“I’m sorry Dave, I’m afraid I can’t do that.”
This seemingly innocuous reply by a piece of artificial intelligence (AI) known affectionately as HAL, from 1968’s 2001: A Space Odyssey, planted the seed of a story that would be retold over and over in pop culture: a terrifying vision of the future, in which robots conclude that human intelligence is obsolete. AI has been a part of our social consciousness for so long now that it’s not only taken for granted, it’s also widely misunderstood. On the one hand, many of our current technologies were shaped or inspired by the last 50 years of science fiction — on the other, much that has been envisioned is still far beyond the realm of possibility. The flip phones of the 90s may have borne a striking resemblance to Captain Kirk’s communicator, but I suspect that neither myself nor even my great, great granddaughter will ever be “beamed” anywhere. Even with the exponential technological growth of the last half century, our creative minds consistently transcend what’s possible — but it’s exactly this creativity that drives us to discover ways to make our dreams reality. Nowhere is this progression more obvious than in AI and machine learning, an area that’s seen huge growth as a result of the “Big Data” invigoration of the last 15 years. Machine learning — and within that, Deep Learning — has the very real potential to take us somewhere brand new, enabled by the ability to capture trillions of data points via an Internet of Things and unprecedented computational power and data access speed. However, before we start asking our Amazon Alexa for investment advice, we should understand what AI (and more specialized subgroups) actually represents.
AI, Machine Learning, and Deep Learning
At its simplest, AI is the ability for a machine to mimic intelligent human behavior. Speaking more technically, it’s when a machine is capable of processing its external environment (abstract or physical) and taking logical actions based on predefined rules after assessing all the variables. For example, the last time you were stopped at a busy intersection, you probably weren’t aware that you were leaving your safety up to AI. However, the induction loop sensors and the programmed logic that uses them were very much aware of your presence, and were designed to send you safely and efficiently on your way — without the need for human oversight and intervention. The changing of traffic lights at an intersection, which governs hundreds to thousands of cars and human lives every day, is a clear and simple example of how certain aspects of our lives are better governed by “thinking machines” than human intelligence. Machine learning takes it a step further, moving beyond classic programming logic to a place where artificial intelligence can actually make decisions and recommendations based on factors and data analysis far beyond what could have been conceived or programmed by a human being. The current renaissance of machine learning (a term first coined in 1959 by computer gaming pioneer Arthur Samuel, for my fellow Wargames geeks), is a result of the new and growing presence of data points in all aspects of our lives. The effectiveness of virtually all of the key machine learning algorithms (regression, k-nearest neighbor, probability and Bayesian theory, weighting and ranking, parametric and non-parametric estimation, etc.) is directly correlated to the amount of data and computing processing speed available, both of which have undergone a dramatic increase in the past decade.
Perhaps most exciting to me is the concept of deep learning, in which neural networks are constructed to mimic the human brain — processors called “neurons,” which are connected by synapses and are activated based on environmental variables. Amazingly, the original idea for deep learning was also originally conceived of 50 years ago, when the artificial neurons were called “perceptrons.”
Unfortunately, unlike many other visionary concepts where optimism has fueled real solutions, convincing refutations (later proven false) of the perceptron model stymied investment and growth for decades. The happy ending? We’re entering a moment where Deep Learning is being acknowledged as potentially the most exciting facet of the AI ecosphere.
Machine Learning in Supply Chain and Manufacturing
How is machine learning really applicable? Well, speaking to my own realm of expertise, we’ve only scratched the surface of what it can do for manufacturing and supply chain solutions. Data volumes and variabilities that previously couldn’t be captured let alone assessed are now available to help us create and improve processes. Some of the key potential machine learning improvement areas in this realm could include (but aren’t limited to):
Manufacturing operation optimization, including yield improvements, reduced downtime, and increased quality.
Procurement, strategic sourcing, and costs management.
Supply chain planning and forecasting, and material availability.
Improved Overall Equipment Effectiveness (OEE), including finding and isolating the most and least influential variables.
Inventory optimization and entitlement.
Real time monitoring to create intelligent, agile manufacturing and scheduling.
Machine load levels and optimization.
Asset tracking and supply chain visibility.
Predictive maintenance and process visualization, analytics, and APIs (connected factories).
Quality control driven through visual and sound pattern recognition .
One beautifully simple method for creating solutions is a Digital Factory, in which both randomized and planned events and data can be driven to highlight and stress test various use cases. For example, at Endeavor Consulting Group (ECG), we’ve created one using virtual reality, digital IOT sensors, blockchain, and supply chain enterprise systems to devise and model solutions within a realistic but virtual world.
Digital Factories can help us look at anything from a hundred rows of multi-level warehouse space to multiple distribution centers in various time zones and any other conceivable real-world scenario. From there, we’re able to quickly explore and test out solutions, preempting costly mistakes and wasted time.
Where No One Has Gone Before Unlike the apocalyptic visions of science fiction, the real business value in thinking machines lays in driving a blend of “specialized” machine learning with “general” human learning to create innovative solutions to our problems. ECG is excited to be an active leader in this dynamic space, enabling new supply chain and manufacturing paradigms through the use non-human intelligence — we have a long way to go before reaching the final frontier, after all!
Chris Chambers is a Systems Solutions Architect and Managing Partner at ECG | Follow Chris @clchamb13