There are three dimensions to consider when using AI models:

1) It uses vast amounts of training data which is currently lacking consent/payment for input into AI models for training data. Models aren’t constantly accessing this data once they have been trained all they have is a model of language rather than a knowledge database that they are querying.

2) User inputted prompts /data shape the patterns that are found. There is a considerable amount of manual intervention and engineering to ensure something sensible is produced after the prompt.

3) Output is based on the prompts and training data and any content generated has unclear copyright status. There is no direct link between the training data and output so even if output looks like training data you can’t draw a line between the two, this makes it harder to call out copyright infringement.

For more technical detail on how AI models work see this Ada Lovelace explainer.
 

Return to AI Guidance & Resources