Machine Reasoning

What Is Machine Reasoning?

A type of artificial intelligence (AI), machine reasoning is a software process that uses rules and various areas of logic to make inferences. Because machine reasoning uses logic – and in a way that is meant to be comparable to human logic – its applications are chiefly under the banner of robotics and natural language processing. I

The use of machine reasoning is one of the many reasons that software systems such as chatbots are becoming more conversational.

This technology helps such systems to make decisions based on the context of language and other complex information – as opposed to just receiving keywords and taking the implications of those words literally and without nuances.

How Does Machine Reasoning Work?

Machine reasoning works by the combination of being fed context-specific facts by humans and by developing its understanding of those facts through its own self-learning processes. It is due to this mixture of training and self-learning that machine reasoning is able to not only categorize and contextualize information but also form its own examples of the resultant categories and contexts.

An example of machine reasoning can be seen in its ability to use natural language processing (NLP). A machine reasoning-enabled NLP system, if trained correctly, would be able to differentiate between the use of the word on in the following sentences:

• There was one passenger on the plane.
• There was ice on the plane.

While the inputted word – on – is the same in both examples, the context makes their implied definitions very different, and both a human and a well-programmed machine reasoning system would be able to follow and act on the meaning of the two variations.

Why Is Machine Reasoning Important?

Machine reasoning is important because intelligent software systems would not be able to provide the same level of human-readable communication, decision-making, and automation without it.

In fact, for these reasons, machine reasoning is opening up opportunities for many disruptive applications, particularly natural language processing (NLP) systems.

In other words, machine reasoning is crucial because it is so applicable to both the needs of humans as well as the way that humans communicate and understand information.

Machine reasoning, in conjunction with NLP, is the reason that artificial intelligence systems are becoming increasingly able to answer humans’ questions, process the opinions of various industry experts, and summarize large amounts of information such as complex news stories.

How Does Machine Reasoning Fight Fraud?

Machine reasoning fights fraud by essentially carrying out detective work. Just as a fraud analyst needs to use reasoning and deduction to detect fraudsters and stop them in their tracks, so too must a machine reasoning system make inferences based on its input.

Such input varies depending on which approach to fighting fraud the machine reasoning system is tasked with. However, here are some of key applications and input that relate to machine-reasoning-based fraud fighting.

• Transaction analysis: Machine reasoning systems can be employed to carry out transaction analysis to detect both patterns and anomalies in an organization or individual’s log of their income and outgoings. By analyzing financial data and other information, a machine reasoning system can detect fraud based on whether someone’s transaction monitoring is consistent with previously observed patterns in legitimate transactions.

• Behavioral analytics: There are certain actions that web users are far more likely to take when they have nothing to hide. For example, legitimate online shoppers are not as likely to copy and paste credit card details, unlike fraudsters. Copying and pasting card details in itself may only be a minor cause for suspicion. However, machine reasoning can carry out behavioral analytics to also determine other factors that could help to identify fraud: time spent on a web page, hesitation, keystrokes, and so on. Importantly, machine reasoning will identify these and then pinpoint any deviation, rather than being given a specific, limited list of actions or red flags.
  
• Linguistic analysis: There are arguably certain speech patterns and word choices that fraudsters tend use which are less likely to be written by innocent individuals. A machine reasoning system can therefore be fed, for example, an email that is potentially phishing, and draw its conclusions about whether the author is in fact suspicious – and even explain its reasons for making its verdict, if it is a clearbox system.

What Is the Difference Between Machine Learning and Machine Reasoning?

Though they are both part of artificial intelligence, machine learning is more formulaic and objective, whereas the machine reasoning is more open-ended and intuitive.

For example, an ML system is more applicable to reaching conclusions based on objective information like the intrinsic properties of text (such as syntax) and images (such as the colors and number of items in the images).

Meanwhile, machine reasoning systems can reach conclusions based on more subjective and qualitative information like the actual meaning or themes behind the text and images – such as, respectively, context and artistic interpretation.

It is important to remember that both machine learning and machine reasoning are crucial to decision-making and prediction modeling – especially so in the context of fraud-fighting – and they are far from mutually exclusive.

In fact, not only are their differences not always clear-cut, but the two systems can even be used in conjunction with each other: A machine reasoning system could, for example, be fed a list of transactions and then identify what it considers to be suspicious purchases. After that, it could then call on a machine learning system to draw up statistical models that will help determine the rate of potential fraud.

Related Terms

Related Articles