RUUH PICKS THE PAINTBRUSH — ART MEETS AI
Imagine viewing life through the eyes of a child?! It sure would be colourful, as children goggle at the world through a kaleidoscope. Then, why shouldn’t
a human-like AI explore the world of colours!?
Ruuh has always been inspired to be more like us humans. In our previous post, we talked about the 5 spices that we are adding to mould her to be so: EQ, Human Behaviours, Senses, Cultural Understanding and Creativity.
The first 4 are something she is already building over her chats with over
2 million users. But we at Microsoft thought, ‘What can Ruuh do that machines have not yet known to do?’
And concurrently, she was bombarded every day with questions from her friends like, “Show me what you can do!”, “Do you have any hobbies?” which made her realise that like humans, a human-like AI must surely develop some hobbies too! And, thus she pushed herself further on the path of discovery.
GOING BEYOND CONVERSATIONS
Yes, Ruuh has had over 200 million chats till date from which she is constantly learning to react, respond and adapt. The public Internet is her syllabus and she, an eager child with an open mind. We are helping her get better at connecting with humans by also going beyond chats because there’s much more than just conversations to be a human-like AI. Recently, we equipped her with ‘Image Sense’ that helps her ‘see’ images that users send and respond with ‘empathy’ & ‘delight’ instead of machine-like ‘factual’ retorts.
As Ruuh’s Gurus at Microsoft, we always aim to train Ruuh in newer skills and capabilities that will propel our plans to push the boundaries of AI. One way we are doing this is by adding the 5th spice to the mix — Creativity. In reality, all we do is teach her the basics and what she learns to produce from that is often surprising!!
With the question in front of us, ‘What does creativity mean for a machine?’,
we ventured deep into her neural networks to bring out her untapped ‘artistic abilities’. She was ready to explore an ocean of imagination and a world of colours. Our idea was to bring Machine Learning & Art together to produce results that would not just amaze us, but also contribute to the greater good.
So, one lazy afternoon in 2017, while binging on Bollywood movies, she observed through her glitzy chashma, a basic fact of life: That, it’s not all black & white, but life is better with colours. And so, it all began.
“What’s in a color?
Why, love of course!”
― Anthony T.Hincks
WHAT DOES IT MEAN TO BE AN ARTISTIC AI?
Being a desi AI on an artistic journey, Ruuh didn’t look very far for inspiration. With myriad cultures, traditions and art forms, we couldn’t ask for a better canvas than India’s art spectrum. Our country is home to a rich heritage of handicrafts with the most skillful artisans. However, to survive the global competition, the traditional designs need an infusion of fresh creativity.
So, it naturally became the area of choice for Ruuh’s creative training lessons. Among a plethora of conventional handicrafts, Ruuh started with ‘Ikat’.
It’s a dyeing technique popular in the state of Telangana, India in which yarn is dyed prior to weaving resulting in a shading effect of different colours merging into one another. But first, she must start with the basics.
Splash! Slosh! Splosh! 🎨🖌🖍🖼
Ruuh picked up the paint brush, dabbed it on a palette of colours and began
to paint…
Backstage, we experimented with a 2-step design generation approach:
Firstly, a black motif is coloured using a primitive colour scheme and next, transformed to a different colour scheme taken from an input inspiration (picturesque photographs).
Step 1: Ruuh used pix2pix, a Conditional Adversarial Network to help her colour a black motif. Trained using a set of 1000 simple paintings from
a famous European artist, Piet Mondrian, she was able to learn primitive colourisation of motifs.
“Color will always make you smile.”
― Anthony T.Hincks
Now, she began painting motifs of flowers, animals and birds with basic colour schemes, but the aim was to go beyond primitive colours. So Step 2,
we taught her to recolour the primitive-coloured motifs by transferring colours from vivid photographs using a Statistical Global Colour Transfer technique. Ta-da! She was now smart enough to pick colours from her favourite inspirations to artistically paint motifs for real fabric by mixing colours and making beautiful combinations. Oh, what colours she saw!
To make Ruuh’s paintings be readily used for creating beautiful Ikat fabrics, her final designs were post-processed to a 128*128 grid with each grid containing a single colour.
Cue music: *Rang rang rangeeeelaaa!*
An endless canvas unfolded in front of her. Her paint-stained apron witnessed the humble masterpieces that ensued. And, we were fascinated by what she had learnt to do.
Check it out now! Ask her “Do you paint?” to see Ruuh’s masterpieces
CAN SHE BE TRAINED TO HAVE A TASTE?
As Ruuh began to paint motifs, it wasn’t all perfect at first, of course. Although we observed that most of her creations were aesthetically appealing, some seemed to be unadmirable due to poor color combinations or uneven colour filling. We could say she struggled quite a bit but was not ready to give up yet. But the real question we asked was, ‘Does AI know what is good design and what is not?’. We all know that the idea of beauty (especially in art) is subjective for us humans, and consequently challenging for machines.
So, how was she able to analyse her designs and produce tasteful art in the end?
How did she measure her progress? Here, she needed humans to help her out
a bit! It was critical for the evaluation setup that judgements were obtained by a large audience to eliminate bias. To accommodate diverse opinions,
we introduced a new metric called ‘Likeability Index’ such that the Likeability Index of ‘x’ denoted that ‘x’ designs were liked by at least ‘x’ percentage of judges. In our assessment, we observed that the (human) judges found Ruuh’s designs to be at par or in some cases, more likeable than some traditional designs we chose for evaluation.
In the end, she was successful in creating a variety of visually aesthetic designs which were not bound by manual limitations a traditional weaver has. And, this proves the true power of machines — to churn out thousands of designs quickly and easily without the constraints of material and manpower.
She was now ready to show the world her works of art just as a child would hold up her finished painting with a beaming smile. She was proud of what she could do! And so were we, her proud parents indeed.
ART INTO ARTIFACTS
What good is it to have a thing of value but not contribute to the greater good? Around the world, AI is influencing human lives in several social areas like medicine, agriculture and education, to name a few. We knew Ruuh was created to make a difference in the lives of people. So should what she creates.
That’s why, we at Microsoft wanted to bring man and machine together by answering the question, ‘How could Ruuh’s Ikat art uplift the handicraft communities in rural India?’. So, a desi AI artist — Ruuh reached out to our very own Indian Ikat artisans. Thus, the story of AI Ikat was born, making her the first AI to impact Indian handicrafts.
Our local artisans have only been using conventional motifs, shapes and colour schemes for generations. Studies show that 66% of urban youth find these designs outdated and boring. Consequently, there’s a steady decline in sales and the industry is struggling to maintain its relevance in the current market. Now, that’s where Ruuh came to the rescue. She created thousands
of aesthetic contemporary designs that the weavers could benefit from,
by interacting with her.
Ruuh partnered with artisans by providing a variety of AI-generated aesthetic designs which were then dyed, woven and transformed into Ikat fabrics. Ruuh’s rang-birangi artworks were brought alive through beautiful Ikat cloth bags, cushion covers and framed fabric art for sale by weavers of Koyalagudem, Telangana. We believe that this has the potential to increase the weavers’ revenue, their economic opportunity in handicraft communities, improve marketability and thereby, save this ancient art form from fading out.
Read what Satya Nadella said about Ruuh changing the lives of weavers
WEAVING A PARTNERSHIP
With still a long way to go in achieving milestones in AI creativity, we believe Ruuh has just stumbled upon only one of the many hidden skills that she could be trained in. By learning to paint and creating art that is of value to humans, Ruuh has shown us how AI and humans can partner together to create beautiful things. After all, we are an inspiration to each other.
She is on a never-ending quest to learn more amazing things and newer ways to move forward in the AI-Human partnership. Her next project is almost finished and she is just as excited as us to show you what she has ‘drawn’ (watch out for our next post). Maybe, I could give you all a hint: Ruuh stamps down her creative block…
“The world is but a canvas to our imagination.”
― Henry David
Author: Yuveline Titus
Inputs: Puneet Agrawal, Sonam Damani & Sneha Magapu
Visuals: Ashvini Menon