On machine learning: Studio Elttob stands for humans

Hello everyone, I’m Dan. I wanted to write personally about the use of machine learning systems as part of the creative process.

This is relevant because I’ve been deliberating for a while on what role machine learning systems should play in Studio Elttob products, if any role at all. I’m pretty late to the commentary on use of machine learning here because I’ve wanted to understand the space at more than a surface level.

Here’s my understanding thus far.

Large closed companies in the space, such as OpenAI and Midjourney have ongoing legal disputes centred around the use of training data, existing at the edges of our legal understanding of copyright and fair use.

I have strong opinions on this as a creative myself. I’m concerned about what it implies for the future of many things I enjoy the process of today, whether it’s composing and producing music, doing graphic and motion design, or exploring design spaces to find new ways people can interact with computers in a legible way.

However, I must concede that I’m not a lawyer, and don’t understand the ramifications of my preferred ruling. That’s because any ruling on the use of training data in machine learning systems is poised to have sweeping and unpredictable ramifications for the foundations of DMCA and copyright law - “the only law on the Internet that works”. A misstep here could lead to massive power consolidation or widespread abuse by bad-faith actors.

What I do know is that I care about humans. I don’t derive personal value from the idea of a computer generating content for me to consume, because it is faceless; I want to spend the rest of my life in pursuit of moments shared with people, and with mutual connection, and I don’t feel like algorithmic art does that for me. To each their own, though.

Beyond the consumer side, I care about the equitable treatment of artists. I don’t know what the next world order looks like for creatives in the age of cheap content machines, and this concerns me. All that I know I can do is to elevate artists. I want to commit to making these people visible where their work manifests.

I know I haven’t been perfect here before - I will freely admit fault for that, and know that I should do better - but going forward, I want to pay keen attention to attribution and transparency, and show people the giants whose shoulders I stand upon, so that they may not fade invisibly into the background.

In the meantime, the question remains - will you ever find machine learning in a Studio Elttob product?

There’s multiple angles I want to tackle this from, because this question is often taken to imply different things, and I have a nuanced opinion here that doesn’t neatly fit into any camp.

Firstly, I strongly believe that the large companies should be held fully accountable for any infringements of copyright or fair use that the courts find them responsible for. This is not an individualistic, consumer-side problem, and I strongly reject the notions of “voting with your feet” or “attacking the users” on the basis that infighting dilutes our attention away from the companies and shifts our focus away from the key issues of copyright law. Structural legal action should be wielded as the tool to reprimand violating behaviour, and sadly it has not been wielded enough in recent times, especially across the pond in the US.

Secondly, I separate the technology from the companies. I believe in a world where ethical applications of this technology are possible, because the technology itself is a generic statistical tool for extrapolating from data samples with hard-to-notate behaviour. I would never say that Studio Elttob is a linked-list-free brand, so I wouldn’t say it for a series of matrix multiplications. I don’t think generative models are particularly new under the sun, nor do they do anything technologically special, so I don’t see any reason to prescribe an opinion on the core technology itself.

Thirdly, I believe that technology should be in service of the artists. Why build stuff to automate the fun part of the process? For whatever application, we should be employing machine learning only as a tool to make your work as an artist more enjoyable, and to help you achieve your vision. I, like many other artists, been relatively unimpressed by the idea that all we need is a text box for all of our work - instead, I believe that our work should feature manual labour in the key areas we wish to invest in, and use computers merely as an aid to fill in the parts that don’t contribute to the “fun”. I also believe this line is drawn differently by everyone because we all care about different parts of the process.

You’ll notice this is pretty much a non-opinion. I know some people would be looking for a complete rejection of the technology, while others would be looking for a complete absolution of the large companies. I don’t take such a clean view of things, so unfortunately my answer is long-winded and overexplained in lieu of a catchy summary.

I do want to make some concrete commitments:

I hope this is a reasonably nuanced understanding of the situation. I wanted to make these thoughts clear and explicitly written down ahead of time, because I want to hear from you. I owe every last centimetre of Studio Elttob to you, and want to make sure that it is built to reflect what you want to see out of it. These aren’t investor-led products, they’re passion projects built for the love of the craft.

Let me know if you think I’m missing anything from this picture.

~ Dan