Human beings spend a lot of their time making decisions. One of the most powerful tools we use to make decisions is reasoning. Meriam-Webster defines reasoning as the process of thinking about something in a logical way in order to form a conclusion or judgment or, more simply, the ability of the mind to think and understand things in a logical way.
We may not always make decisions this way but, when required, humans have long engaged in creating formal arguments that include logical chains of reasoning based on facts and logical arguments. Not limited to Sherlock Holmes or the TV show House, reasoning is something we might use in a legal contract, a formal policy document, or when analyzing requirements for eligibility for everything from renting a car to qualifying for Medicaid. Logical reasoning done well is reproducible, explainable, understandable, and defensible. With reasoning, we get the right answer and can understand why it’s correct.
Logical AI: a revolution in the automation of human reasoning
For many years, computer scientists have sought to create decision making tools based on the way humans reason. Early expert systems in the 1980’s and 1990’s were quite limited, but they were central to the first wave of commercial success of Artificial Intelligence (AI). Outgrowths of that technology, business rules and decision trees, have become well established as distinct market sectors. Now ubiquitous, they automate millions of mission critical business processes every day. But automation of more complex human reasoning and knowledge remained illusive, due to technical limitations.
In the last fifteen years, game-changing advances occurred in logical AI, the field of computer science concerned with the automation of logical reasoning. These advances have come together in the form of Rulelog, a knowledge representation and reasoning (KRR) language that underlies the Flora-2 and Ergo Suite systems, and related technologies that grew out of logic programming. This next-generation logical AI may well revolutionize what is possible in ways we are only beginning to realize.
New applications based on logical AI advances
A raft of new applications based on these advances in logical reasoning are coming. We are hampered more by imagination at this point than by technology. Examples of such systems already being prototyped include:
- A banking regulations system that automatically and with high accuracy discovers whether a given financial transaction is allowed or prohibited, while furnishing explanations and an audit trail.
- An electronic health insurance contract that automatically tells the patient how much of a given claim is covered, or whether a claim has been denied and why.
- A textbook, enhanced by a chatbots tutor that interactively teaches a student any subject from accounting to zoology and can answer questions – accurately, in depth, and with interactively navigable explanations, not just at the level of keyword searches.
Technical advances that make it possible
Expressivity. In the past, logical AI systems were hampered by limitations in how English (or other natural language) sentences could be expressed in logical terms. Only relatively simple, formalized types of language could be encoded. Now almost any actionable knowledge that can be expressed in a natural language can be encoded in a logical language, without ambiguity and with semantic meaning in depth.
Scalability. Business rules and decision trees can represent only shallow knowledge. They scale well with data, but poorly with respect to knowledge. More complicated – deep – reasoning is out of bounds for these systems. The new generation of logical reasoning systems have no such limitations. In addition to arbitrarily complex chaining, the amount of rules and facts that can be handled in a given system has increased dramatically.
Flexibility. Information is constantly changing, and often presents situations where one statement – rule or other complex sentence – might conflict with another. Whereas older systems are unable to resolve conflicts, modern logical AI allows for priorities to be assigned to different rules so that those with higher priority are selected over those with lower priority. Updating with new information in the past might have necessitated a huge amount of work (done by expensive specialists) to refactor and re-optimize the knowledge bases. Today, due to smart cacheing and advances in formalized common-sense reasoning, new information and knowledge can be added on a ‘just-in-time’ basis without the need for such refactoring.
Explainability. It is often not enough to know the answer: many times we need to also know why a certain conclusion was reached. This is the power of logical AI systems: because these systems reach their conclusion due to a logical chain of reasoning, the reasoning can be made transparent. Interactively navigable explanation trees can be generated showing every step of how a conclusion has been reached. If one needs to justify a hiring decision, a claim rejection, or a regulatory compliance approval, an explanation can be crucial to acceptance of a decision or conclusion that was reached by a machine.
Learn more about logical AI and Ergo.
Janine Bloomfield, PhD is COO and co-founder of Coherent Knowledge, maker of Ergo Suite software.
image credit: https://commons.wikimedia.org/wiki/File%3ABenedict_Cumberbatch_filming_Sherlock.jpg by Fat Les (bellaphon) from London, UK (Flickr) via Wikimedia Commons/Janine Bloomfield