Practical AI puts Theory into Play for Business and Industry
Practical AI (artificial intelligence) is what happens when the training wheels come off of data science theory. Then, an AI algorithm is applied to a practical situation to solve problems. As a result, AI rides out across the interface between theory and practice, into the world where machines and software interact with humanity.
Scary, right? Not necessarily so.
After all, many of us already have quite a bit of practical artificial intelligence applied to our daily lives. Consider that artificial intelligence describes a data algorithm “trained” or pre-programmed to detect and respond to various defined input and/or user behavior. Then, look around your kitchen, laundry room, on your wrist, that smart device in your hand or sitting on your countertop.
You already have many relationships established with practical artificial intelligence, and more are on the way.
Recently, I sat down and spoke with Tipton Loo, VP of AI Practice at a global IT solutions company. We talked about just what this machine/software/human interface “looks like” when practically applied. Here’s what he had to say.
Consider that the hallmark of true practical AI involves applying techniques like deep learning and neural networks to solve real-world problems in an efficient and automated way. This subset of machine learning enables computers to recognize cognitive functions and interact with humans in, well, a human way.
As a result, cognitive computing permits computers not only to recognize patterns of responses. But also, the computer can differentiate between visual images and sound patterns for applications such as computer vision and speech recognition.
In order to “train” cognitive computers to “think” in this manner, artificial neural networks are created, “much like the neural networks the human brain uses,“ describes Loo. “These networks have various layers, much like an onion.”
As the computer is trained to think by itself, “the required cognitive logic is built into the layers of the neural network. Then, the neural network can be applied to locate, categorize and combine associations for decision making,” says Loo.
Consumers benefit from data science application of artificial intelligence.
In addition, Loo feels that “transfer learning takes what people have done before and allows them to apply these models to their specific use cases.” In other words, why reinvent the machine-learning coding wheel for each new application?
One type of neural network, convolutional neural networks, is used for image recognition. Recently, Loo’s team worked on a multi-factor authentication application using facial recognition. This application can be employed on many practical use cases from banking (ATMs) to retail.
His team started by leveraging and customizing an existing, open-source image recognition classifier model. Then, they developed an application which utilizes a video camera to capture, register and recognize faces to match their respective IDs.
As a result, the smart ATM use case was used as an example of where the face could be matched to a debit card and PIN identifier. Furthermore, this application was deployed on a low-powered device. Consequently, Loo mentioned, “this AI was practical, affordable and easily distributable across ATMs.”
Why does application of practical artificial intelligence matter?
Well, the next time you use an ATM, you want that machine to recognize your face and be able to detect subtleties in facial expression. For example, is that person in line behind you friend or foe? Should an emergency response alert be triggered based on granularities and anomalies detected from the image and facial recognition?
In addition, personal assistant devices and online chatbots allow humans to converse with machines through speech recognition and natural language understanding, both of which are a big part of AI. Companies are interested in “what people are typing into a chat box or saying to a machine,” says Loo. “What are the implications from the types of words used to express feedback?”
These AI applications, combined with sentiment analysis of customer service surveys and even social media, are used to gauge consumer perception about a company’s brands or services. As a result, organizations make decisions impacting whether they recalibrate their products and/or build new ones to meet consumer perception and sentiment.
In addition, practical artificial intelligence algorithms do not get “bored.”
Practical AI applications also have solid roles in manufacturing and industrial arenas. Automated application of output from data science plays big in predictive as well as preventive maintenance algorithms used in industrial applications.
Picture this. A factory floor hums incessantly with machine-generated noise. In addition, the often long (or non-existent) periods between machine breakdown or failure are unexciting, although desirable. Humans responsible for maintaining machinery become bored. As a result, they can make mistakes due to drops in vigilance.
In another use case, Loo’s team captured sensor-generated data from smart industrial devices, measuring vibration, temperature and pressure. “First, these data are fed into an algorithm that detects anomalies. Then, that algorithm allows AI to pre-empt machine failures. If ‘something’ doesn’t look or sound right, the AI responds appropriately.”
An AI algorithm happily, predictably, consistently and unfailingly continues to perform the functions it was programmed for. According to Loo, “AI is mathematical, primarily based on if-then logic. Algorithms don’t get bored. Also, these algorithms can perform functions at scale. There’s no overtime involved. As a result, they are more precise and cost-effective solutions.”
Practical artificial intelligence as an anthropology to it.
Not only is a Fourth, and potentially Fifth, industrial revolution emerging as machines, software and humanity converge via practical artificial intelligence. I view convergence within the industrial Internet of Things ecosystem as creating a new form of workplace anthropology. I asked Loo about this concept.
Loo compares current application of AI algorithms in business and industry to the changes automation catalyzed in the textile industry in the 19th century and then again in the 1950’s through 1970’s. The number of weavers grew in the 1800’s, despite the introduction of automated looms, in a large part due to increased demand for now cheaper, more intricately-designed clothing. Later, new jobs such as machine technicians were created.
Ultimately, Loo sees AI as “something positive”, as the technology continues to “disrupt how people work.”
In addition, Loo extrapolates that practical artificial intelligence will “create new jobs, requiring retraining of existing workforces. ”As a result, the workforce, even the hourly workforce, will be trained to apply critical thinking skills into an increasingly augmented, interactive and intelligent work environment.”