Real Applications of Machine Learning
So everyone has heard of Machine Learning and Artificial Intelligence right? It’s all over the trade shows and the news, Google even started giving AI driven web experiences preferential treatment. What no one simply spells out in plain English is business cases that Machine Learning can tackle. This article will remedy some of that and help you determine if a Machine Learning solution is right for you.
Chat Bots
Chat Bots were one of the first applications of Machine Learning to really take off with consumers. With the ability to reduce the percentage of messages sent to help desks they are extremely popular with E-Commerce platforms. Thanks to companies like Google we now have tools like DialogFlow that make deploying a chat bot easy. These are quick to configure for sending or receiving Phone Calls, Instant Messages, Emails, Slack Messages, etc.
The end benefit that Chat Bots can offer your company vastly depends upon many unique conditions. Before purchasing one consider the current costs of your help desk. If you only have to pay less than a few thousand a month it may not be worth the investment. However if the need is simple Chat Bots can be surprisingly affordable and easy to use.
Optical Character Recognition (OCR)
Optical Character Recognition is the fine art of converting text in an image into something the computer can understand. That text can then be used anywhere within your digital systems. Older companies with years of paper logs find will find this particularly useful, imagine how much data is sitting unused!
Teapot Inc. is a simple example case, they manufacture Tea Pots. Imagine now that they have used paper logs to track hourly production for the last twenty years! Previously to update to a digital system and take advantage of advanced analytical systems they would have had to pay someone to manually enter all that data. Now they can scan all the documents and automatically extract all the useful data simply and efficiently.
The primary difficulty with this kind of work is finding trained developers and a reasonably accurate OCR platform. Google Cloud Vision has the best accuracy, as well it is affordable to process documents by the millions. Given a mid sized investment and a good team you can now access years of previously untapped data. The main cost driver is complexity of the form structure and the number of documents to process.
Advanced Analytics
So what does advanced even mean here? Programmers are better at analytics since adding Deep Learning to the AI tool chest. The end use case comes in the power of DL to predict future outcomes. This is valuable in our previous example where we downloaded twenty years of production data for our companies assembly lines. Now we can predict how many teapots will be made in 2020. With a little work we can even see any places where we might be lagging behind the ideal.
Say for example the data from the last 20 years indicates increasing teapot production by an average of 2% / quarter for the first 10 years consistently. However in year 11 suddenly that number flat-lined to 0.5% / quarter! This would allow you to go back and figure out what changed in year 11 to reduce growth. (My theory is they stopped giving employees tea breaks.) Far more complicated analysis can be performed to determine any KPI’s your organization is tracking, all we need is input data.
Advanced Analytics can provide nearly endless benefits so long as your organization is at the stage where they are ready to invest heavily to bring this kind of system on-board. Sadly because the technology is so new no real standard for each industries data has been established. This means almost all truly valuable insights to be gained here have to come from contracting highly experienced data scientists.
Quality Assurance
I’m glad you’ve stayed with me this long. You’re about to learn about one of the coolest use cases for AI I’ve encountered thus far. Lets imagine that on those assembly lines before we have a multi step process to make our teapot. The first step uses heat / pressure to mold aluminum into a teapot shape, step two adds a lid, etc.
Now imagine that the factory got too cold in step one. This would cause the teapot to fracture completely later in step six. Normally this would simply be factored into the final QA before shipping them out. However now we’re spending all that money taking defunct teapots from step one all the way to step six. Worse still we have to pay someone to throw them away!
The solution is to deploy a tiny sensor into each step in the process with a camera, thermometer, barometer, etc. With that data we can train a Neural Network to alarm in step one when there is a defect. Given a robotic arm it can even automatically remove the defective teapot, suddenly Teapot Inc. is saving millions in QA!
Getting this kind of system set up takes on average two years to fully tune and get accurate. Over the following decade it would then save millions in damaged products, man hours, and raw materials. Make sure before hiring someone that they have experience interfacing with industrial machinery, the amount of data produced is staggering.
Industrial Automation
Many industries have process’ that require specialist human workers to stare at gauges and respond to changes in the system. AI promises to free those workers for more creativity driven tasks, all while offering greater reliability and safety. Imagine a factory that handles the heat treating of industrial tools (heating massive hunks of steel to thousands of degrees and rapidly cooling it to add strength.)
To automate the entire process of heating and cooling the pieces all while monitoring humidity and preventing any pieces from shattering would normally take years for a human to master. By gathering data and carefully training a DL Neural Network we can automatically and safely heat treat any type of tool with minimal human intervention.
These sorts of systems are usually highly custom and should only be purchased by larger companies who can spend one to two years testing and onboarding them. However similar to an AI based QA system it can save you hugely over the coming decade. Make sure as you have the patience to have a team fully test and develop a quality solution over years, you really don’t want to skimp here.
Anything you can Imagine
With machine learning is just out of its infancy and into its early childhood, use cases are still being uncovered every day by engineers and researchers worldwide. If your company produces data and needs any task automated more than likely AI can do it now or in the next ten years. We are lucky to have the chance to witness such a massive shift in technology!
