In 2019, when we first set out to build a company around the idea of making AI approachable to everyone, we knew that AI literacy was a crucial piece of progressing towards that vision. It was then, while we building the first iteration of what would become FieldDay that we started building our brand not only to communicate what our company was about, but as an educational tool to help people grasp some of the core concepts involved in building AI systems.

One important thing to mention here is that we were coming at this with a pragmatic lens on then current, widely available technology. LLMs were still a few years out in becoming mainstream and so we focussed this effort on what you would now consider ‘classic’ machine learning.

While today the concept of RLHF and finetuning are standard vocabulary in AI — terms that even a somewhat interested user of ChatGPT will be familiar with — the idea to make ‘everything teachable’ and to build consumer-grade tooling to make this happen was still a new concept a couple years back.

We were on this mission with the product that would eventually become FieldDay, but we knew that no matter how much work we would do on the product’s design, there would always be a learning curve in adopting the concept of collecting data to get to the outcome you’re looking the algorithm to get you to.

That’s why we decided to embed this educational mission deeply in our brand communications. Nelson Schöller led this effort and helped us bring these concepts to life in a way so that anyone can make sense of them in their mind — while also making them visually appealing enough so that they would have a chance of overpowering the rampant use of electric blue brains and Wall-E style robots in the conversation about AI.

We ended up creating a short video that played front and centre on our website for a couple months as well as a set of [insert actual number] still frames that would be used in presentations and wallpapers within our team for many years to come.

The process of getting there was tough. The reason why poor illustrations of AI are so ubiquitous is because these concepts are by definition abstract and almost impossible to visualise. We decided to try anyway. You can see some of our explorations below, walking a fine line between being too literal — and risking becoming outdated too quickly — or too abstract and getting lost in translation.

One thing that was important to us was that all concepts were illustrated with consistent main characters (the wooden balls and blue cubes) and could seamlessly transition from one another – both for the video but also for easier comprehension by the viewer.

In 2022, we were pleased to see Google DeepMind start a much bigger initiative along the same lines: commissioning artists to illustrate more (and more complex) concepts of AI (and emerging technology) with their ‘Visualising AI’ initiative. Even better, they released all the images created under Creative Commons licence on Unsplash, almost as a subversive act in order to replace bad visualisations of AI step by step.

Inspired by the DeepMind project we decided to put our visualisations on Unsplash as well. You can download them here, or use them through various integrations Unsplash offers in Figma, PowerPoint, or Canva.

The response so far has been great with over 1M views and hundreds of thousands of downloads.