To Fight Institutional Capture, a New AI is Here
all new case.science
Well it has been a while hasn’t it? I’ve been working on case.science, which long time readers will know all about. There’s now a big update, which marks a significantly better direction to take the project in.
Case.science got started as a way for AI to help society to understand arguments, and evidence that’s hard to find, or perhaps has been actively concealed. Something digger readers know all about. To combat this, you would find the data and science, ‘upload it’ to the AI, then it would help everyone explore it at their own pace. Shouldn’t more people know about all the weird things Fauci did? You’d ‘build a case’ on that - a repository of articles and science papers - then an AI would assist everyone in navigating through that material. Imagine it like a wiki page but an AI answers user questions instead of faceless corporations editing the wiki copy to snidely editorialise the story.
This is still the aim of the project, but things have changed in how you can build these cases.
Case.science now acts as research assistant - type in your question, and it will zip off onto the internet to try and find you the answer. It will look in pre-designated places for data relevant to your questions. It’s designed to look for science, but it’s also good at doing news and general searches too. Whatever you ask, caseAI translates that into a few custom searches, and it tells you what it found.
Great!
But what about reading what it found? Well, caseAI can do this too. As a matter of fact, if you like the look of an article, just ask it to read it for you. Maybe even ask it to look for something specific in the article - it will try its best to find it. You can even paste in a URL and it will try to read it for you.
Why have I done this?
This new direction allows the AI at case.science to look through really valuable datasets instead of relying on ‘top data’ from the legacy information system. This control we have about where caseAI looks means we can correct some of the ‘institutional capture’ of our information ecosystem. Right now, the paradigm of ‘ask me a question and I’ll tell you the answer’ relies on trust - just what is going on in the background there? Where exactly did you look?! With case, we can make sure it’s looking in places that might otherwise be excluded.
For example, what happens if the first port of call for ‘what are people saying about X’ is that the AI looks through Substack rather than the NYTimes or The Guardian? Well case.science can now do just that, it can actually search through substack for posts on particular topics - a great example of how we can champion independent voices. Can we do this for science papers too? Because if you search for science on Google you’re most likely to get the ‘top science’ that was put there. Right now, case.science looks on pubmed, zenodo, arxiv, biorxiv and a few other places. If you have ideas on where else it should look, let me know. You can even use the feedback tool on case.science to send your thoughts, just tell the AI you have some feedback!
‘Cases’ are user created lists of links and data that support a particular idea. You can now make them through the AI itself. Just ask. The caseAI can help you make the case, and it can add links to the case. Want to make one about Fauci? Get searching for supporting data, and add it to a case. Case.science will eventually become a respository of human created ‘cases’, and it’s these cases which I believe will make the biggest contribution towards a more coherant society. Human curation is going to be a very important way to push back in the age of AI.
As well as ‘cases’, case.science now has conversations which you can share if some interesting insight is developed. Conversations are a really natural way to research something by collaborating with the AI, searching, and reading articles together to build up a long conversation about a topic. If your conversation was interesting and insightful, you can share it. The conversation goes from private to public, and you’ll get a new link to share with friends.
Because case has access to news, it’s also a great way to stay up to date without having to look at doom scroll feeds. You can ask it for a news briefing about AI, or medical science, or anything really. By using AI to read things on our behalf, we can get away from feeds designed to trap us. To be blunt, I think feeds are basically evil - they are programmed to keep you using them as long as possible and you have to wade through the manure to find your diamond dust. Try case.science instead as a way to get your morning information burst on a particular topic.
As ever this is an alpha product - there are definitely things that will not work perfectly right now. Please let me know by using the feedback tool. If you like what case is doing, it would be great if you subscribe, as this will remove all restrictions on how many articles and searches the tool will perform for you in a single day. You can search and read away to your hearts content. If you are intested in the project more seriously, you can contact me via substack, or again, use the feedback tool on case.science. Maybe substack would be interested in it since it searches their platform?!
If we don’t stake a claim in the rapidly developing AI territory, we risk condeming ourselves to yet another system that’s ripe for abuse. Just how substack is challenging mainstream news voices, we need to have something similar to challenge mainstream AI engines. If you like that - give case a try. As ever, shares are deeply appreciated. The audience we get is the audience you give. Comments open.
Sounds like a very interesting useful piece of A1. All the best going forward. I think Steve Kirsch would enjoy giving it a workout.
This is fantastic! Thanks Phil for doing this work. I never had any success setting up a case that worked (not that I tried real hard), but this already is working great. I think I'll be using this a lot. It's VERY helpful.