Coming soon

The Atlas: seeing the problems and the world the AI evolves in

An AI must be able to visualise the problems and the world in which it evolves: the Atlas is that representation, and the one on which I am building my AI. It is enriched by every certified result and every dead end; my challenge results were obtained by recording them in this way. Its public part, the open Atlas, is in preparation; it is the subject of this page.

The Atlas is not a black box: every entry carries its certificate and can be read. Entries can be added, removed or replaced; since every computation replays identically, the effect of a change is measured exactly: change, replay, compare.

In mathematical research and scientific computing, a lot of work is redone without anyone knowing: a solved problem is attacked again, an explored path is taken again, a dead end is rediscovered. Failures in particular get lost: what did not work is rarely published.

I am preparing an open Atlas of certified computations. Every result will appear with its certificate, checkable by a small public program. Areas explored without success will appear too, in two clearly separate forms: “nothing exists here, and here is the proof”, or “nothing found within this computing budget”. A failure becomes an explored area, useful to everyone.

What the Atlas will make possible

• Never recompute what is proven: a certified result is borrowed, not redone.
• Avoid dead ends: a lead already explored without success is flagged, with the computing budget spent.
• Start from an established result: every result can be re-checked on any machine, and a researcher can extend the work where it stops.
• Check without trusting: a small public program checks each certificate in seconds.
• See the links: when a result uses another, the Atlas shows it; if one falls, you know what depends on it.
• Save computing power and time: for researchers, leaderboard referees and research funders.