The most infringing chair design

Over breakfast this morning – which included some rogue bacon & cheddar sausages from Aldi – I told my partner the story of how James Bridle made a chair in collaboration with AI. She’s an intellectual property rights lawyer, so her immediate reply was ‘I wonder what would happen if it included any protected or unprotected designs from real chairs’.

James points this out in his post:

It should be noted that this chair is built on the stolen labour of everyone who’s ever put something on the internet (including many who passed centuries before the internet was invented). The energy use is not good. I didn’t make this with ChatGPT: I made it with a partial history of all previous chairs, and I held myself back from making it “better”. But it’s something to think with.

Chairs as a design are quite interesting. They’re all basically the same, but there’s a breadth of design possibilities with chairs: in function, aesthetics, materials, context and the intended user, all sorts.

But what if you could use large language models to produce the most infringing chair design possible, combining elements of designs for chairs embedded in the training data of the LLMs?

I don’t know how you’d go about it. Presumably you’d need to see whether a particular LLM was trained on protected designs, using prompt injection to get the data out. Then you’d need to write prompts asking for instructions to build a chair that specify where to draw inspiration? I don’t know.

What would be the point? To see what happens next. There’s a great big legal question mark for such an infringing design, with regards to who or what is responsible for the infringement.