AI Safety Sphere

[Update: This post is unfortunately no longer relevant as I have stopped developing the website. It is still kinda cool, but it can be difficult to have people collaborate on a mind map. Also there is less of an issue of finding relevant papers with recent advancements in search]

AI Safety Sphere is a collaborative mind-map dedicated to mitigating AI existential risks. The platform welcomes researchers, enthusiasts, and experts to contribute their insights, bridging gaps in understanding and ultimately fostering an open space to explore ideas.

Inferential Walkways

One of the primary challenges in AI safety is understanding how individual research efforts fit into the broader landscape. As a newcomer in particular, it can be difficult to grasp the inferential steps and assumptions underlying various works. Eliezer Yudkowsky’s concept of “Expecting Short Inferential Distances” highlights this issue. There is often a lot of background knowledge and context one needs to start making useful contributions to most problems (particularly those in academia). And worse, we are biased to not believe this is true.

1. Oblivious of the problem ---> 2. Understanding the problem ---> 3. Making contributions to the solution

These are the steps every collaborator must take before making useful contributions. I argue that the jump from. steps 1 to 2 is far greater than that of 2 to 3. In the words of Einstein, “If I had an hour to solve a problem I’d spend 55 minutes thinking about the problem and 5 minutes thinking about solutions.” This platform aims to address this need by providing the context necessary to onboard collaborators by default.

With this structure, you are free to roam each other’s inferential walkways. Each node in the tree will always be one inferential step from the node prior, and so regardless of how deep you go, you can always ground yourself by walking up the tree of inferential steps.

Filling Knowledge Gaps

AI Safety Sphere allows authors to identify and present holes in understanding, which can then be filled by themselves or the community. Unlike the traditional way research is stored and presented, where valuable work can be lost in the sea of papers and articles, this platform ensures every contribution addresses a specific need in the community, and work is not shoved down by the ever growing flow of research.

This structure encourages users to spend more time thinking about how structure what work needs to be done, and what further questions need answering.

One specific example of this in practice is how authors can easily direct feedback by creating specific discussion thread nodes configured to the needs of the feedback.

Lowering Barriers to Entry

This is certainly not universally true, but joining the field, is often a path of many years of education, before one can start making significant contributions. Society cannot afford to have such high barriers to entry with this important of a field currently lacking members.

There are a few ways this website seeks to lower barriers to entry:

If a reader gets confused they can always walk up the inferential walk way to enhance understanding
There is less of a burden on the author to have all of the answers. Instead they are more encouraged for the questions they ask and the problems they highlight.
Far faster to find things. Anyone familiar with programming can tell you how much faster tree search is to list search. (at least if the branching factor is low enough)
It is often annoying when there might be valuable information in an article or book but a vast majority of the text reiterates knowledge you already have. With a centralized tree, overlapping knowledge is slim to none

If you are an experienced researcher, you might be concerned that this website caters to much to newcomers and thus won’t be as effective. Instead, however, the map takes care of providing the context for your work by default so you can stop repeating yourself and get more done.

Images

If you want to try it out for yourself, this is the link: https://aisafetysphere.com/.

How you can help:

Try out the site and give feedback
I am in need of people to help further develop the site. Email me at aisafetysphere@gmail.com if you are interested.