Digital native: Face off

If you try the Google Art and Culture App, you will be depicted in frames created by your colonial masters.

artificial intelligence, AI, Google Art and Culture app, digital photography, Google, facial recognition

Can we teach AI not to be racist? That depends on the database it is learning from. (Image Source: Thinkstock)

You MIGHT have seen people posting their selfies with their art dopplegangers, and have been dying to check out who you might resemble from the annals of art archives. Google’s Art and Culture App, which shot to the top of iTunes and Play Store recently, came out with one of the most fun applications of selfies to entice people into exploring art. It allows people to upload a selfie, and then zooms across the cultural databases Google has access to, and finds your lookalike, picked from the vast history of portrait art. The results have been hilarious, surprising, and exciting; and, this is probably the first time that a meme has sent millennials looking at art, which would otherwise have been shrouded in museums, without more than a glance.

There are lots of positive things about this feature: it uses the most common genre of digital photography and makes it a gateway to explore visual art cultures, which, otherwise, remain inaccessible due to the cultural weight of art history and museums. It shows the potential for customisation of art, where your own face becomes the entry point into the world of portrait-making across centuries. It exploits the biometric facial recognition features which have now become commonplace on social media and digital cellphones to open up a new way of thinking about our faces, and how we identify them. It encourages people to start recognising that the selfies might look like a new form, but they have fascinating histories of how we have seen, documented, and celebrated faces through time. The app has clear pedagogic, exploratory, and memetic values fuelled by human curiosity — and is enriched by big-data-driven predictive algorithms.

Yet, the app’s feature has also led to some uncomfortable conversations. Right now, the access that the app has to the cultural databases is limited to white, Anglo-European data sets. Which means, that if you have a face that mimics the racial characteristic features identifiable as Caucasian, the matches are quite phenomenal as you see your face travel over time through different genres. However, if you happen to be non-white, then the app immediately starts struggling with your selfie. If you are a non-white person but light-skinned, you will still be matched to portraits of European faces. If you are any kind of East Asian, the app will perpetuate the stereotype of how “they all look the same” and keep on giving a limited number of choices as your match. So, even if you think that you are as different as chalk and cheese from the person next to you on the road, the app might just think that both of you look the same. And, if you try this app as an Indian, you will find that unlike the glorified images of Renaissance artists, you are being depicted in the stereotypical frames that were created by our erstwhile colonial masters — you will see yourself largely through their eyes.

It is easy to scream racism at this Google app, because it obviously does predictive matching based on racial profiling of faces. However, the app itself is just a symptom of a much larger problem in machine learning: the question of the database. The problem with the app’s “racist” matches is not that it was designed to be racist, but that the corpus of data that it is exposed to — in order to learn and form these predictions — is limited to certain centres of power. It reminds us that artificial intelligence is only as good as the data set that it learns from, and if the datasets that are used as a standard are predominantly from the first world, it is not going to be a surprise that the AI will falter when faced with non-standard faces.

Even as the world turns digital, this app is a stark reminder of the huge inequalities of data-modernity that map our globe in a new way. It is important to realise that digital India thus, would not mean just producing these apps and techs in the country, but also claiming ownership and stakes in building diverse global databases that represent us in fairer ways. It also shows why the effort of digitisation is not merely in producing users and building connecting infrastructure; but, in digitising and carefully curating the vast corpus of historical archives — textual and visual — into databases that are going to inform and structure the AI of the future. Google’s app will slowly roll out for a global usage, but the problem of racist algorithms is not going to be solved by a tweak. It is going to need structural attention at building databases that are inclusive, diverse, and representative of the multiplicity that our digital futures need.

Nishant Shah is a professor of new media and the co-founder of The Centre for Internet & Society, Bangalore.

Tags:
Eye 2018