A businessman in a city environment using facial recognition technology to access an app on his smartphone
PHOTO: Shutterstock

In recent years image recognition has been hailed as a critical identifier of digital media to causal customer needs. But can it be an intrusion? 

Marketers worried about its application should pay attention to the facial recognition debates emerging today. The choices can have a downstream impact for martech decisions in which managing image recognition is crucial for customer experience.

What the Debate Is About

On June 9th IBM  CEO Arvind Krisna wrote a letter to Congress announcing that the company will no longer develop facial recognition technology. The concern behind the decision is that facial recognition systems are biased in identifying people of color, creating misuse for mass surveillance and racial profiling.  Shortly after IBM's announcement Microsoft and Amazon promised not to sell facial recognition tech to law enforcement.    

Business and public leaders are seeing the liabilities in how image recognition systems are deployed. The most haunting misapplication to date is the wrongful arrest of Robert Williams, a black man whom Michigan state police misidentified through an image recognition match to a watch thief suspect. To worsen matters, officers arrested him at his home in front of his young daughters and wife. Detroit has experimented with image recognition through its controversial Project Greenlight surveillance program that uses citywide cameras to match still images and video footage to a criminal database.

Facial recognition is a variation of image recognition. Marketers have used image recognition for identifying product images in social media streams to better understand product mentions and to determine quick customer engagement responses. Social media platform have developed shopping features based on the technology. Pinterest Lens, for example, uses image-recognition technology to identify products users discover with a smartphone camera and then purchase. Facial filters used in Snapchat and other mobile apps is another facial recognition example.

The concerns center on how image pixels are processed to identify a person can be questioned. While an image recognition system works for an item consistently, recent studies on facial recognition has revealed an accuracy gap in identifying facial details well. Many such biases exist within the commercial programming applications businesses use today. Researchers Joy Buolamwini and Timnit Gebru released a landmark 2018 study declaring that gender and skin-type bias exist in commercial artificial-intelligence systems.

Related Article:5 Marketing Considerations for Facial Recognition and Eye Tracking Software

What Should Marketers Learn

A key message marketers can take from the facial recognition debate is that brand managers must thoroughly understand how deep learning-based solutions can be misused. Such understanding is vital to tying model details to specifics consequences so that a team can deftly understand the misapplications risks that can harm brand objectives, let alone customers.

Another critical message is that marketers must know that the level of transparency brands need to share regarding image recognition usage can shift. From data drift from training session to wide range of accuracy errors, poor performance within machine learning systems in general can develop over time. Skipping continual verification can diminish messages crafted to customers regarding how privacy, data, and even messaging is being managed. Snapchap’s recent Juneteenth facial filter, criticized for its tone def message for the holiday celebrating emancipation in the United States, is an example of what can be released without a cohesive overview of technical prowess and user experience.

These key messages will impact the choices for a martech stack meant to manage programmatic campaigns. Being able to verify accuracy requires considerable effort. Many programming bugs in a machine learning initiative like image recognition are not obvious to spot, and can take time in identifying where they occur in a training and testing process. Thus martech stacks should be select to enhance a teams’ capacity to develop collaborative data processes like CI/CD and observability. Doing so can implement changes that better align with programmatic dynamics.

Moving forward marketers should be alert for the consequences of recent facial recognition legislation. A few cities, such as San Francisco and Boston, have issued facial recognition bans. In addition a congressional moratorium was introduced to address the topic for federal agencies.

Debate regarding facial recognition systems will certainly continue. Meanwhile brands will certainly continue to use image recognition to identify products and services. Yet it is important to understand the liabilities as well. Understanding the risks can teach marketers the right guard rails needed to prevent a brand from being seen as creepy.