)—and between those who embraced AI but rejected symbolic approaches—primarily connectionists—and those outside the field.

Prolog provided a built-in store of facts and clauses that could be queried by a read-eval-print loop. The store could act as a knowledge base and the clauses could act as rules or a restricted form of logic. As a subset of first-order logic Prolog was based on Horn clauses with a closed-world assumption—any facts not known were considered false—and a unique name assumption for primitive terms—e.g., the identifier barack_obama was considered to refer to exactly one object. At the height of the AI boom, companies such as Symbolics, LMI, and Texas Instruments were selling LISP machines specifically targeted to accelerate the development of AI applications and research. In addition, several artificial intelligence companies, such as Teknowledge and Inference Corporation, were selling expert system shells, training, and consulting to corporations. Insofar as computers suffered from the same chokepoints, their builders relied on all-too-human hacks like symbols to sidestep the limits to processing, storage and I/O.

Automated reasoning techniques can be used to compute new tables, to detect problems, and to optimize queries. The existence of a formal language for representing information and the existence of a corresponding set of mechanical manipulation rules together have an important consequence, viz. Logic eliminates these difficulties through the use of a formal language for encoding information. Given the syntax and semantics of this formal language, we can give a precise definition for the notion of logical conclusion. Moreover, we can establish precise reasoning rules that produce all and only logical conclusions. In talking about Logic, we now have two notions – logical entailment and provability.

Looking at the worlds above, we see that all of these sentences are true in the world on the left. By contrast, several of the sentences are false in the world on the right. We hope this work also inspires a next generation of thinking and capabilities in AI.

In symbolic reasoning, the rules are created through human intervention and then hard-coded into a static program. Although we will emphasize the kinds of algebra, arithmetic, and logic that are typically learned in high school, our view also potentially explains the activities of advanced mathematicians—especially those that involve representational structures like graphs and diagrams. Our major goal, therefore, is to provide a novel and unified account of both successful and unsuccessful episodes of symbolic reasoning, with an eye toward providing an account of mathematical reasoning in general. Before turning to our own account, however, we begin with a brief outline of some more traditional views. The difficulties encountered by symbolic AI have, however, been deep, possibly unresolvable ones. One difficult problem encountered by symbolic AI pioneers came to be known as the common sense knowledge problem.

Consequently, also the structure of the logical inference on top of this representation can no longer be represented by a fixed boolean circuit. Historically, the two encompassing streams of symbolic and sub-symbolic stances to AI evolved in a largely separate manner, with each camp focusing on selected narrow problems of their own. Originally, researchers favored the discrete, symbolic approaches towards AI, targeting problems ranging from knowledge representation, reasoning, and planning to automated theorem proving. Another area of innovation will be improving the interpretability and explainability of large language models common in generative AI. While LLMs can provide impressive results in some cases, they fare poorly in others.

Each method executes a series of rule-based instructions that might read and change the properties of the current and other objects. Natural language processing focuses on treating language as data to perform tasks such as identifying topics without necessarily understanding the intended meaning. Natural language understanding, in contrast, constructs a meaning representation and uses that for further processing, such as answering questions.

The purpose of this paper is to generate broad interest to develop it within an open source project centered on the Deep Symbolic Network (DSN) model towards the development of general AI. In the next article, we will then explore how the sought-after relational NSI can actually be implemented with such a dynamic neural modeling approach. Particularly, we will show how to make neural networks learn directly with relational logic representations (beyond graphs and GNNs), ultimately benefiting both the symbolic and deep learning approaches to ML and AI. Symbolic processes are also at the heart of use cases such as solving math problems, improving data integration and reasoning about a set of facts.

They are also better at explaining and interpreting the AI algorithms responsible for a result. However, this also required much manual effort from experts tasked with deciphering the chain of thought processes that connect various symptoms to diseases or purchasing patterns to fraud. This downside is not a big issue with deciphering the meaning of children’s stories or linking common knowledge, but it becomes more expensive with specialized knowledge. Symbols in the language represent “conditions” in the world, and complex sentences in the language express interrelationships among these conditions.