From Nebula To Star

After the interactive survey, I have run some of the prompts by the participants through a tool for AI art generation. Below are some examples.

Semi-structured interviews with some of the participants followed up, understanding more about their perception on AI-generated art. People were primarily being amazed, entertained and very curious about the AI art works.

“Wow, are you sure this has been produced by AI?”

“How has this actually happened? Can you share the details?”

“Can you show me how it’s being made?”

All of the interviewees started analysing the AI-art work – some were comparing and contrasting between their work and the AI-generated one. Some were analysing the potential ideas that stay behind the AI-work and how they were represented, wondering could the algorithm can actually go that deep? Discussing the meaning out load, most of the participants became aware of their own selves, guided just by minor questions on my side. Some of the sessions actually had a therapeutic effect (as shared by the participants), besides being entertaining and educational.

“This could easily be a created by a human. I can find ideas on multiple levels there.”

“Look at all these similarities. I actually can’t believe it. Have you showed the algorithm my drawing?”

“It’s so interesting – it makes me express what I am feeling now”

“Never before thought that I actually have all that understanding inside me. And look at that – it’s art by a machine.”

Some interviewees did feel alienated from some of the works – where shapes were not clearly defined and colours were either too bright or sharp.

“Not sure if I can clearly express it with words, but I can’t comprehend some of these pictures. They just don’t resemble anything I am used to. On some level they just don’t feel very real.”

I found a lot about the interviewees during these sessions, and all of them, when prompted again changed their answer to the question whether they would consume AI-generated art. At the end of each interview, it felt as if the participant was getting a bit closer to understanding the machine, but also – to understanding themselves.

The same happened to me – I came out of these interviews rather different. There was this urge to create.

During my research I have stumbled upon an interactive talk with Abby Mitchell from IBM – An Introduction to Open Source Quantum Computing.

I loved every minute of it – a very complicated matter was explained in an understandable language and for less than one hour my knowledge in the area has significantly increased. I am fascinated by the enormous opportunities this whole area opens up.

Just a week after the workshop IBM shocked the world with massive news. Advance to a 127-quantum-bits chip was achieved – twice as many as the previous processor. This is a huge breakthrough that is completely changing the quantum computing status-quo.

What’s next?

Moving forward I have created an interactive survey and addressed participants from various ages, nationalities and backgrounds. There is no specific demographic group targeted. I was pleasantly surprised to see the enthusiasm from the participants and the fast responses.

Below are the survey questions and screenshots with the answers.

The next step in my journey led me to machine learning models built in Python in (an unknown before to me Google tool) – Colaboratory (or Colab).
I used a notebook created by Hillel Wayne. It’s based on a notebook by Katherine Crowson (github, twitter), which was simplified to be more accessible to non-programmers. The original technique was discovered by https://twitter.com/advadnoun.

With my level of technical knowledge I was able to generate an “artwork” with a text prompt.

Am I actually an AI artist, I was wondering, whilst waiting for the 25-min run to spit out my first ever visual from the text input of “Morning on the Bulgarian seaside”.

 

Wondering about how AI “make its own decisions and interpretations” and having some discussions with engineers, I have signed for a Creative coding workshop with Damien Borowik in UAL, where he reveals more about the Creative coding course he is teaching. I learned about the Processing software – a completely unknown medium for me and about the huge engineering & art community around it. I completely enjoyed writing my own code and seeing it visualised into an art piece.

At the end of the evening, two ideas really fascinated me – the existence of random() and noise() functions.

We humans have found a way to generate randomness by a machine! The Perlin noise has been developed back in the 1980s by Ken Perlin – not even a novelty idea. It turns out we can actually just randomly pre-program a glitch or a bug or anything we want. What a powerful tool to have! And what a philosophical debate of who actually makes the decision as soon the function is called! Regardless of the answer, the essence is that there is a way to “recreate digitally” the incidental shake of the painter’s hand that makes the art work unique…

This was one very entertaining evening I spent, where a seed has been planted – can I become an artist by doing what I really enjoy – coding?

Having had observations on the AI art scene, I went to speak to a pop culture artist who likes “embracement of technology” – Timothy Gatenby. During the session we have discussed the uniqueness of the AI art works and originality – “it doesn’t necessarily have to be made from the hand of the artist, as long the initial idea is theirs”. As an artist, seeking to push the boundaries, he is interested in the new ways to communicate with the audience and excited about the opportunities. He sees technologies and AI in art much more as an opportunity than a thread.

Art is evolving together with society, and as an artist, one may either ride the wave or fall behind. I would argue that an AI algorithm is just an enhanced / different tool (let’s say a brush and canvas vs a digital camera), it is something more than that, by being able to “make its own decisions and interpretations”. AI-generated art has the potential to become a new normal in the near future and artists may divide even further than just digital and physical – but also ones that are using AI.

Below is the full recording of the interview.

Here are the questions running through my head in the recent days.

Can AI mass-produce unique art? A friend or a foe for artists and artisans? (Should this even be a question?)

What makes art unique? Where is the line of uniqueness drawn between two different pieces of art?

Are we in the dawn of a new art-consumer reality – can art be tailored for the taste of the consumer?

Can there be a base piece of art, which then can be then tailored via AI to satisfy the consumer’s taste?

• start with a random choice
• start with a pre-defined choice

Brainstorm of areas, where AI algorithm can challenge the artisan status-quo:

→ embroidery patterns

→ pottery

→ calligraphy

→ beer production

Brauer AI Serves Machine Learning And AI On Tap, Shows Us How To Brew The Perfect IPA

Foreseeing the future, I am questioning:

Can AI teach art?

Can AI predict the future of art?

How does art look like after the AI?

---

Current reads

https://theconversation.com/the-price-of-ai-art-has-the-bubble-burst-128698

https://www.bbc.com/culture/article/20181210-art-made-by-ai-is-selling-for-thousands-is-it-any-good

https://aiartshop.com

https://medium.com/swlh/pricing-my-art-with-ai-625dbe442a

https://huggingface.co

https://www.smithsonianmag.com/smart-news/ai-casts-new-doubt-on-national-gallerys-prized-peter-paul-rubens-180978771/

https://theconversation.com/how-a-team-of-musicologists-and-computer-scientists-completed-beethovens-unfinished-10th-symphony-168160

https://nwsh.substack.com/p/ai-artisans

https://academicsupportonline.arts.ac.uk/learning-resources/17352

Recently I have been pondering some questions related to style and AI in my mind.

How is style defined?

What does style depend on, what are the characteristics?

Can an AI algorithm be unique?

What makes an AI algorithm unique?

If the same system is trained with the same data, would it produce different results, if prompted with different questions?

Would an AI algorithm copy the style of the author / creator or “create a style” of its own?

If an algorithm is trained with Maya Angelou’s poetry, would then it create Maya Angelou’s poetry style?

What would happen if an algorithm is trained with Pablo Neruda & Maya Angelou’s poetry? Would the style of its creations be a blend between both poets?

Can a natural language processing algorithm develop its own style?

And if so, would that make it more “human”? Would human characteristics be attributed to it?

I am doing short interviews with acquaintances – most of them software engineers, trying to see beneath the surface of their digital understanding on the matters above. And whilst I receive quite binary answers with linear deductions on the “technical stuff”, some of the more provocative questions actually reveal deeper layers – converting the software engineer into a philosophical human being. And yes – style copying / replicating is one of the most common usages of AI in art today – visually satisfying and entertaining the human eye, but there should be much more to it all. That additional bit I want to explore.

Meanwhile I have also encountered this very personal and moving publication about an AI algorithm supporting a writer in putting some of the most difficult words they will ever write down.

https://believermag.com/ghosts/

I am absolutely amazed by the therapeutic potential of an AI technology and how it can be harnessed in the future! AI art is currently perceived primarily for entertainment, but can we not use it for educational purposes as well? To educate us how to look at the world with a new fresh pair of eyes and to educate us on how to better look at ourselves and reflect.

Being subscribed for Arthur C Clarke Center for Human Imagination newsletters, I have seen an upcoming workshop that changed the course of my research. Bridging the Gap between Subjective and Computational Measurements of Machine Creativity aimed to provoke a discussion between artists, engineers and academics (most of them being experts in more than one field) on what creativity is and how can we measure it.

As interesting and entertaining as the overall workshop was, here are my more specific observations and analysis that fuelled my subsequent research.

• AI-generated art is a very niche field, only a handful of people around the world are actually involved in it and even less are doing research in that area

With more art galleries hosting AI artists’, and recreations of artworks by late artists (examples with Rembrandt and Beethoven) popularising AI art and the introduction of NFTs, enhancing the trading of digital art, AI-generated art is gradually making its way in the overall art scene. However as there are no huge implications, currently it is still a vastly unexplored area that in my opinion hides a huge potential.

• In order to create AI art (or an AI-art generator), one needs to have an engineering background or at least a very good technical understanding.

Without understanding of the basic principles of machine learning – neural networks, weights on the different nodes and training data sets, one cannot just easily create AI art. It requires both an artistic idea / approach and technical knowledge.

• There are very polarised opinions in many areas and during the discussion time there were two camps formed: “artists” vs “engineers”.

We have all heard about “human vs machine” debate. Well, interestingly there was nothing like this here – it was the good old “human vs human” debate, where “artists” were confronting “engineers” and vice versa for lack of support and clarity in each-others fields.

• Most of the regular workshop participants – neither defined as artists or engineers (like me) encountered some misunderstandings in 1) what the artwork actually is 2) where is the dividing line between the human and the machine creativity.

An obvious sign that AI-generated art is not very easily perceived by the public (or even other AI artists). It needs an explanation. Is the art only what is visually perceived or is it also in the making? Observing the other participants and myself – there was one major feeling – alienation from the art works.

This workshop was a very heavy encounter with AI-generated art and it made me realise that during my research so far, I have always gravitated around AI, trying to utilise it through a different perspective, but never actually getting into the core. Diving right into it, putting it into an actual focus of my research is what I really want to do.

Intro

My initial research was reviewing the mentoring experience both from a mentor and mentee perspective, with the aim of redesigning the interaction via a gamified framework. This aim has developed from what I have perceived as a gap in the mentoring process indicated in my research, where those interviewed were looking for direction/ next steps in their different journeys but lacked the mechanism to facilitate this. The framework this project is looking to develop is not aiming to ascribe answers to mentees, but to help them in generating their own, with or without the support of a mentor.

Such being the case, during the last iterations of my research, it has transcended into deeper exploration of the self-mentoring* experience and how AI technology can enable and facilitate it.

Current research question

How can the mentoring experience be reimagined – enhanced and facilitated through AI technology?

• By providing and guiding users through a structured framework?
• By utilising existing data to provide a more beneficial journey?
• By harnessing similar data to inform decisions and gamify the experience?

Research Methodology

Semi-structured interviews

In order to test and further develop my hypothesis, I have chosen semi-structured interviews as a primary research method. Whilst testing the framework and due to the sensitivity of the questions, I am asking the interviewees to type / write down their answers and not disclose them before the end of the interview. Still, the majority of the respondents chose to voice out their answers. To establish trust and observe the natural behaviour of the interviewees (as much as possible over a video call), this methodology proved to be working. (see Appendix 1.0)

Research

Primary research

General findings

Confirmed hypothesis of a working framework: participants gained more clarity for their next steps / had an answer / direction on how to proceed to solve the issue in their area.

They needed more support when providing answers – more than 75% would prefer pre-defined answers to choose from, instead of them coming up with the answers. 50% would like to be able to complement pre-defined answers with their own.

Preferred method of answering the questions – speaking out loud (voice) or typing on computer.

Dreams

All participants found it hard to define what their “wildest dreams” in the particular area were. As expected, the more the area has been narrowed, the clearer the answers on their side were.

Fears

One thing I encountered was how open the participants were about talking about their fears. In most of the cases they were also very well identified. Participants found this step to be the easiest when answering the questions.

Strengths

All participants found it hard to identify their strengths, related to the particular area they were working on. After some initial guidance on my side (suggesting sample answers) they quickly picked up, still, I would assume due to humbleness and modesty, they did not recognise and reveal their full potential.

Challenges

Initial findings are showing respondents at this step are quite often defining “impossible to overcome” challenges. However with some slight guidance and discussion, they are transitioning into their own..

Answers and findings

Coming naturally, sometimes subconsciously → confirming the hypothesis the framework is functioning.

Secondary research

Bibliography

• Donald A. Norman – Emotional Design
• Edward de Bono – Six thinking hats
• Tom Kelley & David Kelley – Creative Confidence
• Phil Knight – Shoe dog
• Barry Schwartz – The paradox of choice

Sources/references for the near future:

• Henry Petrovski – Success through failure
• John Thackara – In the Bubble, Designing in a complex world
• Nir Eyal – Indistractable
• Nick Bolstrom – Superintelligence
• David D. Luxton – Artificial Intelligence in Behavioral and Mental Health Care

Publications

• The Invisible Teacher: A self-mentoring sustainability model by Dr Marsha Carr

Watson School of Education University of North Carolina Wilmington

• Understanding artificial intelligence ethics and safety

Alan Turing Institute

Research of Mental health apps approved by NHS – NHS website

Interventions

Framework creation

Based on previous primary and secondary research and self-reflection, I have set up the following framework, which served as the basis for my latest iteration:

Area of focus → Dreams → Fears → Strengths → Challenges → Answers

Testing data set

As a result of the interviews, I have gained some initial data, which can be utilised as a testing dataset to train a neural network.

Next interventions

→ ML model (dependant on gathering a larger and higher quality dataset)

→ Framework for frameworks: why limit to one framework, when a framework for frameworks can be developed? There is great potential in using Ai to guide both mentors and mentees to models and frameworks tailored to their specific needs. This is an area for further investigation.

Concluding

My research had led me on a journey, with initial findings influencing the more detailed areas of investigation. This has meant that I have had to be careful to avoid inbuilt bias within the research. In particular, consideration must be given to AI ethics, where both confidential and highly personal data must be treated with care to prevent bias in future recommendations and any other AI implementations (predictions, etc.). The knowledge that the current research is generating is not extensive enough yet to provide assurance of an AI mentorship experience applicable and suitable for all, due to the current small subset of people interviewed. However, expanding upon this with a higher volume of qualitative data will allow more reliance and certainty to be placed on the findings. Therefore, for the next phase of my research, I am looking to extend my interview program, targeting specific types of interview candidate based on the findings of current research, ensuring that initial key findings were not anomalies, but indicative of wider trends.

Appendix 1.0

→ Trust: There would have to be a relationship between the interviewer and the interviewee that transcended the research, that promoted a bond of friendship, a feeling of togetherness and joint pursuit of a common mission rising above personal egos.

→ Naturalness: As with observation one endeavours to be unobtrusive in order to witness events as they are, untainted by one’s presence and actions, so in interviews the aim is to secure what is in the mind of interviewees, uncoloured and unaffected by the interviewer.

Source: Adapted from Woods, 1986