How Getty and Shutterstock are combating algorithmic bias to make marketing more inclusive
Despite honest efforts to improve diversity and representation in stock imagery, the algorithms behind most image search engines – powered by historical download patterns – can stoke the flames of harmful stereotypes and impede diverse representations from appearing in search results. Now stock image companies are under pressure to change more than just their database of images and videos.
Stock image sites including Getty Images are on a mission to improve diversity and inclusion in stock media
This fall, Google updated the algorithms that power its image search function in an effort to surface more diverse images in search results. The move aligned with a pattern seen at Google and other major tech players of late – a pattern of actions taken to combat the increasingly publicized problem of algorithmic bias.
Image search is a particularly palpable manifestation of algorithmic bias. The algorithms behind most image search engines are based on keywords, engagement and popularity. In other words, the more times an image or video has been downloaded by users historically, the more likely it is to appear at the top of search results for its respective keywords.
Nowhere is this phenomenon more obvious than in stock imagery. And this problem has, of course, shaped the marketing and media sectors more broadly, since news outlets, content creators and marketers often use stock images and videos in their own work.
So how is the industry approaching the challenges of representation, diversity and bias in stock imagery?
‘The beauty of the shades and textures of humanity’
In response to widespread demand for change, many of the top stock image sites have introduced new initiatives to improve diversity and representation in their media. Getty Images launched its first diversity-focused collection, the Lean In Collection – created in partnership with Sheryl Sandberg’s Lean In Foundation – in 2014. The collection aimed to increase representation of women and girls in leadership roles. Then in 2019 came Project ShowUs, which, alongside Dove and parent company Unilever, media agency Razorfish and creative collective Girlgaze, sought to expand definitions of beauty. Project ShowUs was also one of the first major initiatives to cast real people rather than models in stock photos; now, the trend is picking up steam across the industry.
Alongside a range of other stock imagery collections – including projects focused on representing Hispanic and Latino people, people of all ages and people with disabilities – Getty Images is creating guidebooks and toolkits that include relevant statistics and best practices for more inclusive depictions. Designed for media and marketing professionals, some of the latest include an LGBTQ+ guidebook created in partnership with Glaad and a new diversity, equity and inclusion toolkit developed alongside Citi, which in October rolled out to 17 markets globally.
The company’s director of creative insights for the Americas, Tristen Norman, says that these efforts speak to a larger philosophy. “We have an ethos at Getty: if you change the image, you can change perceptions. And that’s been our mission ... especially within content creation. We understand how powerful visuals can be in upholding certain ideas about certain communities, and also breaking them down. So we’ve tried to embody that in all of this work.”
And marketing professionals agree that humanistic, diverse representation matters. Ana Ceppi, who serves as senior advisor, Hispanic at Edelman, says that early in her career as a marketer, she rarely saw women portrayed in leadership roles – and even less often saw Latina women depicted as anything other than “maids, nannies or ‘bombshells.’” Even when the marketing and media spaces began to offer women more seats at the table, she says, Latina women were left out of these portrayals. “Nearly every consumer segment – men, women, LGBTGQ+, differently-abled – have been portrayed as stereotypes [and] predictable archetypes,” she says. “The biggest gap in today’s marketing is overlooking intersectionality – the beauty of the shades and textures of humanity where most of us live. Get to know me, understand me, elevate my voice and my story – and see me for me.”
It’s not just the big players such as Getty Images, Shutterstock and Storyblocks that are trying to answer Ceppi’s call to action. In response to growing frustration regarding the lack of diversity in stock imagery, a number of entrepreneurs have launched their own diversity-focused stock image businesses. Some of these include Nappy, dedicated to depictions of brown and Black people, CreateHER Stock, which aims to improve diverse representations of women in stock imagery, and TONL, which promises premium culturally-diverse images.
Getting to the root of the problem: algorithmic bias
Although it’s clear that culturally-conscious, informed representation matters, there are also more fundamental problems at play. Norman admits that even Getty’s efforts to create more diverse depictions and its guidelines for portraying different populations don’t effect real change without behind-the-scenes work. “[The collections and toolkits are] highly external – you know, the bells and whistles. But it’s also forcing us to really interrogate what’s happening within. [We’re] thinking about our keywording, about how our contributors approach things, about who we hire ... and who’s leading these conversations. I’m an Afro-Latina myself and being at the helm of these conversations has been really powerful – versus having someone who might sit outside of the community.”
And while all of these initiatives are bound to help move the needle toward a more diverse, inclusive future for stock imagery, the most palpable change is likely to come from one effort in particular: overhauling the algorithms that underlie stock image search functions. Getty’s various efforts to expand its database with more diverse images offers users more choice when it comes to media that depicts people of varying ethnicities, races, religions, abilities, ages and more; however, relying on representation alone as a driver of change almost guarantees that efforts will fall flat – because the underlying search algorithms, rooted in historical data, remain the same. Even when the pool of images and videos has expanded to include more diverse representations, images of able-bodied white people tend to show up at the top of users’ search results, since those images have been used more often in the past.
“The bias is built in. When historically you’ve been using non-diverse images, they will continue to be preferred by the algorithm,” says Julia Muller, associate director of search at B2B marketing agency Just Global. The solution, according to Muller? Changing the algorithms entirely, even if it means sacrificing short-term performance. “Performance is king when it comes to algorithmic smart bidding. To disrupt this, algorithms can be adjusted to prioritize diversity at the expense of a short-term performance drop to accumulate data to bid more effectively and improve performance going forward.”
Arielle Garcia, chief privacy officer at UM Worldwide, has written about algorithmic biases for The Drum previously. She spells out what a new paradigm might look like: “As with any engagement-based algorithm, there need to be controls in place to prevent the perpetuation and reinforcement of bias. The more that people have historically searched for or interacted with images that reinforce stereotypes, the more those images will be surfaced by the search algorithms. One example of how this might be mitigated is to have an additional algorithmic layer: one that surfaces a balanced distribution of imagery that reflects the makeup of society – surfacing images that depict people of different genders, ages, races and ethnicities, for example. The practical effect of this would be that when someone searches for stock images depicting ‘nice hair,’ ‘bride’ or ‘firefighters,’ they would see diverse and representative image results that don’t reinforce previously-held stereotypes in word associations.”
But Garcia says true accountability requires even more than updated, anti-bias algorithms. “Stock image and image search sites have an opportunity to drive meaningful change through transparency and accountability – for example, by monitoring and reporting on interactions and downloads against diverse and representative imagery, by reviewing their image tagging for bias, analyzing the word associations in their algorithm training data and having a mechanism to review uploaded images associated with sensitive tags.”
Shutterstock, another major player in the stock images space, is especially focused on its tagging process. Meghan Schoen, chief product officer at Shutterstock, tells The Drum: “Our global contributors play a large role in ensuring our search function is diverse and inclusive. When they upload content to our platform, they are responsible for tagging their own keywords and search terms. They bring diversity to our content from the way they look at the world, their cultures, language and background. A big focus for us is to educate these contributors on the value of inclusive language. Educating people helps us to educate our artificial intelligence (AI) tools and bridge the gap between technology and creativity.” She explains that, more broadly, the company is “constantly reviewing and updating our AI search function to ensure we combat any historic offensive terms and potential biases that may be built into the platform.”
Moving beyond representation
As far as Getty is concerned, the company’s head of data science Andrea Gagliano admits that, as with other image search engines, when a user searches for and downloads images that reinforce existing stereotypes on Getty or iStock, that data is aggregated and used to inform search results for many users. And though it’s “a technically hard challenge” to disrupt these patterns to combat bias, Gagliano says the company feels “a responsibility” to create change.
How the company goes about catalyzing change, however, varies based on a number of factors including the needs, culture and history of different regions. “We really need to be considerate of [perspectives] within our society and across the world ... within different regions. Our search results are always changing every single day and they’re different for each region, so based off of both historical patterns in any given day, as well as new and incoming content that we need to expose, these initiatives are really customer-driven. We use customer research to inform what we do and use our local offices globally to really understand what that community and region really needs when we’re talking about diversity and inclusion in our search results.”
Ultimately, all the effort, Norman argues, only propels lasting positive changes if marketers and media organizations move beyond representation and seek to deepen their understanding of diverse and underrepresented populations – and include and uplift them. “If we’re only looking at representation or diversity in isolation, we’re missing such a large part of the picture. What’s amazing about having a diverse representation of races or ethnicities, gender identities, sexual orientations, ages, bodies, abilities, religions and so much more is that it creates a beautiful mosaic of differences worth celebrating,” she says.
Unfortunately, she says, many brands are still “slotting underrepresented people or communities into their work without recognizing – let alone amplifying – these differences or the nuances that make each aspect of identity unique.” This approach can result in furthering stereotypes and tokenization. “Where we want to get to instead is true inclusion,” she says. “We must create space for people to show up in their truths, whatever that might look like, and leave our assumptions about what is ‘aspirational’ or ‘appeals to the masses’ at the door. The world is clamoring for that specificity ... and as an industry we need to show up and do our part to make that happen.”