Artificial Intelligence explained
using eight artworks
A journey through the world of Art to help you understand AI
At first glance, Artificial Intelligence and Art have little in common — except for a few letters. Artificial Intelligence (AI) has been around for only a couple of decades, whereas some works of art have been telling stories and captivating viewers for tens of thousands of years. In this blog, we will use eight artworks to tell a completely different story: the fundamentals of AI.
I’ve always been a firm believer that true experts are to be able to explain their field to anyone who is interested. And given the impact that AI has and will continue to have on our businesses and societies, we need different narratives to reach a wide range of audiences. At Prosus Group, we invest heavily into training ‘AI Translators’, to give everyone in the organization a basic level of AI understanding so that they are able to identify AI opportunities in their field, and know how to act on these opportunities. This is my version of that narrative to explain the basics of AI to a non-technical audience — leveraging my background in art history. When you’re interested in either art, art history or understanding the fundamentals of AI, keep on reading. Disclaimer: due to my roots, you might see a ‘bias’ in the dataset I’ve used towards Dutch and European art!
At Prosus Group, we invest heavily into training ‘AI Translators’, to give everyone in the organization a basic level of AI understanding so that they are able to identify AI opportunities in their field, and know how to act on these opportunities.
1. Invisible enhancement
17th century painter Johannes Vermeer has painted a number of masterpieces that still enchant us today, including The Milkmaid and The Girl with the Pearl Earring. His secret? Allegedly, he used an optical tool called a camera obscura: a small and dark room with a pinhole, through which a scene is projected onto the wall of the room — like a camera without a film. It still takes a skilled artist to transform this projection into an artwork. However, the camera obscura might have helped Vermeer to create aesthetically pleasing paintings.
The same holds for Artificial Intelligence: AI applications are often running in the background to create invisible enhancements. In popular media, AI typically gets depicted as invasive, humanoid and unearthly, like an army of killer robots that will come to take over the world. In reality, AI is usually the opposite of foreign: it’s products that you are already familiar with, but better. It’s a car, but then self-driving. It’s a social media app, but more personalized. It’s farming, but smarter and more efficient.
2. The missing piece
Not all works of art have endured the test of time. From the original decorations of the Egyptian Tomb Chapel of Nebamun, only fragments remain. They offer us glimpses of what it would have been like to walk into the ancient Tomb, but have missing pieces or gaps. We can try to guess what would have been depicted on those missing pieces, based on the rest of the artwork, or on other similar works of art.
The most common application is AI is to make predictions, which you can think of as trying to fill in the missing piece of an artwork. In AI, you use the information that you have (about the past) in order to predict missing information (the future).
For example, you can try to predict which song someone might like based on what you know about this person and what you know about other, similar people. This gets harder when you have less data about this person, or less data about similar people. In the terms of the artwork, filling in the gap is more difficult when you’re only looking at a very small piece of the total artwork or there are no other artworks out there that are similar. The more data you have, the more accurate your predictions become.
In AI, you use the information that you have (about past) in order to predict missing information (about future).
3. Simple things, repeated
Have you ever looked at a pointillist painting up close? It’s all in the name: artists like Georges Seurat, Paul Signac and Camille Pissarro experimented with paintings consisting of a collection of small dots. From a distance, these dots blend into colors and shapes. But the fundamental building blocks of this painting style are very simple: anyone can paint a dot.
At the risk of making AI seem very un-cool, you could say the same about AI algorithms. The mathematics behind AI algorithms are actually very simple: mostly addition and multiplication. The trick is that these steps are repeated many times — billions of times actually for very complex tasks. Luckily, we can use powerful computers to make those computations faster.
4. Copying is allowed
Think: Roman art. What do you see? Likely, you are thinking about marble statues. What if you were told that many of the best known Roman statues are actually copies of Greek statues? Would that make these artworks less impressive, or less effective in their function to make you admire them?
Copying is also very common in the world of AI — though we call it open-source. Data scientists typically don’t write their own algorithms, especially when they work on applying AI in business. When done in the right way, re-using existing algorithms can be a fast-lane to creating impact through AI in your company.
What is different and unique in each application, is the data that is used to train the algorithm as well as the process of bringing that unique data into the right shape for an opensource algorithm. Data scientists tweak certain model parameters to make it suitable for their application and their data. That wasn’t very different for the Roman artists. You often see a tree trunks appearing to support the statue: the original (and lighter) Greek bronze statues didn’t need these fortifications.
5. Don’t be like Van Gogh
Vincent van Gogh is not only famous for his paintings, but also for his tragic life story. The current fame of his works forms a stark contrast with his fame when he was alive. He only sold a few paintings during his lifetime and constantly struggled with poverty, despite his efforts to find potential buyers. It was Van Gogh’s family that successfully promoted his works and managed to exhibit them after Van Gogh’s death.
The relationship between a data scientist and the rest of the organization that she works in, should be the opposite of Van Gogh’s solo journey to find a buyer for his work. Data scientists shouldn’t be working in isolation on algorithms that no-one understands or appreciates, but instead work in close cooperation with the rest of the organization. Internal subject matter experts, trained as AI translators, help identify opportunities for predictions or AI-driven automation. Product or project managers lead teams that integrate AI tools to create business or customer impact. Software engineers help to scale AI algorithms to work real-time and thousands or millions of times per second. Creating impact through AI takes much more than an algorithm — even though that’s typically the part of the field that gets most attention. It takes a full organization to efficiently and effectively identify customer and business problems — and build solutions that work.
It takes a full organization to efficiently and effectively identify customer and business problems — and build solutions that work.
6. Change is essential
and photograph of the painting, ca. 1935, with additions (left)
If you were to travel back in time to the moment that your favorite work of art was painted, you might be very surprised by what you see (by many things, obviously, though I’m talking about the painting here). Over the years, many of the world’s most famous works of art have been partially or fully overpainted.
The most obvious reason for overpainting is to repair damages and discolorings. Like paintings, AI models need to be kept in good condition. They are adjusted and ‘restored’ regularly, for example because the underlying customer behavior has changed. This typically means that models need to be re-learn based on more recent data, which is called re-training.
Some overpaintings are arguably no improvements. Over time, one of Rembrandt’s self-portraits accumulated a big hat and long curls. This is also a risk for AI models: after retraining, you need to check whether the model still produces the results you initially intended.
7. The paradigm shift
For the majority of the 19th century, artists were painting realistic, life-like scenes. Fast-forward to the early 20th century and you see a wide range of abstract, impressionist and cubist artists in style. What happened in the meantime that caused this shift away from realistic art? Photography.
These major changes in the artistic movements cannot be attributed to photography alone. But what’s clear is that there’s no return to the era before photography — realistic paintings as ‘visual records’ no longer have the same appeal as they used to, before photography was invented.
We cannot attribute the current high time of AI to the abundance of data and advances in computation power and AI algorithms alone. Though again what is clear, is that there’s no return to an era without AI. Changes such as personalization and automation are here to stay — and will only become more common over time.
There’s no return to an era without AI.
8. Part science, part…
Some artworks trigger a feeling of ‘I could have done that’ — or in more extreme cases, ‘my dog could have painted this’. Without insulting your artistic skills, I would dare to argue that this is typically not true. What has made some of these seemingly simple artworks so successful and famous is a complex dynamic involving timing, reputation and originality.
As someone who appreciates the value of making AI sound simple, I have to acknowledge that not everything about AI actually is simple. We’re not yet at the point that just anyone can build an AI model, even though companies are investing in democratizing AI. At Prosus Group, we are also working hard to scale AI knowledge and we have developed amazing educational programs in-house for different roles and levels — from product managers to executives. Until we do get to the stage where anyone can train an AI algorithm, our AI teams will continue to balance science, entrepreneurship… and a bit of art.
I’d like to thank Nishikant Dhanuka and Zulkuf Genc, my colleagues from the Prosus AI team, for their suggestions and help in editing. If you’re interested in reading more on the topic of Art and AI, I can recommend sources such as this blog. There are also many great free resources available to continue learning about AI, such as the Elements of AI course. Also if there are any further questions or suggestions, feel free to reach out to us at datascience@prosus.com. Enjoy your learning journey!
