the industrialisation of plagiarism

machine art and its discontents

An interesting argument made to me a few weeks ago about the debate relating to the use of copyrighted material to train AI systems.  The case was that everyone trains their own brain on copyrighted works – often in ways in which it’s very obvious that those copyrighted works have had an influence on their own later creations.  This process has to get very blatant indeed before anyone thinks it’s worthy of comment, let alone a matter of intellectual property.  Furthermore, when we’re talking about human beings, a lot of people have the moral intuition that we should allow more use of copyrighted material to produce derivative works.[1]

So why should it be different for AI?  If Noel Gallagher can wear his influences so prominently, why shouldn’t ChatGPT?

I think one entirely defensible response is something along the lines of “nah, not buying it”.

As I’ve argued previously on this ‘stack, I don’t agree that recurrent neural network/transformer models are doing anything that is particularly analogous to anything that human beings do, particularly not when it comes to the way they creatively use existing material.  But even if that objection was overcome and someone was able to convince me that AI systems have actual non-metaphorical mental representations and create things in exactly the same way as we do, I would still have a more fundamental reason to treat them differently.

Which is that – the industrialisation of a process matters.  A steam loom does exactly the same thing as a hand loom, but ignoring the fact that it does it a hundred times faster was an amazing social and economic disaster, from which the living standards of the English working class took nearly a century to recover.  You can’t just say “this is doing the same thing as something that’s currently allowed, just more efficiently”, because that kind of efficiency has to be considered at the level of the whole system.  (We all know what the medical name is for a structure that grows without concern for its host body).

More generally, thinking about the obvious (immiseration of the creative sector) and possible (degeneration of the entire cognitive environment) consequences of the industrialisation of content production makes you understand that the whole “boomer/doomer” or “techno-optimist” paradigm is a bad way of thinking about the underlying system. 

Industrialisation is a system-level decision, and it needs to be made consciously.  It’s almost always correct to do something better and more efficiently if the possibility arises; that’s pretty much there in the meaning of the words.  But that’s a local fact; you can only go from here to saying “and so it doesn’t need to be regulated and should just be encouraged to go out improving things as fast as possible” if you’re really sure that the global properties aren’t different from the local ones.

Which means you need to be very sure that the new process doesn’t consume finite resources which aren’t currently priced.  That it doesn’t create externalities which are negligible at small scale but which multiply up.  That there is enough spare resources and cognitive energy to manage the reorganisation of the overall system which will be made necessary.  And so on.

In general, humanity has been pretty bad at planning major transitions.  So far we’ve got away with it.  But we might just have been lucky.