Automation is changing the way customers interact with restaurants. While person-to-person ordering is still prevalent in America, traditional systems are increasingly being replaced by automated ones, especially given the dramatic increase in demand for contactless ordering spurred by the COVID-19 pandemic. One of the next steps in automated ordering is the introduction of artificial intelligence (AI). Think Siri, Alexa, or Echo at the drive through window, taking your order. While this technology is still being developed, the industry isn’t far away from implementing AI ordering systems. While a development like this has many upsides for both restaurants and customers, it also raises a number of concerns the industry needs to address and grapple with before adoption.
One of the most pressing concerns is the tendency of AI technology to learn racial biases and perpetuate discrimination. For most, this concern comes as a surprise. AI is not a person – so how can it hold racial biases, and how can it perpetuate discrimination?
The answer, once you look, is actually very intuitive and simple to grasp. As Richard Sharp explains in his piece Machine Learning Needs Bias Training to Overcome Stereotypes, AI learns about the world by observing it and making inferences. One of the key features of AI is that it mimics – it models its approach to interaction based off of observations of others’ interactions. Biases that exist and affect daily interactions will inevitably translate into a machine learning system’s understanding of human interaction. Therefore, just because an algorithm itself doesn’t use a person’s race or ethnicity as an explicit variable, it can learn to treat people differently based on race and ethnicity through its learned approach to interaction.
In this sense, an AI ordering system and a person-to-person ordering system have this in common: the potential for discrimination. This means that the time and energy companies have devoted to implicit bias training in person-to-person setting must be matched as they make the transition to automated ordering systems. The approach just needs to look different.
In the case of person-to-person ordering, a good example of a company devoting time and resources to implicit bias training is Starbucks. In 2018, two black men were accused of trespassing at a Philadelphia Starbucks and were subsequently arrested when an employee called the police. The incident led the CEO of Starbucks, Kevin Johnson, to publicly apologize and commit to doing better as a company. In the following months, the company implemented a bias training program that entailed the closure of all U.S. locations for four hours so employees could be led in group discussions and trainings focused on identifying implicit biases held by all.
The single-day event was followed by other efforts to train Starbucks employees in preventing instances of discrimination, as the company worked with experts to put together 12 educational modules to be implemented at stores over the course of the year. The modules covered a diverse array of employee-customer interactions, and intra-store interactions, such as those between employees and managers. While there is no perfect approach or training course, Starbucks is an example of a company’s leadership deciding to devote time and resources to bias training, with the assistance of experts and leaders such as Sherrilyn Ifill, the president and director-counsel of the NAACP Legal Defense Fund, who advised Starbucks on their bias training program in 2018.
Transitioning from person-to-person ordering systems to automated ones changes the nature of restaurant-to-customer interactions. The Starbucks bias training modules implemented in 2018 would clearly need to be updated to reflect a restaurant environment with an AI ordering system if Starbucks were to automate their ordering system. That being said, the transition to AI does not mean there is any less need for time and resources devoted to bias training. But instead of only having modules and educational tools, companies need their engineers solving bias and discrimination in AI.
Companies can’t singlehandedly change the world around them – they can’t stop the AI technology they use from observing bias in interactions. However, they can train the AI to recognize bias and not make inferences based on it. This is a technological problem that requires a technological solution. Engineers are capable of doing it – but executives of big restaurant chains in board rooms across the country need to recognize and make investments in it. Executives shouldn’t wait for another incident like the one that happened in Philadelphia in 2018 to get serious about bias training. Ensuring that AI doesn’t perpetuate harmful forms of racial discrimination is an issue that should be addressed before those systems are implemented.