The annoying thing about unconscious bias is, well, we’re not conscious of it. So the first thing we did was hold up a mirror to ourselves. We looked at the images we’d downloaded over the last 10, 5 and 2 years and had a massive wake up call. And it turns out the whole industry needs one too.

For example, according to the Unstereotype Alliance’s last report, just 22% of ads feature people from a mixture of ethnic origins.

Why is this still happening?

It’s estimated that 35% of the images agencies put into culture are from image libraries. We discovered that image libraries will give you your own personal algorithm. It serves you more of what it thinks you like. So if you are unconsciously choosing stereotypical images, it serves them again.

So we created Reflection. On a budget of just £80k.

The plain fact is the unbiased images are there, we as an industry just aren’t choosing them.

But whether you are a huge agency network or a single freelancer, Reflection can help.

Far from being a judgmental tool, it simply reflects back the choices you’ve previously made over the last 5 years, and scores them against the current demographic of your nearest large city.

This measure is called The Reflection Metric.

You can improve your metric with every choice that gets you closer to your nearest city’s demographic.

Of course, Reflection is still early stage tech. But even in this first iteration, in its simplest form, the impact could be huge.

Because imagine if we all started using Reflection?

We could all be reflecting society as it really is. In the images we have the responsibility of putting into culture every day.

We want Reflection to be the tool that changes the entire communications industry, globally.

Whether you work in advertising, design, media, film, publishing or content, whether you are a global chief creative officer or a one-person design studio, we want Reflection to be as ubiquitous as Photoshop. With an annual subscription, users will be able to open up powerful diagnostic tools that allow global companies and agencies to identify the bias in their media output in both moving and still image.

And we want Reflection to grow as well. We need the AI to evolve and learn to recognise disabled and transgender people as well as individual ethic groups that are relevant to specific countries - a ‘reflection’ of society in India or Japan looks very different to what it does in the west.

Globally, we want the reflection score to be the recognised metric that for D&I imagery standards.

1.

Reflection is a Google Chrome Plug in. It analyses your last 5 years of image library downloads using proprietary AI. Detected faces are then classified for gender, age and ethnicity based on FairFace model published by Karkkainen et al.

2.

We wanted to track images – but not be intrusive– so we whitelisted a number of image sites and performed our checks only on those.

3. We reflected user’s choices back to them by showing the images they’ve selected – and storing a small low-res thumbnail of each image.

4. Users now had access to their own personalised dashboard. Here they could see their own Reflection Metric. This is calculated by comparing their image choice history to the demographic make-up of their nearest large city – based on sex, age and ethnicity group. The aim is to improve it with every new image choice they make from that point.

5. The tool also prompts the user at the moment of image search – suggesting terms to better reflect their nearest large city and improve their Reflection Metric.

6. Tested in three markets over a 6 weeks and saw individual Reflections Metrics dramatically increase, collectively moving from 66/100 to 91/100.