AI/IA in Design Process: An Attempt to Make Data-Aware Creatives

Abstract

This article describes the process and outcomes of a pilot workshop conducted at the International Design for Social Development Conference, Gandhinagar, 2019. The workshop was an attempt to familiarise designers with the use of Artificial Intelligence (AI) and Machine Learning (ML) in their work process. The outcomes of the workshop involved first hand experience in collaboration with ML and the consequently arising inquiries in the dynamics and ethics of collaborating with machines.

Full text

AI/IA in Design Process: An Attempt
to Make Data-Aware Creatives
Yatharth 1, Harshali Paralikar 2
1,2, National Institute of Design-Paldi, Ahmedabad, Gujarat, India
1 yatharth_y@nid.edu, 2 harshali.para@gmail.com
Abstract: This article describes the process and outcomes of a pilot workshop
conducted at the International Design for Social Development Conference,
Gandhinagar, 2019. The workshop was an attempt to familiarise designers with
the use of Artificial Intelligence (AI) and Machine Learning (ML) in their work
process. The outcomes of the workshop involved first hand experience in
collaboration with ML and the consequently arising inquiries in the dynamics
and ethics of collaborating with machines.
Methodology: The workshop sets the context by inviting participants to put
down what they think or know about the terms ‘Artificial Intelligence’ and’
Intelligence Augmentation’. This paves a way for an interactive presentation on
the history of tools and how they can augment intelligence by allowing human
and machine skills to complement each other.
Second part of the workshops aims to demystify the word ‘Artificial
Intelligence’ and introduces various Machine Learning methods through
activities in which the participants learn how to identify and draw a subject
they never encountered before. This allows the participants to draw parallels
between human learning and machine intelligence. The next few activities use
contextual puzzles and games to draw the limitations of similarity between
Machine and Human intelligence in the current state of technology as of 2019.
The next segment of the workshop gives the participants a chance to
collaborate with two different machine learning algorithms. Teaming up in
groups of three, participants playfully negotiate with the AI to imagine and
tangibly prototype an AI+Human designed Chair.
The workshop concludes with participants contextualising the new chairs they
create with the AI, presenting them in an exhibit followed by a roundtable
discussion on data, creativity, ethics and future use of such algorithms in design
processes.
Keywords: Artificial Intelligence, Intelligence Augmentation, Human-Machine
Collaboration, Emerging Design Practises
1 Introduction

Fig. 1. The workshop poster
This article is an open ended enquiry into a future of Design where creators freely collaborate with machines and
augment their creative abilities without loosing the integrity of their Design practise. We present a detailed report
of the ‘AI/ IA in Design’ workshop that was conducted at the International Design for Social Development
Conference at the National Institute of Design, Gandhinagar. The report takes the reader through the context of
the workshop, its intent, conception, execution, outputs, and insights.
2 Background
The National Institute of Design is India’s premier design education, service, and research institution. It was
established in 1961 based on Charles and Ray Eames’ India Report. The historic Ahmedabad Declaration on
Industrial Design for Development-1979 completed its fourth decade in January 2019. The declaration was
jointly drafted by UNIDO and ICSID (now WDO) in close cooperation with NID, Ahmedabad way back in
1979. To commemorate 40 years of this seminal declaration, NID in association with the World Design
Organization (WDO) held a two day International Conference on Design for Social Development on 14th and
15th of February, 2019. The conference revisited the vision enshrined in the declaration and restated it in the
light of Sustainable Development Goals (SDG).
As a part of this conference, the authors conducted the AI/IA in Design workshop which reflected upon the scope
of emerging technologies in Design practises. The workshop pertains to the 9th sustainable development goal of
Industry, Innovation, and Infrastructure. The workshop was also a part of one of the author’s graduation thesis
titled ‘Craft Futures’ which explores how emerging technologies inform crafts and vise-a-versa.
3 Objective
The term artificial intelligence is known to all but understood by few. In the last couple of years, there has been
an increased hype in the possibile capabilities of AI and it taking away jobs and overtaking humans. Both the
authors shared a collective need to weed out this notion and replace it with a truer, experientially informed
understanding of AI amongst people. More importantly, the authors themselves explored machine learning as a
tool and felt that a similar experimentation within the Design community would not only allow for greater
creative possibilities, but would be healthy in the evolution of the profession itself.

The workshop was a pilot attempt to familiarise designers with the use of artificial intelligence in their work
process. In its subtext, the workshop aims to destroy the popular notions of AI and present it in the light of a tool.
The workshop aimed to create AI aware designers who are ready to collaborate with machines treating them as
‘augmentators’, and not terminators.
4 Methodology
The workshop was a three hour long controlled session with 20 to 25 participants.The proceedings of the
workshop were split into three major parts. The contents of these parts have been mentioned here briefly and
later explained in detail.
Part 1 Meaning and history of tools. Introduction to Machines as Tools and Addressing mis-information on AI
Part 2 Interaction with AI to augment Design Process, experiencing different levels of Human-Machine
involvement in design process through two separate team activities.
Part 3 Reflection and group learning, discussing applications of AI in individual projects, application in themes
of conference, reflecting on implications and ethics of AI and Designer collaboration.
5 Tools
Tools used included Slideshow, writing tools (pen, pencil, paper) and model making tools like clay, cardboard,
cloth. An Image Classifier built on top of Ml5js using DarkNet(ImageNet DataSet), Image Generator using
DCGAN in Tensorflow(Training dataset collected from Pinterest)
6 Material
Materials needed for the experiment were: A mobile phone, Pen, Paper, Modelling tools and a Computer.
7 Process
7.1 Part 1: Human Learning and Machine Learning
Introduction to Tools Humans have limited physical and mental abilities. What makes humans more intelligent
than the rest of the animal kingdom is that they’ve figured out a way of doing more than just what they are
naturally capable of. To give simple examples, humans don’t have claws or fangs to defend or to offend, to break
hard food food. So the early man augmented his physical abilities with spears and arrows to defend, offend, poke
and pierce. Humans don’t have large working memories to remember and recall information so they augmented
their cognitive abilities with abacuses and writing to document/ record and calculate large chunks of information.
Tools have played a crucial role in the evolution of mankind and in the augmentation of the physical and mental

capabilities of the homo sapien specie. Tools not only compliment us where we lack, they extend our potential.
Tools not only make human life easier, they change the way humans live.
Fig. 2. Marshall McLuhan on media shaping humans
Doug Engelbart, the founding father of the field Human Computer Interaction (HCI) and the inventor of the
mouse, proposed the theory of intelligence augmentation by ‘de-augmenting’ the pencil. He tied a brick to a
pencil thereby making writing with it many times harder. When it is hard to do the low-level parts of writing, it
becomes near impossible to do the higher-level parts of writing: organizing your thoughts, exploring new ideas
and expressions. The message was: a tool doesn’t just make something easier, it allows for new,
previously-impossible ways of thinking, of living, of being. Writing especially wasn’t just a way to record things,
it led to the creation of mathematics, science, history, literary arts, and other pillars of modern civilization.
Fig. 3. Engelbart's de-augmented pencil

Intelligence Augmentation As human intelligence augmented with tools, tools also evolved and became more
sophisticated owing to the growing human intelligence. They became more complex, more connected. They were
able to achieve exponentially more than what humans couldn't. Humans also became more and more intelligent.
Their physical and mental capabilities expanded enormously owing to the sophistication of the tools they created.
The wheel, the pencil, the vehicle, locomotives, computers, the internet. The caveman, the tribal man, the
colonized man, the industrial labourer, the urban citizen- tools have made us who we are, and we in turn have
made them. They empower us, we enhance them. The most recent, most sophisticated tool that everyone is trying
to understand lately, artificial intelligence. But before that we must revisit its predecessors- the personal
computer and the world wide web.
Demystifying AI The personal computer is a machine that fulfills everything an individual can do- hence today,
almost everyone around us, has one personal device. The world wide web goes a step further. It fulfills the search
for information in seconds- something that a lone individual could have done in a far more greater amount of
time. The Artificial Intelligence, takes some more steps further- simply put in continuation to the last example- it
gathers data on what you’ve searched to help you search better.
Fig. 4. Left:A google search of AI, Right: A venn diagram explaining machine learning
So what really is Artificial Intelligence? The word comes in different names pertaining to different techniques
used in its development Deep Learning, Machine Learning, Neural Networks, Genetic Algorithms. For the
purpose of this workshop we will be talking about the Technique called Machine Learning, which has lead to the
recent explosion of “AI” Applications.
Table 1.Getting to know Humans and Machines better.
What are humans good at?
Empathy, Creativity, Compassion, Imagination,
Humour, Common Sense, Reflection/ Introspection
What are computers good at?
Computation, Arithmancy, Repetition
Accuracy, Precision
What are humans bad at?
Computation, Arithmancy, Repetition
Accuracy, Precision
What are computers bad at?
Empathy, Creativity, Compassion, Imagination,
Humour, Common Sense, Reflection/ Introspection

Machine Learning Machine Learning is really simple to understand: It is a machine, which can learn. A
machine which is made to learn by strategically feeding data. Humans are really (really) good at it already. You
know because you are constantly learning something. like right now. If you show a machine many pictures of
cats, it can tell you what a cat is. If you make tell it draw cats, it will make funny drawings but overtime with
right training, it will learn to draw a cat. If you tell it if “billi” is “cat”, then it will learn that. On an even broader
scale its something even simpler, it's something which recognizes patterns. This what we humans do too and are
really good at.
Say, a cat is cat because it has fur, it meows, it has 4 legs, it has whiskers. We know its not a lion because it’s
smaller. We are constantly looking for patterns around, we are really good at this too, its trivial for us to
differentiate between different breed of dogs or read what someone's expression means by looking for what that
configuration of eyebrows, mouth, eyes and the overall posture mean. We learn these patterns overtime with
experience and are born with some. Learning by association
Fig. 5. Machine learning by classification
Summing it up, machine learning is a machine which can learn how to recognize patterns and create new ones.
It's not intelligent. It cannot understand context. It is not going to kill everyone. These are the things human are
good at. It won't take your job ( hopefully ) but it will take up the parts of your job which involve repetitive
engagement with patterns. Machine learning is highly task specific. and it can make mistakes.
7.2 Part 2: Exploring alternative workflows of Design process with ML
The goal in this segment of the workshop was to familiarize participants with use of AI through a small project.
Time alloted was roughly 60 to 80 min to the entire section two. Experiment 1 was followed by Experiment 2.1
and 2.2 which ran parallely

Experiment 1- Basics of Machine learning The goal was to give a revision lesson in how learning happens in
humans and machines. Time alloted was roughly 20 minutes. This was a classification based Machine Learning
experiment.
Fig. 6. Cards from Experiment 1- basics of Machine Learning
Fig. 6. Experiment 1 in action, being conducted by Yatharth
Participants were shown 10 images of Llama and some natural random objects like trees and goats. This was
titled Training Set 1. Each image was labeled. A test was conducted with Test Set 1, which contained 10 more
images added to Training Set 1. This was a mixed dataset with mostly llama, some goats and some alpacas. Test
Set 2 was not labelled. The participants were asked to note down the number counts of each category .
Participants were introduced to Training Set 2 Introduce which included photos of alpaca and llama in alternate.
All were labelled. A second test was conducted using Test Set 2 which was a mix of Test Set 1 and Training Set
2. The participants were asked to make a note the number of counts as they observed in each category again.
In conclusion, participants were asked to draw a llama and alpaca from their own understanding. A discussion
was held on the differences/ similarities in the two an whether the participants could identify these
characteristics. In conclusion, participants were told that this experiment is the basics of Generation.

Fig. 7. Demonstrating how generative models work by drawing Llamas and Alpacas
Concluding Notes The success rate of identifying a llama from an alpaca was better when mistakes
were made in identifying a llama and then corrected. With every correction, new information on the
visual characteristics of a llama was gathered and this made identifying easier.
In the process, the participants also learnt what is NOT a llama, learnt the visual characteristics of the
alternative in the data set, the alpaca. The process aimed to drive home the realization that the brain
may have developed certain biases/ misconceptions as a result of exposure to many images and one’s
preconceived notions. These biases make it difficult and confuse self in the decision making process.
Participants attempted to overcome biases by repeatedly making mistakes and correcting them.
Experiment 2.1: The goal of this experiment is to experience an AI led design process. Resources required are a
camera phone, papers, pencils, pens, and modelling material. The experiment was carried out in a team of 10
people which was split into smaller teams of 2. Duration of the experiment was 45 mins. Through the
experiment, the participants had to find chairiness of things.
[14 |Adapted from: 100percent Chair Project ]

Fig. 8. The Chair Identifying App
The Chair Identifying App was custom developed for this experiment. (The application was built on top of
classifier example from ML5JS which is using the ImageNet dataset, the app would give out the combined
probability of ‘Barber Chair’+’Folding Chair’+’Toilet Seat’+’Rocking Chair’ since the ImageNet dataset does
not cover an overall category of ‘chair’; we will look at the implications of this custom dataset later in the
workshop results).
Each group of two was asked to collect ten images of complete objects that could be a surface to sit on. These
objects had to be made of a solid substance and should be available in the vicinity. These had to not be designed
purposefully for sitting (ie. no stools/ chairs) Participants were asked to gather all collected images in a google
drive folder. A total of 50 images were collected. Each sub-group then ran their 10 images through the Chair
Identifying App. Participants were asked to jot down how much percent chairy each image was. The top five
chairiest images from the 50 collected were selected and then further prototyped by each group.
In a Nutshell The Designer’s role in this experiment was that of collecting and feeding nonsensical, abstract data
to the machine, and developing its design decision. The Machine’s role in this experiment was to take the crucial
call of selecting an appropriate design direction using all the pre-fed data on good design.
The status quo of design power was: Machine 75% Human 25%
Concluding notes Through this experiment, we allowed the machine's discretion to guide our design process.
We fed the machine abstract images and based on its intelligence, it chose the ‘chairiest chair’. We manifested
the machine’s design decision into real prototypes. This way of working may seem extremely counter-intuitive,
stupid even. But it was important to explore this variant of status quo of power with the machine.
Experiment 2.2 The goal of this experiment is to experience a human led led design process. Resources required
are a camera phone, papers, pencils, pens, and modelling material. The experiment was carried out in a team of
10 people which was split into smaller teams of 2. Duration of the experiment was 45 mins.
[13 |Adapted from:The ChAIr Project ]
Each group of two was asked to collect 10 images of what they felt were ‘nice chairs’ from their vicinity. There
would be a total of 50 images, which then were collected in a google drive. These images were then plugged in
those images in a data set that feeds the images to a program.
This program will create permutational iterations of those images to give possible representations of new chairs.
The groups were asked to pick the top >5 ‘chairiest chairs’ and based on their discretion taking notes of why they
made those decisions and then further prototyped.

Fig. 9. Participants at the task of designing chairs
(The program is a Generative Adversarial Network(GAN) which has been trained on 3000 Images of ‘Classical
Chairs” scraped from Pinterest boards, The program generates synthetic images of ‘chairs’ from its training
dataset)
Fig. 10. Left: Data set, Right: Generated chairs
The Designer’s role in this experiment was that of collecting and feeding curated, relevant data to the machine,
and developing its design decision. The Machine’s work here was to provide iterations in form and structure
based on the ideations it was fed. The status quo of design power was: Machine 25% Human 75%
Concluding notes Through this experiment, we merely used the machine's computational power to aid us in the
repetitive, arithmetic part of the design process. The machine was fed concepts, and it provided combinations of
those concepts to show possibilities. We used our design discretion to choose from those possibilities, developed
it further and manifested our own design decisions into real prototypes.
After prototyping both the teams were asked to imagine a context and use of the final product and write a short
note about the story of the chair they prototyped.

Fig. 11. An outcome of the first experiment, humans- 25% and ai-75% power
7.3 Part 3: Ethics of AI
Fig. 12. Insights that emerged from the concluding discussion
Lastly, the participants gathered all their designs and put theme to display at the end of the workshop. Discussion
was held on thoughts and reflections of the workshop. Some nudges were:

What was expected? What was unexpected? What was quirky? What was regular and mundane? What was
irregular, new? What was your wow moment? What were your strongest feelings, at what point? What was
similar in the two workflows? What was starkly different in the two workflows? Do you have similar interactions
with a machine in your design work? Where? How? How might you interact with machines in your daily work
such that they can help you design better? How to efficiently and effectively use the tool of AI such that it
enables Human IA? What went wrong in the project such that it made you feel something is not okay?
8 Conclusion
Fig. 13. A short representation of the process followed in Experiment 1
One of the groups put a picture of a toilet seat in the program, the AI classified it as 96% chair. They were
excited, We were excited, This unexpected result lead to probably the most exciting creation of the workshop , A
Toilet Chair. What kind of traditional design-process could have come-up with such an aesthetic-functional
design object. The creation of collaboration between Man Machine has since then become the poster child of this
workshop, It almost refers back to Marcel Duchamp's Dadaist masterpiece : ‘Fountain’. The way Dadaism
questioned the Modernist world, A human-AI relation can fundamentally change the way we look at and practice
design.
Fig. 14. A serendipitous coincidence or an emerging pattern?
The workshop highlighted the need for constructive critical-conversations about Machine Learning ( What it is,
How it works, Who makes it ? ) and how we can ( and if ) potentially incorporate it in our workflows. It also
called for a relook on what exactly is creativity and how much of it has do with our humanness and originality.

Most importantly, It made us reflect on what we do with a tool and what in turn it does to us, and that perhaps
being aware,critical and conscious is what we are doing and what we must do.
While reviewing the programs for writing this paper, we found out that we accidentally labelled ‘toilet seats’ as
‘chairs’. The program didn’t make an eccentric design decision, we did.
9 Reflections
Although we managed to touch upon some pointers of Machine-Learning and Design, we are still long way from
even scratching the surface of understanding how the promise of Machine learning would playout. The workshop
showed two different ways of approaching design process but in its scope it was mostly limited to forms of the
chair.
Design is not just a study of aesthetics or problem solving but of societies, norms, economies, ecologies,
technology, sciences, environments, more and interaction of all of it. Technologies like Machine Learning are
going to impact all the fields of study we know and maybe don't even know yet, A conversation about its impact
on design is incomplete without talking about its impact on all other fields of study too. Not just academics, Its a
conversation everyone should be a part of - The policy maker, The citizen, The Intellectual, The non-intellectual,
The old and The young.
A data aware designer is a small step in this direction. One workshop at a time.
This workshop is a project in progress and would perhaps stay that way for a long time to come.
Taking into consideration the overwhelming reaction to the workshop from the participants and the organizing
team alike, the authors intend on fine tuning the workshop and conducting in creative, scientific, and
humanitarian institutions.

10 Authors Bio
This project was conducted by Yatharth and Harshali.
As of 2019, Yatharth is an exhibition design student at the National Institute of Design, Ahmedabad. His
interests are in emergent collaboration of Man-Machine along with the intersection of
Media-Technology-Society and making technology easier to understand and create.
Harshali Paralikar is a product designer-researcher. Her interests lie in the inter-relations between the fields of
Behavioral Sciences and Emerging technologies. She graduated from the National Institute of Design as an
Industrial designer majoring in Product Design.
References
1. UNIDO ICSID, NID,  Ahmedabad Declaration on Industrial Design for Development

. 14 Jan. 1979.
2. Lilley Sam, Men, Machines and History: The Story of Tools and Machines in Relation to Social Progress

. 1948
3. Engelbart Douglas, Improving Our Ability to Improve: A Call for Investment in a New Future
3. Engelbart Douglas, Augmenting Human Intellect: A Conceptual Framework

.

1962
4. McLuhan Marshall, The Medium As The Message

, 1967
5. Chen Angela, @chengela, A pioneering scientist explains ‘deep learning

’,

The Verge, 16th Oct, 2018
6. Hao Karen, Giving algorithms a sense of uncertainty

MIT Technology Review, 18th Jan, 2019
7. Accessible AI Tools by Shirin Anlen, MIT Open Documentary Lab, docubase.mit.edu
8. Sagmeister and Walsh, Casa da Musica
9. Grey Richard, How to see climate change through other eyes

, BBC Future, 21st Jan, 2019
10. R2D3, A Visual Introduction to Machine Learning
11.ML5js, Image Classifier Example

,

Classifier Source Code
12. Tahoon Kim, DCGAN-Tensorflow

, ,

GAN Source Code
13. Philipp Schmitt, The ChAIr Project

, ,

Inspiration for experiment, 2018
14. Radical Norms, 100percent Chair Projec

t ,

Inspiration for experiment, 2018
15. Patrick Hebron, Rethinking Design Tools in the Age of Machine Learning

,

2016
Acknowledgements
The authors extend their deepest gratitude to Prof. Tanishka Kachru without whose help and guidance this
workshop would not have been possible. A big thank you to the organizing team of the DSD conference for their
support in realizing the workshop. Lastly, the authors thank the Ayaz Basrai and Critters Collective for being a
source of great and constant inspiration.
fin