Linkability of AI systems is a threat to the very meaning of a 'web'

[dwsinger]

Current ML systems have problems with linkability; not only is their output often not linked to sources, as the document says, but it's often not possible to link to the system to get the same generated output. Those using the ML system are basically forced to copy-paste, which loses (a) the fact it was ML generated and (b) what generated the output, under what conditions/requests/parameters. Would a 'best practice' for ML systems be to have a request also emit a URL that can be used to re-generate the same output as before?

At the moment, even the same person going back later to teh same ML system may (will?) get a different result, and different people asking the same question may well, if the system is tailored to the requester (e.g. by knowing their preferences).

It's hard to imagine what we mean by a web – whose threads are URLs – if one can neither link to the content generators, nor have the generated content link out.