Motivation: For the sake of advancing AI

We are doing this R&D work partly to make demonstrations of more advanced AI systems. We expect that our stance on AI will not necessarily be a popular one. But this is an important “minor” strand of AI research dating back to Alan Turing:

"As time goes on the [computer] itself will take over the functions both of [programmers] and of [users]…The [programmers] are liable to get replaced because as soon as any technique becomes at all stereotyped it becomes possible to devise a system of instruction tables which will enable the electronic computer to do it for itself. It may happen however that the [programmers] will refuse to do this. They may be unwilling to let their jobs be stolen from them in this way. In that case they would surround the whole of their work with mystery and make excuses, couched in well chosen gibberish, whenever any dangerous suggestions were made." -- Alan Turing, 1947.

On average, advanced AI would bring in new ways of working, and would facilitate broad access to high-quality training. This agenda could serve to focus the mind of technical workers, but not many are pursuing it presently.

Motivation: Technical experiments become easier

Even in the present time, without relying on any speculative AI futures to magically appear, we can benefit from pursuing the agenda above. Accordingly, we are doing some applied work with existing software that will give us a set of further tools and levers to work with.

Representative Prior Work


PlanetMath users created a reasonably large informal mathematical knowledge base together. On the way, we came up with several technical demos and sketched possible previews for upcoming features. One possible direction of work we looked at would be to focus on building a comprehensive category theory knowledge base.

Modelling the way mathematics is actually done

In this paper, we talked about how mathematics is situated somewhere in between ‘games’ and ‘storytelling’ in its complexity. We proposed to build computational models of informal mathematical reasoning. Subsequent work continued on in this direction, using ideas fromdialogue games and argumentation theory more broadly.

A sketch of a plan

So, having gotten together around these ideas, we’re having online chat, frequent short meetings. We’ve talked about maintaining a blog that would describe what we’re learning and developing. So, roughly speaking, we will try to develop a curriculum through the blog. We also have this wiki, that any of us can edit, which we can use as a staging ground for more developed blog posts. Our thought was that blog posts might move in the direction of more developed outputs, whether products or research papers. We want to use some ideas adapted from Scrum to build a shared awareness of what’s going on. However, we want to be careful not to become “managerial” since everyone is currently here as a volunteer, working on topics of his or her own interest. We want to provide mutual support and fun. Reflection, whether in writing, or by recording and listening again to conversations, should help with that. We are not constraining things to come out in a purely structured curriculum, or any other form of product development. “Users” and “customers” may appear as we release things we are happy with and expand our little community.

“Rousseau says, someone who has been properly educated will be engaged in society, but relate to his or her fellow citizens in a natural way. ... We naturally look after our own preservation and interests. By contrast, amour-propre is an unnatural self-love that is essentially relational. ... Thus, amour-propre can contribute positively to human freedom and even virtue. Nevertheless,amour-propre is also extremely dangerous because it is so easily corruptible. ... In its corrupted form, amour-propre is the source of vice and misery, and results in human beings basing their own self worth on their feeling of superiority over others.” — IEP

A possible formulation: short correlated sprints as opposed to random behaviour

“Two people working together 4 hours a week for two weeks” could serve as an approximate unit of work. Once we have amassed a few outputs from this kind of effort, we will have some evidence of the kinds of things that we can realistically achieve. So far, our workflow has been more based on solo activities and informal conversations, but short robust team-ups continue to be an option!

Hypothetical conversation: In my next post I want to integrate something that I learned from you about PL. I want to drive in the direction of synthesis, as hard as I know how to right now. This depends on everyone having free time to invest in this. Start a blog where we think about what's the overlap in terms of learning?