I’ve been making a website which tracks metascience relevant seminars – the Global metascience seminar monitor.
The site subscribes to emails, scrapes seminar webpages (takes RSS), updates daily, takes updates and corrections by email automatically and publishes a calendar feed so you can see events in your calendar automatically.
It’s half a useful tool to keep up with an interesting and evolving area, and half an experiment in using AI. I vibe coded the website, using my openclaw agent running Claude, but the site itself runs entirely without commercial AI (it uses some local models to do ingest incoming info and decide if it is relevant) — the motivation was to avoid building in dependency on a subscription to any AI company.
The monitor is a function which I think has wide relevance to our age of information overwhelm, and to the specific capacities of AI. The AI may not be very smart, but it has practically infinite bandwidth and practically infinite patience. Rather than replace the things I am good at and might enjoy (going to seminars and learning from them, is one example), I can use the AI to do something I am relatively bad at (keeping up with 1000s of emails about seminars, picking out items I might be interested in).
The monitor also illustrates another angle of the current AI moment — true personalisation. There was a lot of talk about personalisation during the early digital era, including talk about personalised advertising. Digital technology is could be infinitely customisable, true. Surveillance companies like Meta have a vast amount of data on each of us, also true. But the digital technology of the last 20 years has not been personalised, and what is possible with AI has shown that up. Targetted advertising was mostly a con to sell to the tech companies’ true customers (advertisers), while the apps and websites we ended up with provided lacklustre personalisation at best, tending to push us all towards a homogeneous experience with a thin layer of different content on top.
(The events feed on Facebook is a particular grievance of mine. There was a point at which nearly every event had a Facebook page, nearly every friend of mine told Facebook what they were doing and what they liked, and I told Facebook what I was doing and what I liked, and yet the ‘events you might like’ suggestions were consistently mediocre. The fact that Facebook couldn’t even make a modest effort at pretending to meet this need shows clearly how their priority was never personalisation for the user).
With AI you can ask for content based on your preferences, and also build bespoke software based on those preferences. I wanted a metascience seminar monitor, and AI meant I could build it (I could, maybe, have built it without AI, but it would have taken six months). I have a whole host of personalised projects, from a text-editor which integrates with my notes, to a system which allows me to file expense claims by sending the receipt to the agent on WhatsApp.
It’s interesting to think about the things AI is particularly suited for – things where I don’t have the bandwidth, things where I don’t have the persistence, things which are annoying (like filling in forms), things which I might forget…And things where the product which is served for everyone else isn’t exactly what I want for my purposes.
The recent history of tech is one of enshittification, and the history of everyone is more admin, more forms, and a world increasingly filled with impersonal, infuriating, bureaurocracy. There are lots of reasons to hate or fear AI, but if it lets us subvert or avoid some of the homogenising, flattening, and constraining tendencies of the digital world, and make things which are particular to what we, as individuals what and need, then that’s a good thing.