Machine Learning and Artificial Intelligence
BBC images
This brief was in collaboration with BBC R&D, the creative technology department of the BBC.
The commission was to create a dataset of images that could accurately illustrate the concept of Machine Learning (ML) and Artificial Intelligence (Ai).
Starting point
The images we create would then be used to accompany the news stories or any other media that would need a visual representation of their content.
So we started by looking at the images that the BBC was already using to see what we could do different and/or better.
Like the images on the BBC, the images provided by a simple google search looked very similar and used many tropes linked to technology.
Most of those tropes occult the understanding of those principles rather than explaining how they work. We made the decision to not use any of them.
Visual Research
In a lot of stock images about Ai, you have the letters Ai wrote to make sure that people understand what the image is about. As mentioned before there are also very often zeros and one floating into space. I wanted to use those two tropes in my image, whilst trying to show the human labour behind it. Ai and code don’t appear out of nowhere, they are the work of programmers.
To highlight that I vectored a couple of images of people working at their desks and had them feed the numbers to the Ai in the middle. The letters are filled with a superposition of those workers.
I followed the codes of what an image explaining ML looks like. They are often made of a diagram showing the different steps of the process of teaching a machine. However, there are usually mostly words and arrows but no visual representation of what it looks like.
That’s why I decided to use images of squirrels, they are an animal that everyone would recognise and they have a lot of special features. A human would not think twice about what animal it is if it saw a cat or a squirrel. However, a machine would still have to go through the process of identifying the different features and comparing them to its database.
I decided to make a third image on this topic after thinking about the fact that how ML and AI work, also raises the question of what it is made of. This reminded me of cooking recipes, with all the ingredients that you need to make a dish.
I decided to make a list of the different ingredients, from programmers, code, the actual circuits, training dataset, and so on. Then I thought of using a funnel as a way of showing all those elements mixing to give Ai devices. So at the bottom of the image, I traced an amazon echo, a Google dot, and Siri.
I wanted to show the process of how facial recognition works by showing the matching process between the face and the model. I also wanted to introduce the notion that the system doesn’t see a face but only the mapped grid,
so I decided to not have faces shown.
The faces are from the website This Person Does Not Exist.
Final Images
I created a series of images presenting the process of facial recognition. I decided to take the example of the face id feature present on phones.
I took a picture of myself using my phone and used that
as a base for the image. I added white to act as what
the camera sees during the authentication process. I layered dots on my face, as a visualisation for the grid
of measurements.
The base grid is in blue, as a reference to the blue that images usually have. Then the dots on top are green to signify that the authentication is correct and that the face and the model match.
I chose one of the grids as the base to have all of the other faces compared to. I placed it in the middle and coloured it blue. I animated each grid to come from one side, stop on top of the reference, and then give its place to another face grid.
To make the matching process more evident I decided to highlight the places where the grids matched each time. By doing that, the visual became more visually interesting and revealed more about the process and how it works.
Full image dataset of the group