The two AIs - problems of obscure interests on the internet.
The two AIs - problems of obscure interests on the internet.
Thus far, I’ve only written about local history which was one of the major things I’ve wanted to write about for some time. That said, I do intend for this blog to also be a bit more general with some of my other thoughts featuring, this is a bit of a mix between some random thoughts I’ve been trying to make sense of, reflections on tech and my day job combined with some rambling about how annoying I often find the internet and social media. Especially when it comes to connecting to other people, which it should be much better at.
The first ‘AI’ stands for ‘Algorithm Internet’ - not my phrase but very apt. Essentially this is what has reduced the internet to an endless scroll of dopamine tweaks and hiding all of what used to be called ‘long tail’ sites and interests. For me this is most visible on YouTube and it's worth mentioning, I pay for YT - and so generate 0 ad revenue as I don’t get ads.
I have a number of obscure interests (what really!) and despite some very obvious clues, YouTube pretty much refuses to recommend vids based upon them, as it's likely most of the channels uploading will be very small and thus not generate much engagement. For obscure interest number one, I regularly watch an elderly British man (so interest and region both should be flagged) but hardly ever get similar channels recommended to me. The largest channels in this obscure interest are American and oddly even the very largest doesn’t feature in my recommendations - but will come up if I search for said interest. Now regionality in this is very important and even when expressly included in the search will not seemingly weigh the results returned and will instead return the same (slightly) larger American channels.
Obscure interest number 2 does a little better with a couple of channels regularly making an appearance but again tends to try and push larger channels instead of those more suited to me personally. Consumerist interest does a lot better with some very large channels featuring both Brits and Americans pushing their no doubt sponsored expensive wares. Sport interest is the major one - it's a global sport (i.e really global, not American global) and not football so almost 3 in 5 of my recommendations seems to be for this, as it's no doubt generating a much larger global click fest. All of this keeps channels that might have something interesting to say, but only ever a limited audience, down and means that everyone is pushed into a sort of average metric consumer purely to generate the most clicks. This effect is even more pronounced on Instagram. All social media (imho) should push people together with shared interests, but instead seems to define its own interests and pushes everyone to them instead, in order to generate a continuous stream of clicks and thus ad revenue.
The other ‘AI’ is of course Artificial Intelligence, now I run a team of techies and have seen just how amazing it is for coding or generating defined language (terraform for AWS deployments in our case). Anything that has a clear definition and clearly laid out rules and functions. This Christmas break I put an old web application through Claude AI and what came out was remarkable. Cleaned, commented code with full functions. Asked for a new function and associated html elements and it just appeared. Whilst making me nervous about my future career, it clearly works and has utility for defined and clear areas, coding, finance etc all had better adapt.
That said, I usually start testing AI by putting queries through it about obscure interest number 2 and then more recently getting results back from Google AI on queries about local history and in both these use cases, it's appallingly bad. The Google AI generated entire articles that it believed to be about what I was searching for and these ranged from totally incorrect and about something entirely different, to a plausible sounding article that looked to be informative, but instead had conflated a number of things from the same source into a single incorrect whole. It's troubling that this sort of result will find its way to other sources, which in turn will be referenced again and strengthen the inaccuracy and thus create a fiction that can not be disproved beyond primary sources. Wikipedia is already sort of doing this, I also recently found a reference on an obscure page that refers to a very plausible source elsewhere on the internet. Trouble is, the plausible source doesn't actually back up the original reference, and indeed I can’t find any other source from online to my local history books that confirms it, but it's so very believable I don’t really think anyone would question it. It may or may not be true, I just can't prove it either way.
These two ‘AIs’ together worry me when it comes to the internet as a source for information and shared knowledge. The forums of old, where people who knew things would post have been replaced by light and fluffy social media groups with repeated questions but little real understanding. People with interests they wish to learn about are no longer steered into finding the knowledge themselves but given surface level knowledge which is quickly lost by the time the next confused querier comes along. Nothing is preserved and what is, may be totally wrong. People who do know, either lose interest or get tired of being shouted down by the incorrectly but confidently informed. Add to this artificial agents creating content at an unprecedented rate and then essentially re-learning from the same information source is a recipe for total knowledge breakdown.
Anyway, now I’ve got that all off my chest, I’m off to read books published before all this became a thing and hope that that information source at least will remain for some time.