aivas-devel Mailing List for A.I. Virtual Assistance System

Aivas has to learn to make decisions based on all "monitored inputs". What
does this mean?

Lets use the simplest example I can think of: I want it to learn when to
turn on and off a single light, and more specifically, let's say it's the
dining room light, and I want it to learn to turn it on only when it's time
to eat (which may or may not vary from day to day) and if we are eating in
the living room.

If mister house has an control of all lights in the house, and each time
they are turned off or on, it is done through mister house, then a state
change event is done. If Aivas checks (say every minute) for state changes,
then it can build a state table of measured values (i.e. each light has a
state and the time is an input). Aivas should be able to learn my pattern of
turning off other lights as I move to eat at certain times. It should learn
to predict my behavior (by using fuzzy logic and predictive inference) and
determine that the probability of my wanting the dining room light on is
high when it's typically dinner time and I start turning off other lights.

I'm working on building a framework that can check the state of all objects
in mister house and then figure out how to build an inference engine (this
will probably/potentially use a neural network) to analyze that data and
make a prediction about when state changes should occur.

Essentially, Aivas would have to make a decision every minute about whether
to change any/all object states based on previous experience. I'm thinking
the learning phase could easily be broken into a sudo phase, where Aivas
makes decisions, but doesn't implement them, and then the feedback loop lets
her know if she was correct (oh and since I use a female voice for my house,
I'll be calling Aivas a her, feel free to switch as fits your preference).

-- 
Alex Hardman