Artificial intelligence (AI) and machine learning (ML) are complex concepts, and to understand AI/ML, one must grasp the differences between the terminology and various concepts. Many people use terms such as AI, ML, deep learning, and neural networks to describe different aspects of intelligent machine technology. The truth is, there are significant differences between them in what tasks they perform and how they perform those tasks.
Understanding the relationship between AI and ML and human decision-making processes, and providing examples, will help explain how AI is being extended and applied to industrial fields.
01
The difference between artificial intelligence and machine learning
Artificial intelligence is the ability of machines to make decisions just like humans do. Machines can handle recurring situations and choose to handle them in different ways, even when the situations appear to be the same each time.
For example, consider choosing a driving route. My parents recently moved to a town in a neighboring county. When I visit them, I have two different routes to choose from. Each route has its advantages and disadvantages. I choose a route based on my perception of these advantages and disadvantages, filtered according to my thoughts and feelings at the time. Similarly, good artificial intelligence can "consider" data beyond sensor readings and machine operating conditions in order to operate intelligently and efficiently.
The difference between AI and ML is that the former presents the actual intelligent decision-making process, while the latter is the process by which machines collect information to make decisions and provide information for the decision-making process.
Just like humans, machines must collect data and draw conclusions before they can make decisions. This takes time to collect enough data representative of the running system and time to analyze that data to arrive at a result. Data collection involves trying different operating parameters to see how they affect the outcome. This is called “training the machine.”
Data analysis is often described as "reading the story in the data" and then encapsulating that information into simple conclusions/rules for later reference. This is a simplified version of machine learning.
Using driving as an analogy again, after several drives, I chose two decent routes. Route A is a pleasant but winding country road, so it's not the fastest way to my parents' house. However, the travel time is around 60 minutes, varying only slightly each time, rarely by more than a few minutes whether it's faster or slower. Route B is a straight highway, generally much faster, but traffic jams and traffic lights affect travel time. Other route differences might influence the choice. If an autonomous car were to execute ML rules on my route, it would read the following information from the collected data.
The driving time for Route A remains almost constant. If my goal is to pick up my mom for an appointment and make sure she arrives on time, then this is the best option. However, there are no services along the way, so if I run out of gas, would I take the route without gas stations? If I'm stressed, would I prefer to go straight home or try to grab a bite to eat along the way? If my dad is in trouble, would I choose the potentially faster Route B? Another piece of information about Route B is that the driving time varies at different times of the day.
ML sifts through all the data in the process, analyzes it in multiple ways, and finds the rules or correlations behind these "stories".
▲ Self-driving cars will use machine learning to analyze trip data and select the best route.
A biodiesel production system is an example of machine learning (ML) in industrial applications. If ML is incorporated into the process of processing soybeans into biodiesel fuel, both quality and efficiency will be improved. This is achieved by establishing a set of ML rules that take into account humidity, moisture content, temperature, and even soil chemistry (if available). The machine will refine its processes as it operates, thereby continuously increasing output.
02
The difference between deep learning and machine learning
ML is the development of a simple set of rules for making decisions about a process. The interactions between data points involved, where there is a significant correlation between conditions and responses, are far more complex than the rules in ML.
These complex interactions involve a "hidden layer." Hidden layer interactions mean that decisions don't always go from problem to output, but rather encounter other problems or conditions along the way from problem to execution. For example, I need to get to my mom's house on time (but I'm low on gas, or my car isn't in good condition; would I risk breaking down in a remote area? I'm stressed; the hills and fields would be peaceful, but driving past a milkshake shop would make me happier), so I ultimately chose route B.
This hidden layer is the difference between neural networks/deep learning and machine learning. This decision-making process is similar to the interaction of neurons in the human brain.
▲Figure: The hidden layer needs to consider multiple inputs to make a decision, so it does not necessarily follow the fastest path.
03
Applications of deep learning and neural networks
The difference between deep learning and neural networks lies in the depth of the hidden layers. Generally speaking, the hidden layers of neural networks are much shallower than those of systems implementing deep learning, while deep learning systems can have many hidden layers.
Francisco Alcala, an automation engineer at CDM Smith, gave an example of the application of deep learning/neural networks in facial recognition. Even if someone is wearing glasses, sunglasses, has a beard, or hasn't seen them since high school, people can still recognize their face; this is the result of the interaction of hidden layers in a neural network or deep learning system.
Visual recognition is one of the driving forces behind the development of deep learning models. Facial recognition has important applications in security and access control. Identifying the color of labels, containers, or products in high-speed manufacturing environments can impact quality and reduce waste.
Alcala has extensive experience in water supply/wastewater treatment. He developed a Supervisory Control and Data Acquisition (SCADA) program to train water supply systems to better manage pumps and energy to meet demand. The SCADA program can also improve water analysis by predicting and preventing subsequent events based on the occurrence of certain conditions.
Modern industrial systems and machines are already using AI/ML technologies to make decisions, and these decisions are becoming increasingly complex. By understanding the fundamental concepts behind AI/ML, control engineers will have the building blocks needed to implement AI/ML, enabling machines to use available data to operate more efficiently and improve operations.
