The Source-of-Truth Problem
2024-09-04
---
I've been contemplating various different problems over the last few weeks.
- My work has an internal wiki that is supposed to be kept up to date, but large amounts of documentation live in e-mails, spreadsheets or plain local text files. If those don't make it to the wiki, we miss important work when deploying or troubleshooting environments.
- At home I synchronize many copies of my offline microblog file using Syncthing, but some of my devices don't run Syncthing. How do I prevent diverging timelines on those devices? If two files do diverge, what's the best way to resolve it?
- While I find myself less and less enthused about how they're used, I still like the concepts behind cryptocurrency and decentralized finance. I'm interested in how such systems build reliable decentralized ledgers.
- I use a central git repository to manage updates to my capsule. If I don't remember to pull from the central repo to my devices before I start updating, it's likely I'll create diverging commit histories, which are a headache to fix.
As I've thought more about these quandaries, I began to notice that all of them have something in common. All of them relate to a fundamental problem involving some standard of correctness. Some object or set of objects needs to be considered unquestioningly correct, be it a timeline file, a wiki, a blockchain or a commit history. But how can one trust that some document or file is correct?
I began to think about the problem in other respects. It seems to show up in many areas, from political news to scientific discoveries to AI. The more I played with it in my head, the more it all seemed to boil down to one question: what do we use as a source of truth? What do we take for granted that empowers us to make decisions? This is what I call the source-of-truth problem.
Imagine a world that only uses gold coins for money--no credit and no digital transactions--and all fake coins can be detected. To reliably determine how much money a person has, one only needs to refer to physical reality as a source of truth. It's easy to check someone's balance: simply ask him to show it to you. It's easy to change how many gold coins you have, by transferring ownership of your coins or having someone else transfer some to you (such as by buying or selling something). If fake coins can reliably be detected, the only way to game the system in your favor would be to conjure genuine gold coins out of thin air, and that's pretty hard to do.
From this example, we can see that a reliable source of truth has at least three properties.
- It's easy for someone to reference.
- It's easy to change genuinely.
- It's difficult to fake.
Any information system that has these three properties can be used as a reliable source of truth--and if one wants to use a system as a source of truth, one needs to figure out how to give it these properties.
Bitcoin and other blockchains want their balance sheets to be a source of truth so they can form the basis of financial transactions. The fundamental problem they try to solve is how to make their ledgers reliable as a source of truth while still being decentralized--essentially creating multiple sources of truth and having them all agree with each other. A distributed syncing system can make information easy to reference and easy to change legitimately, but making it hard to fake is an extremely difficult problem. I believe that's impossible to do without keeping a full history, and if so, cryptocurrencies will always require a blockchain--and will always require the energy expenditure needed to maintain the blockchain.
A similar issue exists with the git repos I use to update Rob's Gemini Capsule. Each repo is a source of truth with a copy of all the files and a full commit history. Each is easy to reference and easy to update legitimately, but each is also easy to "fake", in the sense that one repo can easily have a commit history that differs from another repo's history. I have to be diligent in remembering to pull from and push to the central repo every time.
In the case of my offline microblog, there is no such thing as faking a change: I either added a post or I didn't. But since there are multiple sources of truth, namely the various copies I have on my devices, adding a change legitimately is not a trivial task. There are some ideas I've come up with on how to get around the problem, but most of them move away from the paradigm of always keeping everything in one file.
Real-life events are an interesting case. Real phenomena, as in humans tangibly interacting with the physical world, are easy to do legitimately and hard to fake, but they can also be hard to reference. In a hypothetical world without long-distance communication, an explosion in a Polish gas pipeline could cause energy shortages in Europe, but someone who live in Spain might have trouble getting to Poland to verify that was what caused the shortage. To get around this limitation, we supplement the source of truth that is physical reality with the source of truth that is abstract representations of reality: writing, images, recorded sound, data from remote sensors, statistical results, and so on.
I feel that in our modern times, many people have completely supplanted physical reality with representations of reality as their source of truth. Data are easy to create genuinely and easy to reference, but we've seen time and time again that data are also easy to fake--and with the rise of AI, even video and audio are becoming unreliable. People see posts on Twitter or Reddit and believe them unquestioningly. News reports on TV must be correct: why would they say it if it wasn't true? Numbers are used to mislead and misrepresent, but people trust that if the calculations are there, they were most likely done in good faith. Data can complement what we experience in the offline world, but for our own sake, it must not replace it.
The source-of-truth problem shows up in so many parts of life, and it appears in ways I wouldn't expect. I haven't performed rigorous logical or philosophical analysis on the subject, and I have no idea what literature might be out there about it. If you've read about topics like this before, I'd be interested to hear about it.
---
[Last updated: 2024-10-06]
Source