What Aaron understood

·

I didn’t know Aaron, personally, but I’d been reading his blog as he wrote it for 10 years. When it turned out that he wasn’t going to be writing any more, I spent some time trying to work out why. I didn’t find out why the writing had stopped, exactly, but I did get some insight into why it might have started.


Philip Greenspun, founder of ArsDigita, had written extensively about the school system, and Aaron felt similarly, documenting his frustrations with school, leaving formal education and teaching himself.

In 2000, Aaron entered the competition for the ArsDigita Prize and won, with his entry The Info Network — a public-editable database of information about topics. (Jimmy Wales & Larry Sanger were building Nupedia at around the same time, which became Wikipedia. Later, Aaron lost an election bid to be on the Wikimedia Foundation’s Board of Directors).

Aaron’s friends and family added information on their specialist subjects to the wiki, but Aaron knew that a centralised resource could lead to censorship (he created zpedia, for alternative views that would not survive on Wikipedia). Also, some people might add high-quality information, but others might not know what they’re talking about. If everyone had their own wiki, and you could choose which trusted sources to subscribe to, you’d be able to collect just the information that you trusted, augment it yourself, and then broadcast it back out to others.

In order to pull information in from other people’s databases, you needed a standard way of subscribing to a source, and a standard way of representing information.

RSS feeds (with Aaron’s help) became a standard for subscribing to information, and RDF (with Aaron’s help) became a standard for describing objects.


I find — and have noticed others saying the same — that to thoroughly understand a topic requires access to the whole range of items that can be part of that topic — to see their commonalities, variances and range. To teach yourself about a topic, you need to be a collector, which means you need access to the objects.

Aaron created Open Library: a single page for every book. It could contain metadata for each item (allowable up to a point - Aaron was good at pushing the limits of what information was actually copyrightable), but some books remained in copyright. This was frustrating, so Aaron tried to reform copyright law.

He found that it was difficult to make political change when politicians were highly funded by interested parties, so he tried to do something about that. He also saw that this would require politicians being open about their dealings (but became sceptical about the possibility of making everything open by choice; he did, however, create a secure drop-box for people to send information anonymously to reporters).


To return to information, though: having a single page for every resource allows you to make statements about those resources, referring to each resource by its URL.

Aaron had read Tim Berners-Lee’s Weaving The Web, and said that Tim was the only other person who understood that, by themselves, the nodes and edges of a “semantic web” had no meaning. Each resource and property was only defined in terms of other nodes and properties, like a dictionary defines words in terms of other words. In other words, it’s ontologies all the way down.

To be able to understand this information, a reader would need to know which information was correct and reliable (using a trust network?).

He wanted people to be able to understand scientific research, and to base their decisions on reliable information, so he founded Science That Matters to report on scientific findings. (After this launched, I suggested that Aaron should be invited to SciFoo, which he attended; he ran a session on open access to scientific literature).

He had the same motivations as many LessWrong participants: a) trying to do as little harm as possible, and b) ensuring that information is available, correct, and in the right hands, for the sake of a “good AI”.

As Alan Turing said (even though Aaron spotted that the “Turing test” is a red herring), machines can think, and machines will think based on the information they’re given. If an AI is given misleading information it could make wrong decisions, and if an AI is not given access to the information it needs it could also make wrong decisions, and either of those could be calamitous.

Aaron chose not to work at Google because he wanted to make sure that reliable information was as available as possible in the places where it was needed, rather than being collected by a single entity, and to ensure that the AI which prevails will be as informed as possible, for everyone’s benefit.