Decoding AI, MI and ML: What’s the difference?
*This blog originally appeared on ARC Advisory's Industrial Viewpoints blog*
The technology industry has long been enamoured with the idea of Artificial Intelligence (AI), but until recently the fruits of that labour have been more science fiction than fact. With social media platforms, cloud providers, and “big data” software companies ingesting massive amounts of data there has been an uptick in the practical applications of AI. In this blog, we’ll separate fact from fiction as we uncover what AI, Machine Intelligence (MI), and Machine Learning (ML) actually mean and how they accelerate Industrial IoT data projects.
AI, Not Robots
Popular culture has long associated AI with robotics. Artificial Intelligence actually is a method to teach computers to make decisions. AI includes Machine Learning algorithms, natural language processing, knowledge representation, and automated reasoning.
There are many methods to teach machines to do our thinking and automate processes but the one garnering the most success is deep learning. Deep learning is a type of ML that analyzes data at different abstraction layers of the neural network – so if you taught the machine to recognize an image of a jet engine, it would abstract elements of the asset to different layers.
It is a long iterative process to teach the machine how to identify things as humans do and subsequently teach it to automate processes and make decisions. Over time the machine will learn to recognize which layers in the network are important to make decisions quicker.
Companies like Google and Facebook get a lot of attention for their efforts. Google replaced the algorithm that has long been the hallmark of its superior search product with deep learning and ML. Industrial applications for AI and ML have flown under the radar, but with the advent of Industrial IoT this has changed. With exponential volumes of data being created by assets and devices, industrial enterprises are in the position to radically improve operations and business outcomes with data.
Applying Machine Intelligence to Industrial Data
Machine Intelligence is an area of Artificial Intelligence that uses machine learning algorithms, reasoning algorithms and methods for automated detection and integration of data sources from any device and system. For industrial enterprises dealing with messy data, MI is the solution to automate data discovery and learning, without the need for manual extract, transform, and load (ETL) methodologies.
What differentiates MI from classic AI is the ability to analyze data streams for continuous learning while classic AI and simple neural networks rely on batch learning. AI also requires that you know what you are looking for, so in the case of an industrial enterprise discovering sensor data for the first time, an obvious application or question may not be known. This capability allows MI to go beyond classification and become predictive.
Bit Stew’s unique approach to MI is in the orchestration of ML algorithms within the technology. MIx technology focuses on expert systems and cognitive computing to execute a range of MI concepts such as semantic data modeling, federated indexing, independent reasoning among others.
The Semantic Model in particular, allows the system (and its users) to understand information being processed regardless of the data source in real time. As data is ingested into the MIx technology stack more of these core competencies are leveraged in a highly structured manner allowing the machine to learn. Bit Stew’s Data Management Workbench relies on the underlying Machine Intelligence capabilities of the technology to create intuitive workflows for data modeling, visualization, and supervised learning for exploring and editing data models and mapping at both design-time and run-time.