Event Cognition Knowledge Representation
In order to build Personal Artificial Intelligence we need to give software the tools to model and understand the world in a similar way our brains do.
One way we might be able to do this is creating an “Universal Dataset”, a dataset capable of integrating all other datasets, and all the information we know about this universe, and imaginary universes. Including contradictions, missing information and uncertainty.
Universal Dataset
We can represent knowledge as “Multiple streams of Functions (assigned to Events) that get Properties (assigned to Elements) as input returning a modified Property as result.”
“The Event Horizon Model proposes that as people experience activities, they
segment them into discrete events.[…] At any timescale, the current event is actively maintained as a working memory representation, and at the same time a long-term memory representation is constructed that can provide a permanent basis for retrieval of information about the event long after it is over.” “Event Cognition” (2014)
This model proposes that our brain understands the world using the Events as units.
But what is an event?
“A segment of time at a given location that is conceived by an observer to have a beginning and an end” (J. M. Zacks & Tversky, 2001)
Building blocks
This knowledge representation consist in four elements: Events, Elements, Properties and Functions.
An Event is an experience (present or past, real or fictional, ours or from others). An Element is an actor (a human or a thing) that will participate on many Events. An Element have a set of Properties defining it (size, shape, color, position on the ownership space), there is nothing outside properties defining an Element. Finally a Function is a mathematical operation taking Properties as input, and returning modified Properties as output.
Where:
- Each Property can be owned by one or more Elements
- The Properties can be modified using Functions
- Each Function can be owned by one or more Events
Also:
- Each Property can be part of other Properties
- Each Function can be part of other Functions
- Each Event can be part of other Events
Example
Imagine we want to introduce the following information to our Knowledge Representation: “The sky is blue”
The most common way to represent it is using the structure: sky -> color -> blue. But on the Universal Dataset things work differently, the information we should introduce is, in fact a question: “What is the most probable color I will see looking at the sky?”.
We will feed the Dataset with instances of events when we (or others) observed the sky. The color will be encoded using vector, for example RGB.
Then, when we ask the PAI “What is the color of the sky?” the inference engine will recall all (or a big number) of instances of observations of the sky when the color was perceived and answer with the most common values, or an average (that might not be blue).
In this example the Element is the Sky, the Property is the Color, the Function modifies the Color to match the perception, and the Event is the observation of the Sky.
The next morning was a midsummer’s morning as fair and fresh as could be dreamed: blue sky and never a cloud, and the sun dancing on the water. — The Hobbit, J. R. R. Tolkien
We are introducing a “fake memory” into the PAI of it directly looking to the sky (Experience model). We can also represent the “memory” of the PAI reading about that experience on a book (Situation model). In this case the Event is the observation (reading the book) of a second Event where a character is perceiving the Color.
- More information about the motivation behind this project: Democratizing Artificial Intelligence
- More information about the Personal AI
Eibriel
eibriel.eth.limo
