Here is the basic policy for AI on this website:
No AI content modification tools are used, ever. Here, by “AI content modification”, I include generative AI, and other machine-learning or AI-based tools that modify or edit content (text, video, audio, etc). And by “AI”, I mean any software that has been trained on a large dataset of images, text, or other data.
Using AI to modify content is considered unacceptable for artistic reasons, since it goes against the basic bottom-up approach of starting with a blank slate as it needs previous creations applied mechanically to operate. I also am generally against AI since its long-term cultural and societal impact is guaranteed to be a net negative for life.
I wrote a detailed explanation in my newsletter. Here is the main thrust:
AI is not just a dangerous technology that will grow independent of our control. Its very existence is an affront to nature and biological life because the spirit of AI is to push intelligence beyond wisdom and replace the intellectual and creative work of human beings. AI represents the apex of overefficiency and our desire to take beyond what we need. It disrespects our existence and is a manifestation of the ethic that life does not matter. By supporting AI, you don’t just support the potential evils it can cause, you affirm and support unsustainability and you fuel the arrogance of the foolish who gamble on the false premise that AI can be controlled and tamed.
What about the definition of AI?
The definition of AI is of course tricky, which is why I specified that it is any software that is trained on a large dataset and subsequently uses that trained model to modify content. Very primitive forms of AI are similar to fitting linear models, such as fitting a dataset to a vector-based model of the form $y = Ax$, so one needs to be careful as such “machine-learning” is too primitive to be really considered AI in any normal understanding of the word. Pure image recognition such as OCR could theoretically be considered AI as well since it is a primitive version of machine learning, but again, such models do not modify any content or attempt to intervene in the creation process, so I am fine with those.
In any case, the definition that seems to make the most sense is one that factors in purpose: if the purpose of an algorithm is to modify content, and its means of modification can only be reached by being trained on very large datasets, then it should be considered “true AI”. Other definitions that only take into account whether an algorithm is “intelligent” are, in my opinion, useless.
What about ethical AI?
According to my ethics, there is no such thing as “ethical AI”. According to some, ethical AI should be a model that is trained on data that is given freely by those who have created it. But that only focuses on a very narrow, data/property-definition of ethics, looking at only the rights of those who have created the data in the first place. This narrow and false sense of ethics fails to take into account the greater responsibility that comes with all creation, and the long-term effects of AI on society whether or not AI has been trained on data given responsibly.
In my opinion, “ethical AI” is a weak and failed attempt to use a fundamentally destructive technology without being an obvious destructive force.
What about YouTube?
Of course, I don’t use AI to make my content. Unfortuantely, by using certain tech platforms, I am supporting AI indirectly. For example, YouTube has recently added various AI features that scans videos. Although I hate this, I have to make a compromise to reach some sort of audience. I do write complaints when I see these new features, but I can only do so much. That’s of course how technology spreads: it makes something a standard and therefore difficult to avoid even when new features creep in that are not wanted.
I will always attempt to minimize my involvement with AI. Unfortunately, YouTube is one of my few sources of income so I have to keep it around. However, at the very least, I’ve already disabled automatic AI dubbing, and I will disable any other AI features that can be disabled.