Writing an AI Policy

It would be very easy to write an A.I (Artificial Intelligence - not Artificial insemination, don’t really have a policy for that yet) policy that says ‘No, just no’ in fact that is the place holder policy that I wrote. But the more I think about A.I the more I think that it’s not that simple.

The term A.I was coined in 19561 and it has grown and mutated many times since then. When I was at University our A.I module consisted of Heuristics and Expert Systems. It might have included more things but at the time it wasn’t something that really interested me and probably wasn’t paying attention.

Heuristics at the time was introduced to us as ‘the traveling salesman problem’ calculating the shortest or fastest route between two towns. Today the maths and techniques used in this is in use all the time calculating routes in Google maps and Wayze for us , and silently running all over the internet routing our data packets across the world.

Expert Systems were thought to be the future of A.I in the 1970s and 80s. They are still used in specialist and niche cases. Technology once developed never really fully dies but slowly fades away, there are still people working on systems programming in COBOL.

Skipping forward forty years and now A.I research has invented Neural networks and other more advanced ways of doing maths with data and now we have generative A.I creating words, pictures, video and music. Mostly slop choking up the internet with bland images, weirdly deformed people and nonsensical articles and books.

This Slop is what we are saying no to. There will never be any articles written by A.I or images of people with six or seven fingers on the weird jelly website or used in any wa in our work.

But to say that WeirdJelly will never use A.I at all is not really possible. Machine Learning is a branch of A.I and Linear Regression4 is a technique used in Machine learning that in its simplest form is used to find the line of best fit of data points on a graph.

I don’t think that creating an A.I policy can purely be about what technologies and algorithmns we will use or not use. It has to be about people. The things that people create and use and respecting their work.

It would be fantastic to help a blind or disabled person tell the difference between ripe and unripe fruit but that shouldn’t mean we steal the work of many artists or photographers to create a training dataset of images.

We also need to think about the wider world. This Generative AI slop is using massive computer resources, wasting heat and clean water. If every time somebody generated a silly cartoon they had to turn on a tap and a heater for half an hour in their home even on the hottest day of the year it would make them think twice about playing with these things.

The final thing we will consider is the quality of work. It’s not just the obvious slop failures but AI can fail in more subtle ways. Code generated by AI assistants can contain bugs and security weaknesses that are hard to spot if you are not familiar with the code that you won’t be if you haven’t wrote it yourself. The internet is now full of of slop articles that contain a myriad of errors which are just embarassing all around.

There will always be edge cases where

So thinking about all the things above, the first version of the weirdjelly AI policy is:

1. Think very carefully about the algorithmns and tools we use. favour Free and Open Source software that respects the rights of its creators over closed source proprietary software that is opaque about what is does with its data. and whose creators share our values.

2. Do not use any generative AI tools to create articles, code, images, videos and animations or music.

3. All articles, code, images, videos and animations or music will either be created by Weird Jelly or created by humans where we know the provenance and will be fairly compensated and credited for their work.

4. When using any more advanced tools that require training models on datasets consider carefully the datasets in use and the energy and resources used by the training and inference. Favour low power computing on local devices over remote cloud hosted computing.

5. If any instances of AI generated articles, code, images, videos and animations or music do slip through the net and are used, when discovered we will remove these at the earliest possbile oportunity and either replace with a human created equivalent or a place holder and fully and transparently explain the process that led the item to be used.

We welcome any one pointing out to us if the know or suspect any item on our website has been created either in part or whollly using generative A.I

So i think that covers it for now. The A.I landscape is changing quickly and this may not be adequate in a few months or a few weeks time but it is a starting point and I think it says what we need to say at the moment.

Any respectful (Human written) feedback is welcome by emailing hello@weirdjelly or on our social media. We are on https://bsky.app/profile/studioweirdjelly.bsky.social and https://mastodon.social/@studioweirdjelly