Benevolent AI
January 31, 2024Wearable Devices CUE1®
January 31, 2024Logical vs Nonlogical Artificial Intelligence
Logic and Non-Logic based AI have divergently evolved to master techniques and methods that allow them each to tackle unique challenges...
Introduction
In contrast to long-established logic or symbolic AI techniques that are based on the formal representation of knowledge and follow explicit reasoning rules, the development of advanced technologies and the availability of large amounts of data have favored the expansion of non-logic or sub-symbolic AI techniques, such as Deep Learning and Neural Networks. The previously mentioned are branches of Machine Learning (ML) aiming to exploit Big Data to make predictions and take autonomous decisions. Logic and non-logic AI have marked differences between them that make them each perfect for what they are designed for.(1)
Logic Artificial Intelligence
Logic-based technologies have been developed as models capable of responding to the needs of knowledge representation, reasoning, and verification, aiming to mimic human behavior.(1)
Computational Logic (CL) was born from the joint effort of Computer Science (CS) and Artificial Intelligence (AI) researchers trying to endow machines with human-like reasoning capabilities. CL has been exploited to offer technology the ability to compute, represent computation, and reason about computation in a human-understandable way.(1)
Knowledge representation has been key since the very beginning of AI since no reasoning can exist without knowledge. Logic-based knowledge representations mostly rely on description and modal logic to represent, respectively, terminological and time-depending/subjective knowledge.(1)
Reasoning based on classical deductive logic is monotonic, while commonsense reasoning is not monotonic. This slight difference is illustrated in the following example: you leave for work, and your reasoning tells you that your house is still standing even if you don’t physically see it; while at work, you are informed a tornado hit the location of your house, so you drop the original inference because of the addition of new information. The same happens if you are told Tweety is a bird and assume Tweety can fly until you’re told Tweety is a penguin. Nonmonotonous logic uses formalisms created to capture the mechanism underlying these examples.(2)
The evolving formalisms and techniques of logic AI have reached a level of impressive maturity while trying to accomplish three specific points: creating intelligent agents, achieving interoperability when also applying commonsense logic, and finally attaining the encoding down technique to allow efficient reasoning.(2)
Nonlogic Artificial Intelligence
Non-logic AI can be classified into symbolic but non-logic approaches (probabilistic like Bayesian Networks) and connectionist-neurocomputational approaches (e.g., Artificial Neural Networks).(2)
Neural Networks (NN) are composed of units (neurons) that are connected by links (dendrites) with a numeric weight. These networks use inputs to generate outputs for the next neuron. The thought that Neural Networks were simple and had theoretically efficient learning algorithms got complicated with the evolution of these systems into thick, multi-layered networks. Thus, researchers have been developing several mechanisms to make them more explainable to human users; for instance, the Backpropagation method is used to translate neural networks into a sequence of repeated arithmetic operations on big sets of numbers.(2)
The best way to explain Neural Networks is in the context of statistical formalisms and methods that allow the user to solve a specific problem (that does not seem to be solvable using logic-AI techniques); for example, handwriting recognition is approached by assigning a value to each alphabetical digit and creating a massive database that software can use when identifying handwritten data.(2)