Overview
The novel coronavirus has caused drastic shifts in consumer behavior and purchase patterns. From new bulk orders to new traffic on different platforms, the sudden changes have affected the effectiveness of machine learning for retailers and companies around the world.
Machine learning leverages data from user activities to find patterns and make educated guesses about subsequent activity, and AI machine learning helps guide many of our daily decisions, from helping inform supply chains to recommending next week’s grocery purchases to assisting in cybersecurity.
As toilet paper and cleaning products topped the charts of most-purchased items during the first weeks of the coronavirus outbreak, artificial intelligence (AI) machines raced to catch up and inform suppliers of stocking needs. From YouTube video suggestions to social media location recommendations, AI systems have tried to keep up with the sudden, abnormal coronavirus-related activity.
When New Normal Is Too Abnormal
At CES® 2020, leaders in the AI space noted how the technology has shown itself to be beneficial in many capacities. And during the COVID-19 outbreak, AI has contributed to combatting the effects of the COVID-19 outbreak, including use in diagnosing patients, predicting the spread of the virus and more. However, at the same time, machine-learning algorithms have struggled with demand spikes and irregular activity that AI programs would normally consider to be fraudulent behavior.
According to the MIT Technology Review, an overload of input data that differs too much from the data the models were trained on has made it difficult for the models to respond to and learn from changes.
From a sudden spike in streaming users making recommendations less accurate, to unordinary purchases of stay-at-home baking and gardening tools causing mistakes with credit fraud detection, algorithms were not equipped for consumers’ high-speed crisis response. Compounding the situation was the constant change in what products were in demand: cleaning wipes one week, active dry yeast the next.
Humans Step In
Many companies have had to intervene to modify and adjust their machine learning models, some even anticipating these purchase pattern changes. For example, Amazon had to tweak their algorithms to better distribute orders among sellers.
Advertising companies using machine-learning models to write emails with top buzzwords and phrases had to adjust their algorithms to avoid certain words and phrases and focus on others.
The current challenges have highlighted the need for further machine-learning training when it comes to crises. Because disaster relief response is usually regional, global crises like the COVID-19 outbreak is a new scenario for which AI developers were not prepared.
Major events like past financial crises and stock market crashes can perhaps help data scientists build machine-learning models that can better respond to sudden and worldwide changes. Teaching algorithms to recognize patterns in both stable and more volatile environments can help create more resilient automations for future needs.
As AI helps solve some of the world’s most pressing challenges, the current crisis has shown a potential path for further machine learning growth.