I recently participated in a bunch of courses in Creative-AI, Generative & Computational Arts. In this writing, I decided to briefly summarize the takeaways from these three courses for those interested in using AI as a part of their creative practice; whether you work in arts, music, performance, etc.
First and foremost, by taking these courses I got exposed to interesting persons working in the field and got to see a glimpse of the type of work they are doing. Some are focusing on tools for creatives, others rather focus on theory of computation in arts — but nevertheless what these courses have in common is that they are trying to bridge the gap between AI and Arts by providing knowledge, methods, and tools accessible for the artists to work with.
A realization I had while taking these courses was that the the line between “traditional” computational arts and AI arts is very fine: sometimes it is possible to do similar things with regular algorithms than with AI algorithms. So it raises a question — what does it actually mean to do AI arts? And how should artists work differently with regular algorithms or AI as an art medium? I reckon that the increasing interest in AI arts can be largely credited on vivid AI imaginaries presented on media; AI with all of it’s emerging possibilities encourages people to imagine of future possibilities, which attracts creatives to work with it. And undoubtedly, as with any technological advance, new opportunities open up for other fields within the society. Therefore, AI as an art material provides new opportunities for artists to expand their practice.
There are a few courses currently available within/related to AI arts, as I did an extensive search of them. The first course I took was AI-Generated Media from MIT Media Lab. The second course was AI for Artists and Musicians from Goldsmiths, and the third one was (not only AI-focused) Computational Generative Art from Simon Fraser University. All of these were online courses; of which the first one was delivered through Maven platform and the two latter through Kadenze. The ones from Kadenze are currently free, the MIT Media lab course was cohort-based and required applying for it, but the materials are freely available online. I think it is crucial that courses and other materials would be delivered free of charge, so the artists from various parts of the world can learn how to use AI and engage in AI arts practice regardless of their socioeconomic background.
I have to highlight, that in the Goldsmiths course a software called Wekinator (by Rebecca Fiebrink) is used and it is free to all of those interested in doing AI art with minimal programming skills. In the MIT Media Lab course the tools used (Google CoLab + GitHub) are also free to use for anyone, and so are the Tensorflow, Keras, OpenAI, machine learning tools freely available for everyone. The latter require a bit more programming experience, so I would recommend starting from Wekinator and CoLabs. There are also a wide variety of website tools (ArtBreeder, NightCage, VQGAN+CLIP) that are fully graphical, but for those interested in getting into the practice and learning to do more advanced things, the above-mentioned tools would be a good start.
The challenge that many creatives face is not the lack of interest towards using AI, but rather that the idea seems intimidating and scary. Many artists working with traditional mediums could provide a lot to the emerging field of AI Arts, but are perhaps scared of using these highly technical tools — and don’t know where to start. Therefore I decided to write this article as a pointer towards that, as I already reviewed a bunch of courses and sources available online.
AI-Generated Media by MIT Media Lab
The AI-Generated Media course was a 5-week course with online lectures, exercises, and a discussion platform for co-learning. The course was a cohort-based one, but many of the materials are freely accessible online through GitHub and MIT website.
The first week of the course focused on covering three different ML techniques; style transfer, deepdream, neural doodle. Style transfer is a technique of applying visual aesthetics from one image to other, deepdream creates “hallucination-like” visualizations based on an image, and neural doodle creates a new image based on a doodled sketch and a source image. The second week of the course included GAN’s (Generative Adversarial Networks) by introducing CoLab codes, which one could use to do “mashups” of portraits, animals, etc. Subsequent weeks covered rudimentary audio and visual deepfake generation, as well as theory on the machine learning algorithms behind them. The exercises took place in Google CoLab. There is a bunch of code available in GitHub, which can be directly opened and run in Google CoLab to get started — with minimal programming experience. And those with more advanced programming experience are able to tweak the code and discover new things within in, which some of the course participants did. Therefore, for advanced learners who already have learnt the basics of existing AI Art tools/methods, it is useful to learn Python in order to be able to extend their practice. But it is not required in the beginning in order to be able to engage in AI Art practice.
AI for Artists and Musicians by Goldsmiths
AI for Artists and Musicians is a video-based course with exercises that one submits on the course forum for peer reviews. So the structure is pretty much the same as the Media Lab course — the only difference is that in the Media Lab course you would have office hours support to ask questions and get help.
Goldsmiths is one of the trailblazing art universities in AI art, as they are teaching a Computational Arts MFA program that has a course in Machine Learning for artists. At this point of time, there are still only a few of these kinds of programmes and courses. This first week of this course started with a concise and brief introduction to ML concepts and then proceeded to installing Wekinator (the ML tool) and trying it out.
Wekinator basically is a graphical interface that one can use to train ML models, so it is very easy to use. One can install different plugins (sound synthesis, web camera, visuals/processing) to be trained as an input or output of the interaction. Basically, this allows artists to convert one type of information into another type — such as converting hand gestures or movements on webcam to sounds or graphics, or mapping music into graphics. Therefore, Wekinator can be used by different types of artists from performace art to music, and visual arts.
Computational Generative Art by Simon Fraser University
The Computational Generative Art is a little bit different from the two courses above, as generative art can also be practiced without ML or AI. Generative art is an approach, in which a patters in computed in a way that it generates new outputs, such as for example geometrical patterns in ancient architectural styles. Therefore, the computational generative art course is extremely useful in understanding the history and basis of computational art, and this lays foundation on understanding AI Art — and what it’s specific affordances are in comparison to other types of computational arts.
This course is using the software and programming language MAX for the assignments; and the course actually requires one studying a foundational course in how to use it before taking this one. Here is a basic course on MAX — which you can take prior to the course.
Conclusion
I will conclude in some final remarks about which learning strategy and where to start. The first point is — don’t be afraid, start exploring the medium. Even if you don’t have the technical knowledge, it is possible to start taking these courses on your own pace and learn as you go. And once you dip your feet into the ocean you will start understanding more, which will allow you to explore more advanced things. But most often, advanced technology is not what is needed; the most interesting AI art projects that I have seen come out of exploration, discoveries, and the artists own perspective and approach to the medium.
So I will conclude this writing with a question and something to think about during the learning process — what type of AI artist do you want to be? What kind of impact do you want to make with your art? What kind of aesthetics do you want to explore, and why? What kind of discourse do you want to have with the society through your art?
— — — — — — — — -
Don’t hesitate to contact me if you are interested in chatting about AI Arts or to share your explorations.
