Artificial Intelligence (AI) has been a buzzword in popular culture for some time, at least since the release of the movie Terminator (1984) and perhaps even longer. Much of the buzz surrounding AI recognizes the power of technology to continue improving exponentially until robots become competent enough to start taking over many of the jobs that humans currently do today. In this blog post, I’d like to demystify the current landscape of AI technology to show how its true use may be most powerful in the software that helps run your supply chain and not necessarily in the robots you may employ.
What is Artificial Intelligence?
Computers were originally created to compute. The 1s and 0s behind computational hardware and software allow them to perform discrete calculations for deterministic algorithms very quickly and powerfully. However, the human brain works a bit differently. As Professor Chomsky famously proved, humans are born with specific “hardware” within their brains which aid in the processing of concepts like language and facial recognition, but the brain’s true power lies in the fact that its “software” isn’t all written at birth. The human mind has the power to learn and solve problems through its interactions with the real world and thereby write its own “software.”
To elucidate the difference, take the example of making a peanut butter and jelly sandwich. For a robot in today’s day, a developer would need to instruct the robot to perform every minor step of the process, like exactly how to grip and turn the lid of the peanut butter jar in order to open it, but a human can be shown the process and understand the minor steps implicitly.
Computer architects and developers have been working on trying to create computers that can gain these same features of learning and problem solving, without the need for explicit human instruction. Below is a taste of some of those ways.
Neural Networks
Mimicking the architecture of the human brain, computers can now also be designed with neural networks as well. After generally creating a computer program with a “goal” to achieve within a set of parameters, the program could generate many different solutions to try solving the problem at hand. In simulating these solutions, the program can retain pathways that lead to the goal and discard pathways that do not. After evaluating a large set of solutions, the program could find the most efficient solution that reaches the goal among all the solutions that it tries. For example, Google’s DeepMind used neural network architecture to learn how to walk
Neural networks are great at helping computers with simple recognition. For example, after being trained on a million pictures of a tree, a program could process another picture and correctly identify if a tree was depicted in the picture or not. The only problem with this approach is that a tremendous amount of examples are required for training in order to produce good accuracy.
Machine Learning
The idea of learning is all about reinforcement. In our example of recognizing pictures of trees, if the program were given positive reinforcement for every time it recognized a tree accurately and negative reinforcement every time it recognized a tree erroneously, then it could continue to improve in accuracy the more it comes across additional examples and feedback.
Applications in Supply Chain
One of the most important aspects of managing any supply chain lies in analytics. Supply chain managers and directors need to be able to track their inventory, lead times, throughput, costs, and a whole host of other KPIs in order to ensure their operations are running smoothly. These analytics are typically used to identify issues and inform business decisions for the future. However, the way of the future lies in actionable analytics.
In the warehouse management space, Warehouse Execution Systems (WES) are taking the lead in implementing elements of AI to manage and release work throughout a distribution center in order to maximize the utilization of material handling sorters and decrease overall time to fulfill. We can confidently see the end of requiring manual wave planning to release orders for picking as WES systems become more sophisticated and more tightly coupled with WMS systems in the coming years. For example, waveless execution is already available as the bedrock of the JASCI’s cloud WMS solution, and Manhattan’s latest WMOS version allows customers to choose between traditional waving and waveless execution.
On a larger scale, if AI-powered computer programs are monitoring the decisions your supply chain teams are making when it comes to restocking inventory to meet upcoming customer demand, they could constantly learn which decisions lead to greater results in terms of money or time. These programs would not only start recommending alternatives but would also be able to provide insights on why, taking examples from past performance, until we gain enough confidence in their modeling capabilities to allow them to make decisions on their own. In fact, JDA’s Blue Yonder solution is tackling this challenge as we speak.
Conclusions
Take note that AI is an umbrella term that encompasses many various algorithms and methods which are all suited for unique applications. If AI is used for optimization, the methods used may be completely distinct from AI that’s used for prediction. There are also some AI techniques that require large data sets, as mentioned above, and other methods that can use much smaller data sets to produce results. There is no single AI solution that will do it all.
However, we are seeing AI take the world by storm with the upcoming rise of autonomous vehicles. We will continue to see advances in AI technology in the coming future, with most advances taking place in the realm of software interfacing with humans before they take place in the physical world directly via robotics. In the nascent stages of AI technology, it would be appropriate to stay critical of how much value and accuracy AI can bring to bear on your real-world applications without having enough good real-world data, but don’t blink too long. AI-driven programs are already seeping into the technology that you use every day, and it won’t be long before you expect more than just basic analytics and user-required input from your supply chain solutions.
Contact Bricz to help you prepare for a present that includes artificial intelligence solutions so you can remain ahead of the curve.
Contributor: Ahmed Salim Supply Chain Manager at Bricz