Osiris Simulation

Alongside practical applications such as those mentioned above, we have used network-theory abstractions to design and build a miniature Osiris, which serves the double purpose of testing Osiris “policy” settings such as the reputation system and offering the same type of analysis that the full Osiris will offer but in a miniature form that can be run on an analyst’s personal computer.

This miniature Osiris is a small market in which programs may send feedback to each other through price signals. Price signals serve as an assignment of credit. This simulation allows Python programs, models, and AIs to coevolve. It can be used for any coevolutionary purpose.

This simulation shows aspects of cognitive synergy between agents having emergent cognitive properties above and beyond those of the individual agents.

Because the agents in this simulation model can be made to run various AI programs, the simulation can also be made to do other things besides simulate a realistic Osiris. For instance, if you simulate a Osiris where all the AI agents are running clustering algorithms, then the simulated Osiris becomes essentially an emergent-level clustering meta-algorithm.

One can carry out various AI tasks (like clustering or prediction) or real world system–modeling tasks (e.g., modeling a political system or a real-world market) by the methodology of creating a simulated Osiris full of simulated agents running actual AI algorithms that are configured and distributed in a certain way. This approach can be used to especially good effect in situations where one AI agent’s modeling process can benefit from feedback from another AI agent’s modeling process.

For example, one such application is feedback between the interpretation of data and multiple overlapping disparate models of the processes that created the data—together, the data and the models create a better model as a whole. In our work on this sort of data fusion through feedback, we use specialized data processors and models that are designed to accept and adjust to feedback. These include a clusterer that can take in exemplar inputs and an agent-based model with special data-absorbing properties that integrate theory with data. A similar approach can be taken with more sophisticated AI methods, such as coevolutionary neural networks, that put parts of neural networks together with other types of AIs in a connectionist ecosystem.