Page 5 - Logistics News - July August 2023

AI

(2015), quarterly forecasts are off by 13 percent on average. This AI: The machine can think for itself. Mimicking human
translates to considerable lost revenue. Any improvement in intelligence using logic.
forecast accuracy thus translates directly to improvement in the ML: Computers can learn by themselves. Machine enabled
bottom line. to improve at tasks with experience.
DL: The next evolution of ML. Algorithms programmed to
Additionally, according to studies by Gartner, 38 percent of train the machine to perform tasks.
organisations identify forecast accuracy and demand variability
as a key obstacle to achieving supply chain goals and objectives, AI is already in your supply chain
and 70 percent are experiencing moderate to high demand AI might feel like something that has not yet been adopted in
variation compared to the previous year. the supply chain or your industry, but the fact is that AI is already
transforming supply chains all around us (Barlow, 2018). Some
Further challenges with traditional forecasting methods are as examples include:
follows: Rolls Royce: Partnered with Google to develop autonomous
• Unable to extract key patterns and drivers. ships to safely deliver cargo (Forbes, 2018).
• Lack of forecast accuracy in the mid to long term. UPS: Uses an AI-powered tool ORION to create efficient
• Highly dependent on human judgement. routes for its fleet, reducing delivery distance by an estimated
• Limited to analysing historical demand, unable to account for 100 million miles (Forbes, 2018).
external factors. IBM Watson: In 2018, IBM launched Watson Supply Chain
• Unable to model and test ‘what if’ scenarios. Insights: “Watson Supply Chain Insights includes advanced
AI capabilities specifically designed to give supply chain
Demystifying AI professionals greater visibility and insights. Companies can
For many, the mention of AI conjures up visions of robot armies combine and correlate the vast swathes of data they possess
and self-driving cars. Even though robotics and autonomous with Supply Chain Insights and Watson and see the impact of
vehicles are exciting applications of AI, there are numerous external events such as weather and traffic.” (IBM, 2018).
relevant applications of AI algorithms that are decidedly relevant
to supply chain practitioners. In the Fourth Industrial Revolution, organisations are
gathering more data than ever before, but data needs to be
A short history of AI translated into information before it can add significant value.
AI is viewed as a very modern field; however, it has been in Traditional forecasting techniques such as statistics simply
existence since the middle of the 20 century. AI techniques and do not have the ability to leverage the wealth of data we have
th
methods have been developed since the 1940s. The famous access to in today’s business world.
Turing test, to determine if machines could think, was developed
by the mathematician Alan Turing in 1950 and the first Stats and ML: A comparison
academic conference on the subject was held in 1956 (Smith et Some might argue that ML is nothing but glorified statistics
al, 2016). and some might argue that it is a mere subset of statistics. The
truth is that ML stands on the shoulders of giants: mathematics
Although the theories and algorithms underpinning AI have and statistics. ML cannot function without the application
existed for decades, the processing power of computers (GPUs) of statistics; however, machine learning is also not equal to
and availability of data (big data) have only recently caught statistics.
up with the calculation and data-intensive requirements of AI
algorithms. Statistics is defined as follows by Lumen Learning:
“The science of statistics deals with the collection, analysis,
AI, ML and DL: terminology decoded interpretation and presentation of data.” (2019). Thus, by its very
What is broadly known as AI consists of several sub-disciplines definition, statistics is a static representation and interpretation
that are commonly referred to interchangeably. But artificial of data. Statistical methods require data to be fed into hard-
intelligence (AI), machine learning (ML) and deep learning (DL) coded algorithms containing strict predetermined rules, which
are not the same thing. All are part of the same scientific field, generate sets of results that can be interpreted by analysts,
but with definite differences that need to be considered. adjusted and resubmitted to the algorithm for improved results.

www .l o g ist i csn e w s .c o .z a JULY/A U GU S T 2023 3