My co-author for this post - Her Catness Fibi. She didn't contribute much beyond a few meows and a lot of purring, but she was very supportive.

We are going through a large change in how we work with information and knowledge. At the risk of hyperbole, I think this change may be one of the big ones, like the invention of writing, printing, computing or the internet. I am talking about the tools we collectively call “artifical intelligence (AI)” and specifically the large language models (LLMs).

There are large ethical concerns using such tools. I have enumerated some myself:

Large-language-models (LLMs) have huge issues:

  • They’ve been trained on information they shouldn’t have (disrespecting ownership and taking advantage of groups of people)
  • They are owned, operated and pushed by some of the worst people / companies in the world
  • Using the services have inherent privacy and confidentiality risks
  • There are large and unsustainable environmental costs to training and using them
  • Their use pushes us away from our direct interaction with whatever medium they produce in - and that feels like it de-values the work of humans in that medium

All those, except possibly the last one, are actually criticism against the hyper-capitalist structure we live in.

I stand by that list. I also think almost every point on it (not the last one) is a criticism of the hyper-capitalist structure the tools exist in rather than of the technique of using tools themselves. Refusing the technique does not repair the structure or even engage with it, it just removes us from the conversation about what the technique becomes.

On exploitation and extraction

The structure within which these tools exist is exploitative and extractive, I don’t think many would disagree with that. The tools themselve are not inherently exploitative or extractive, but the way the current models have been trained are! This makes their use ethically problematic. Is this a question of the tools’ applicability or is it the lack of compensation and consent in their training that is the problem?

The extremely wealthy people and companies that train models are taking the easy way out by just paying people to shut up and license their work, but that doesn’t make the structure any less exploitative. Authors of the works that have been trained on are by and large not getting paid for their work, at best it’s the publishers of that work that get compensated. This was, however, always the (bad) deal that the authors made with the publishers. These new tools are just a new instance of the same exploitative structure.

It is easy to say “this is the deal you made, the publishers now own the rights”, or even “you published this under an open license, you can’t complain”, but that doesn’t make the structure any less exploitative. I don’t think anyone expected the way these models ingest and use the works produced by current and previous generations. It feels exploitative, because it is exploitative.

I would draw a comparison to how YouTube is currently our common library of video content. They started out by offering an incredible deal, taking on the cost and complexity of hosting video for internet delivery. This gave creators a place and a platform, but slowly they have tightened their grip and now they extract increasingly from creators and viewers. The letter of the deal is largely unchanged, but the reality of the deal has changed. The same is true for the current LLMs, they are extracting from creators and viewers, and the letter of the deal is largely unchanged, but the reality of the deal has changed.

I don’t think the tools or techniques themselves are the problem here - again it’s a question of the structure and context they live in. We should always fight against exploitation. This is not new, and not particular to these tools, but it feels new to the people who are discovering that their work is used in ways they never imagined.

On the effects of using the tools

We are changed by the work we do, by the tools we use, and by the structures we work within. Technology is not value-neutral. The tools we use shape our ways of thinking and being deeply.

This is my greatest concern in using these tools: they separate us further from the media we work with. An “AI artist” may not be able to enter into a conversation with the camera or oil-paint like the masters of old. A writer that uses these tools may, over time, lose their ability to write without them. A programmer that depends on models to produce software may find themselves unable to write or even understand the code they produce.

There are parallels here to other tools. The people who produced images with cameras in the 80s moved on to increasingly digital workflows. In the 2010s many people who made visuals were far removed from actually interacting with cameras, and certainly from interacting with actual film. I am not a music producer, but it seems to me the same happened in music: we moved from analog to digital, increasingly producers, DJs, MCs and mixers make the music. Artists still exist, but they seem to be in a different context now than in the 50s or 60s. Is it better? Is it worse? I don’t know but the tools we use shape us and our work.

What can we do?

We can always refuse to use these tools, but we should not ignore them. We will all interact with the output from them, whether we want to or not. They are everywhere. This post was written by me, but I used models to help check my grammar and tone.

What we should do is use the tools deliberately: know what leaves our machines, distrust the output until it passes muster, and keep the knowledge and the judgement ours. When we use these tools it is our duty to use them in the service of human-centric goals, not as accelerators of exploitation!

If such tool use disqualifies my arguments for you, that is a respectable position. We all have our own values.

It is on each of us to decide whether we can use such tools as these. My problems with them seem to me to be mainly with the system that produced them.

These tools have arrived and they will not simply go away. It feels like “Pandora opened the box”, and we are in a new world. It is on us to use these tools to improve the world and fight the structures that would exploit and extract from us and other people. I feel the current moral panic about these tools are an expression of the inhumanity of our exploitative, capitalist society. That is where the enemy lies, not in these tools.