The contributed article was authored by Jonah McIntire of TNX Logistics and Anna Shaposhnikova of Transmetrics. The opinions expressed here are those of the authors, and do not necessarily reflect the editorial policy or outlook of FreightWaves.com.
Artificial intelligence (AI) now powers many real-world applications, from facial recognition, fraud detection, language translators, to assistants like Siri and Alexa. Soon it will be applied to core logistics operations. This should be a golden era of practical AI, when algorithms give way to implementation.
For decision-makers, it is worth understanding some of the basics behind those algorithms to help ensure first experiences with AI in your workplace are likely to succeed. Therefore, the remainder of this article lays out our thinking about two approaches to AI in logistics and where they are most likely to bear fruit.
Two Approaches in Brief
The first approach is what is known as statistical AI, or more popularly as machine learning. It is premised on the idea that a large volume of historical, current and future data has patterns of importance. The software discovers those patterns, with varying degrees of feedback from humans. The patterns, and the models to represent them, act as predictors for data the business finds useful. It is called machine learning because additional experience improves the predictive power of the software.
The second approach is what is known as AI planning. Planning doesn’t necessarily require learning from experience. AI planning is premised on accurately describing the state of the world, the actions available, and goals we want to achieve. With that, AI planning acts as a rational agent trying to achieve goals with allowed decisions and with a model of how the world reacts to them.
Human involvement in AI planning is not to teach the the software, necessarily, but rather to act as a gatekeeper on final decisions. People are said to be in, on, or out of the decision-making loop depending on if they approve decisions, can only halt the process entirely, or are strictly observers.
For completeness, it is worth noting that the two approaches can be mixed. AI planning can include a means to learn from past experience.
In the Logistics Context
Logistics includes areas of application for machine learning and AI planning. To make this concrete, let’s look at two specific use cases in detail. The first is data cleansing. Data quality can be summarized by accuracy, completeness, timeliness and precision. The logistics sector has had mixed success in achieving a consistently high quality of data. That is an issue because so many downstream processes act as multipliers on data quality. Transport planning, customer service, seasonal staffing, inventory forecasts and even safety are impacted. The chronic lack of data quality is not an obstacle to solving an important problem. Sometimes it the most important problem.
Machine learning is a good tool to identify and even correct data quality errors early in the process. In this application, machine learning is predicting the actual data values (or corrections to them) and improves at this through a combination of gaining experience with large historical datasets, combined with human feedback. The result is an increasingly good data quality result despite poor data capture. The models for making these corrections depend on the interactions between data; machine learning makes up for occasional missing or inaccurate data points by learning how those figures should relate to other known data.
Planning is critical in many phases of logistics. From warehouse slotting, pick and pack strategies, dock door and staging usage, transport consolidation and routing, and procurement of all of the above. Given that planning is about decision-making in a known environment to achieve goals, almost all these areas have their own IT systems that just append “planning” to them – labor planning, inventory planning, and transport planning to name a few.
Let’s look specifically at truck planning. In daily truck planning, a decision-maker is faced with a list of loads which must be assigned in some sequence to a list of trucks. Each load and each truck have subtle but crucial variations in their weight, volume, layout, and services such as temperature range, lift gate, and so forth. Sometimes it is also necessary to decide which driver will handle which truck, or even to plan for returnable assets such as pallets or packaging. So there is an almost infinitely large, but knowable, combination of loads that could be assigned to the fleet of trucks each day. And it is clear that a few goals are being sought out via the plans, namely cost reduction and sufficient service level.
This knowable complexity lends itself to AI planning techniques over machine learning. It is an area of modest data volume, structured data, explicit choices and clear goals. AI planning improves on typical planning techniques by exploring the decision space better and faster.
Why it Matters
Executives in logistics need to understand concepts of AI to be savvy buyers. Obviously they don’t make these buying decisions alone, but without some sense of what is available and where its applicable they abandon key technology and business decisions. What we hope that an executive takes from this article is the difference between the two approaches to AI.
Machine learning is suited to problems of large and unstructured data, where greater experience is the primary way to improve outcomes for people or software. AI planning is better applied when the business has clear goals and a scope of decisions made to achieve them.
Also consider what the output needs to be for your problem. Machine learning makes predictions; AI planning makes decisions. By understanding these basics you’ll be better prepared to lead a successful first contact with AI in your workplace.