Executive Loop

The Executive Loop The executive loop serves as the primary control for OSLO’s behavior, decision-making, and generalized inference. Everything perceptible to OSLO exists as a state object, stored in a list.

State Object A state object possesses its own transition vector, as a component of a broader transition matrix, and a multitude of flags and functions. These include a plaintext name, a lambda function, a plaintext verb cloud, a plaintext object cloud, a plaintext descriptor cloud a binary flag to determine whether it is an action or a stimuli, a binary flag to determine if it is a ‘reward state’, and a flag to determine if it returns anything.

Causal Pointer A causal pointer is a set of clocks which upon the trigger of a state object, begin counting down, in the short, less short, long, and very long term, similar to the way a human perceives time. On triggering any state s, four pointers to every other state aside from s are created. If the target of any of these pointers are triggered within the time frame they represent, the occurrence is recorded. Otherwise, they expire. All pointers are stored in a queue.

Reward States & Drive Values A reward state is a special form of state which represents the achievement of desirable behavior. OSLO possesses a set of “drive” values, for curiosity, sociability, survival (remaining online), and efficacy. Particular reward values can be triggered manually during operation to ‘train’ OSLO in real time. There is also another value, ‘sleepiness’. This is because some inferences are too computationally expensive to make during active operation, and instead need to be intentionally done during downtime to gain a more accurate understanding of its surroundings. During normal operation, each drive decays over time by a factor d, and sleepiness increases. States are triggered in such a fashion that the maximal chance of triggering a reward state is favored, and that sleep is prioritized within a set period of time.
Using the pointers, the probability of transition from one state to another whatsoever, and within a given time frame can be recorded. This permits the formation of a transition matrix.