Technology is amoral, but the acts technological advances make possible sometimes have moral implications. Facebook is the most famous company whose executives are struggling with the question of how much they owe the product’s users. After all, Russian oligarchs influenced the 2016 U.S. presidential platform by spreading false information to one hundred twenty-six million users via Facebook. The oligarchs didn’t hack the site. They used it exactly as it was intended to be used. Facebook is facing public scrutiny for gathering data on its users, including data necessary for facial recognition technology, without securing or verifying that data. Yet, it isn’t the only company using artificial intelligence (AI) in a way that may have moral implications.
In March of 2019, VAAK INC. a Tokyo startup company backed by Soft Group Corporation, developed VAAKEYE, a video surveillance technology intended to be used to prevent shoplifting. VAAKEYE uses artificial intelligence to create a personality profile for anyone shopping in a grocery store. It analyzes roughly one hundred personal attributes for each person it detects, including one’s face, one’s clothing, and the speed and direction in which one moves. It also monitors behaviors such as concealment, fidgeting, restlessness, frequently picking up and returning products, and holding products that are in poor condition.
According to law enforcement officials, these are all behaviors common to shoplifters. VAAKEYE isn’t designed to detect whether someone has shoplifted. It’s designed to determine whether someone might. The system of artificial intelligence this crime prevention program uses is called a generative adversarial network (GAN). The short version of how it works: In a GAN, the neural network—the part of a computer that gathers information and determines how that information should be categorized—simultaneously performs two competing functions. One part, the generator, generates new data. Another part, the discriminator, determines whether that data belongs to an actual training set.
In the case of VAAKEYE’s crime surveillance program, the generated data is whatever information the system gathered from each shopper it detects. The discriminator determines whether the data gathered from a particular shopper matches the data sets it was trained on. In this case, the discriminator was trained on profiles of known shoplifters. VAAK INC doesn’t recommend that any arrests be made based on the data its system gathers. Instead, the data should be relayed to store employees, who can interrupt potential shoplifters before the shoplifters commit a crime.
Shoplifting costs retail stores thirty-four billion dollars a year in lost sales, so there’s a strong need for the crime to be more effectively prevented. Even when technology isn’t involved, one doesn’t have to successfully steal an item to be vulnerable to shoplifting charges. In many countries, hiding an item in one’s shirt while shopping is sufficient evidence of one’s intent to steal it. If VAAKEYE is potentially more effective than a watchful store employee, and the information it gathers can’t be used to bring criminal charges against a shopper, why isn’t the use of this program universally accepted?
According to its website, VAAK INC’s mission statement is, “[S]olve social problems with artificial intelligence.” Just because the data VAAKEYE gathers doesn’t lead to arrests yet doesn’t mean it never could. In fact, VAAK INC CEO and founder, Ryo Tanaka, told Tech Crunch Japan that he hopes his company’s technology plays a large role in ensuring future shoplifting arrests. Tanaka says, “In the future, we will further strengthen crime detection and prediction and aim to realize a safer society. By increasing arrests based on video surveillance, we want to increase the [deterrent] effect [on] shoplifting.” Unfortunately, generative adversarial networks (GANS) are only as unbiased as the programmers who create their initial data sets. Unlike a surveillance camera, which records whatever images it receives, a generative adversarial network adjusts its profile based on consistencies in the data sets it receives.
In an April 2019 NPR story, for example, Maddy Savage reported that some Swedish companies used artificial intelligence in hiring interviews, hoping to make their hiring practices more egalitarian and diversify their workplaces. In workplaces that were not diverse, however, the initial data sets programmers created for the artificial intelligence interviewers were not diverse either. AIs did not have the necessary information to allow them to recognize minority job candidates as potential employees. A biased employee profile is unfortunate, but a biased shoplifter profile could lead to false accusations.
The program is designed to detect what VAAK INC termed “”suspicious behavior,” but one could argue that “suspicious” is a subjective term. However, the company’s methods are both legal and effective. Tanaka claims video evidence from VAAKEYE led to ten arrests and a seventy-five percent accuracy rating when the surveillance system was placed in Japanese retail stores as part of a trial. Both VAAK INC and the retail industry warn consumers that attempts to protest stores’ use of VAAKEYE technology will not be effective. Instead, consumers should familiarize themselves with how it works and be prepared for any possible consequences of its widespread use.