inference system

The inference system is a system and a framework that aims to capture to the fundamental qualities necessary for a computer system to be able to infer the user intent. It aims to be “ahead” of the user and figure out what it wants in a non disruptive way.

An inference system is a fundamental part to get closer to human-computer symbiosis

forced categorization of its fundamental qualities:

• abductive inference logic
• seamless inference
• transparent inference
• ubiquitous inference interface
• semantic inference?

abductive inference logic

abductive inference logic is the logic behind the inference system‘s logic (meta-logic).

abductive inference logic aims to produce a logic system that provides the best educated guess (abductive inference).

The inference system aims to provide an adaptive automation of decisions and computations in scenarios that has not experienced yet based on all the information it has available. This implies that the inference system is constantly faced with uncertainty.

inference system goal is to stablish a logic that minimizes disruption and maximizes outcome by inferring information from its components and the particular context of execution.

abductive inference logic factors the certainty of rightfulness of all the potential outcomes in relation to the user intent as well as its catastrophic potential, and makes a decision.

• certainty referes to the likelihood of the outcome of matching the user intent.
• catastrophic potential refers to the likelihood of the outcome doing harm not desired under any circumstance.

inference system outcome does not require perfect rightfulness, only functional enough, but it does require traceability of the reasoning for its actions and a correcting mechanism for its future improvement (transparent inference)

First it should aim to produce an outcome using deductive inference exclusively on previous certain user input. If it can’t guarantee a rightful outcome it should aim for the best possible guess that is not catastrophic, by balancing all its multiple sub-systems. If none of the above is possible user input should be requested.

1. Is there a rightful or functional enough outcome?
   1. Yes
      1. Is it above catastrophic levels?
         1. Yes: Ask user input
         2. No: Use it
   2. No: Ask user input

functional enough and catastrophic are user-inputs that can be progressively tunned.

E: The inference system have high certainty that the user intent is to publish a specific meaning unit (the user has expressed it), but there is a high chance that the meaning unit may be sensitive.

inference logic in traditional algorithms and machine learning models

traditional algorithms could be understood as working purely through deductive inference inference. Their computations are always correct, but requires its data to be concrete and certain. They can’t work with ambiguos information, which is the natural form of language and therefore user input.

E: A math calculation, rendering information, an “if” statement.

Similarly machine learning could be understood as inductive inference. It works exclusively with probabilities, it is much more similar to how meaning and language works with humans, but it implies that its outcomes are not certain. It also transparent inference, reducing user trust and capacity for error-correction.

E: Recommendation and search models used in popular services.

seamless inference

seamless inference is the quality of an inference system of working with the user without obstructing the user intent.

Requesting user input requites a significant cognitive effort due to cost of context switching which distracts from the original user intent and branch of thought.

seamless inference aims to minimize the amount of non-necessary user-inputs.

seamless inference aims for an interaction that can tolerate ambiguity as well an can infer most of the meaning from the context. In many ways it aims to work using similar mechanisms than natural language does.

seamless inference, implies an ubiquitous inference interface with automatic behavior at the cost of a non-rightful outcome. The opposite would be a system that while “automatic” if first requests user preference and configuration in order to ensure rightful outcomes.

seamless inference requires the capacity for the user to adjust abductive inference logic in order to progressively maximize its adaptability and the inferred certainty of user intent.

transparent inference

transparent inference refers to the quality of inference system to provide to the the user the necessary primitives to form an understanding of the ongoing computation and how they relate to the user interface elements.

transparent inference refers to logic that is traceable, but also to the capacity for the user to navigate, find and understand this logic.

E: The user should be able to associate a system sound notification with a visual clue (like a badge). The visual clue should be a responsive user interface component that by interacting the user can trace back intuitively navigate all the details for why the notification was triggered, potentially up to the code itself.

E: The meaning behind a word should be accessible by interacting with the word.

transparent inference is necessary for the user to trust the inference system and to correct non-optimal behaviors.

transparent inference is particularly critical when the system incorporates machine learning like sub-systems. While most machine learning lack traceability the system can make the user aware of it.

E: The omnibar may auto-complete/suggest a meaning unit using abductive inference logic, which may be some sort of weighted algorithm that includes user preferences, user previous actions or natural language processing. The user interface should express the degree of certainty about the suggestion, and the should be able to interact with it and see such algorithm with the current values being used.

transparent inference should be implemented as much as possible at system level and not be left-out to each individual transform. When is not possible guidelines should be put in place to help the developer and the user.

ubiquitous inference interface

In a human-computer symbiosis the interaction with the computer system requires a single user interface that works across all the information domains that the user works with at any given time. ubiquitous inference interface represents this quality.

ubiquitous inference interface is necessary because all information domains are interconnected and feed into each-other (including all user inputs). They’re the contextual information necessary for the understanding of the focal information.

In traditional apps, the omnibar is representative of the closest user interface element to provide a multi-information domain functionality, although only present in limited scenarios.

E: command line interface, web browser’s search/url bar, home-assistants (Siri, Alexa), launchers....

inference system user stories

The inference system should infer:

• The right level of abstraction of a meaning unit
• The best transform to visualize a meaning unit
• The parent predicates that belong to a meaning unit based on the predicates that I assign
• What predicate of a meaning unit to use in a computation just by feeding the meaning unit as argument.
• The optimal behavior based on past events and similar situations.