AI Requires More Than Machine Learning Posted on : Oct 16 - 2018

Lauded primarily for its automation and decision support, machine learning is undoubtedly a vital component of artificial intelligence. However, a small but growing number of thought leaders throughout the industry are acknowledging that the breadth of AI’s upper cognitive capabilities involves more than just machine learning.

Machine learning is all about sophisticated pattern recognition. It’s virtually unsurpassable at determining relevant, predictive outputs from a series of data-driven inputs. Nevertheless, there is a plethora of everyday, practical business problems that cannot be solved with input/output reasoning alone. The problems also require the multistep, symbolic reasoning of rules-based systems.

Whereas machine learning is rooted in a statistical approach, symbolic reasoning is predicated on the symbolic representation of a problem usually rooted in a knowledge base. Most rules-based systems involve multistep reasoning, including those powered by coding languages such as Prolog.

Symbolic reasoning can accomplish tasks that no form of advanced machine learning can, simply because it pertains to a specific knowledge base and not pattern recognition. The confluence of symbolic reasoning and machine learning enables the enterprise to solve an assortment of complicated business problems applicable to real-world situations -- as opposed to simply automating facile, repetitive tasks.

By using machine learning as a feedback mechanism to improve the knowledge of symbolic reasoning, organizations can embed artificial intelligence (AI) into their core business processes for untold possibilities.

Knowledge-Based Rules

Perhaps the most prevalent example of the utility of combining machine learning with multistep reasoning for extensive AI capabilities exists in the internet of things, in which the real-time monitoring of equipment asset management has burgeoned into a lucrative venture for many companies. Regardless of the specific equipment monitored -- whether airplane parts in the Industrial internet or connected ovens and washing machines in consumer homes -- machine learning is regularly deployed to monitor equipment performance and predict failures for minimal downtime. However, by feeding these machine learning predictions into a rules-based system, repairmen or maintenance personnel can see what action is most effective to prevent failure.

In this example, machine learning results feed a concrete knowledge base of engineering principles, operating principles and properties of the equipment monitored. That knowledge, linked together in an enterprise knowledge graph, is the foundation of a constrained processing system of rules about the monitoring and repair of equipment. These rules include information about operations and maintenance typically found in user manuals for the equipment -- whether they are ovens, washing machines or planes.

The symbolic reasoning system, not machine learning, is what guides the human action for preventive maintenance. However, that rules-based system also incorporates machine learning results of the outcomes of repairs to see how best to modify rules. Thus, the results of human intervention are included in the machine learning algorithms for a continuous feedback loop of ongoing progress. View More