AI recursion

AI has made a significant leap the past years, at least for the uninitiated like myself. I am by no means an expert and won’t discuss any technical detail surrounding AI. Simply because I do not know.

I am however not technically illiterate, with a long background as developer. Currently I work as a freelance consultant doing all sorts of different IT projects ensuring compliance with EU legislation, making sure projects are viable and finding ways to reach deadlines.

Whether my background is relevant to the discussion I want to have with myself, I do not know yet. But let’s move this along and see where it gets us.

What feeds AI is enormous quantities of data. Better training data equals better AI.

AI impresses us, because it acts like a co-worker. A co-worker we can not only ask to do tasks but who actually does them. We can ask AI to write speeches, songs, novels, even Powerpoint presentations – you know the pinnacle of human creativity and knowledge – and it just delivers.

There are currently all sorts of legal battles warming up regarding the training content for a lot of these AI-products. More and more organizations, writers and collectives realize that their intellectual property may be used to train the AI without their consent.

The training data is one of the core tenants in AI’s ability to work its magic. It blows our minds currently because it has made such big leaps, but also because it can piggy-back on centuries of human-made content and feed it back to us.

There is already a lot of content spread out over the internet that is created by AI and it will only get worse.

What will happen in 5,10,15 years when the amount of AI-generated content will outnumber the human-made content? Will it not make any difference as the AI-generated content is generated based of human needs or will the utility of generated output diminish over time?

Or will AI over time get better at using the data that is already there? Perhaps get better at classification or using unstructured data?

I think one of the reasons that AI could make such a bit leap within the last couple of years was that it had all this human-generated data to crunch on combined with the legality of using all that data not really being discussed yet.

But perhaps I am mistaken. That would not be the first time.