A   G   I


Research Subject

Recursive self-improvement
- Acceleration of self-improvement is a key to the Singularity.


Optimal AI self-modification
- What are best ways for a program to modify itself? What are good measures of good self-modification? We take experimental approach to the problem of self-modification over theoretical research, using the financial market as a testing environment.


Systematic exploration of the edge of chaos
- We are searching for ways to tackle unavoidable problems caused by introducing feedback loop into a system, such as convergence to trivial states or useless divergence.

Research Subject

Cooperative AGI
-
 It would be impossible to have a complete control over superintelligent AI agent. We are therefore searching for means to reach peaceful coexistence. We are building theoretical foundation for a measure of cooperation also. 
More on this topic ( Click left image. )

--------------------------
AI death
- "Death" is a key attribute of a living entity. We are exploring ways for a self-modifying program to run only once and not being able to run again under any circumstances once it stops.
AI Nirvana
- What could be the "best" objective function an AI can possibly have for itself? We are searching for means to endow an AI agent to have searching capability for such objective function.
Cooperation curriculum
- What are good curriculum for an AI agent to learn cooperation? An example could be learning to cooperate from non-zero sum game and eventually to zero sum game.