In AI, reasoning is the process of drawing conclusions or inferring something new about a domain of interest based on the knowledge we have. It is an essential component of what we refer to as "intelligence." In fact, this is the distinction between a traditional database and a knowledge base. Unlike the expert system's knowledge base, the database can't think and can only answer a limited number of specific questions. 

Types of reasoning

Reasoning can go in one of two directions: forward toward the goal or backward away from the goal. In AI, both are used in different situations.

• Forward reasoning starts with known facts and works its way to the desired outcome. This is an example of deductive reasoning in action.
• Backward reasoning starts with the goal and creates sub-goals that must be completed before the main goal can be completed. This is an example of inductive reasoning in action.

Tools used

Using methods from probability theory and economics, AI researchers have developed a number of tools to solve these problems. The following are some of the most commonly used tools.

Bayesian network

A Bayesian network is a probabilistic graphical model that uses a directed acyclic graph (DAG) to represent a set of variables and their conditional dependencies. Bayesian networks are ideal for predicting the likelihood that any one of several possible known causes contributed to an event that occurred. For example, it could be used to represent the probabilistic relationships between diseases and symptoms. The network can be used to figure out how likely it is that different diseases are present based on the symptoms they have.

Hidden Markov model

The Hidden Markov Model (HMM) is a type of Markov model in which the modelled system is assumed to be a Markov process. Statistical mechanics, thermodynamics, physics, chemistry, finance, signal processing, and information theory are all fields that use hidden Markov models. Hidden Markov models are also used in pattern recognition applications like speech and handwriting recognition, gesture recognition, part-of-speech tagging, musical score following, partial discharge, and bioinformatics.

Kalman filter

Kalman filtering is a statistical and control theory algorithm that uses time series measurements with statistical noise and other errors. In addition, Kalman filtering has many technological uses. Common applications include dynamically positioned aircraft, spacecraft, and ships. Kalman filtering is also widely used in time series analysis for topics like signal processing and econometrics. Kalman filtering is a common topic in robotic motion planning and control, and can be used to optimize a trajectory. Furthermore, Kalman filtering can model the brain's control of movement.

Particle filter

Particle filters, also known as sequential Monte Carlo methods, are a family of Monte Carlo algorithms for solving filtering problems in signal processing and Bayesian statistical inference. Del Moral coined the term "particle filters" in 1996 to describe mean-field interacting particle methods, which have been used in fluid mechanics since the 1960s. In 1998, Liu and Chen coined the phrase "Sequential Monte Carlo." Moreover, Particle filtering is a way to show the probability distribution of a stochastic process based on incomplete or noisy data. It does this by using a set of particles, which are also called samples.

Decision theory

The study of an agent's choices is called decision theory (or theory of choice; not to be confused with choice theory). Economists, mathematicians, data scientists, psychologists, biologists, social scientists, philosophers, and computer scientists all study decision theory, which is closely related to the field of game theory.

Conclusion

Probabilistic reasoning is another name for reasoning under uncertainty. In real-world situations, agents are inevitably forced to make decisions based on incomplete data. Even when an agent senses the world to learn more, it rarely learns the precise state of the world. A doctor, for example, has no idea what is going on inside a patient, a teacher has no idea what a student understands, and a robot has no idea what is in the room it left a few minutes ago. When an intelligent agent needs to act, it must make use of whatever data it has. Many AI problems necessitate the agent's operation with incomplete or uncertain data.

Image source: Unsplash