Our agent technology is the result of many years of applied research and engineering of complex systems. We particularly focus on machine understanding through representation and simulation.
I KNOW WHAT YOU MEAN
Menu
I KNOW WHAT YOU MEAN
Our agent technology is the result of many years of applied research and engineering of complex systems. We particularly focus on machine understanding through representation and simulation.
Data-driven machine learning has become very good at identifying patterns in vast amounts of data. However, it remains unable to understand abstract concepts and so fails to produce explainable results.
Introduction
Understanding why events happen the way they do is the key to cognitive intelligence. However, for an AI to understand causation, it must understand concepts—not just algorithmically predict the following number in a set.
Nowhere is this clearer than in our natural human languages, which encode meaning about the world around us into symbols (words) and rules (grammar). Today’s machine learning algorithms can’t understand these layers of abstraction and logic because far more sophisticated intelligence functions must be in place first.
Cognition refers to the mental faculties of Learning, Reasoning, Memory, Language, and even Imagination. At Titan Virtual, our cognitive agents have these capabilities because the way information is represented internally in their memory is built around the Concept—the same unit our human minds use to solve problems.
Our cognitive architecture hosts an interwoven web of knowledge that grows instantly and infinitely from any English text. Our cognitive agents simulate and navigate this knowledge using natural language and thinking with the same concepts we do, allowing them to use causality to solve problems and predict outcomes.
Breakthroughs
Our technology (transcriber model) is built upon several proprietary breakthroughs in natural language understanding and generation, concept-based representations, and event-driven causal reasoning.
|
Technology
|
Transformer Model
|
Transcriber Model
|
|---|---|---|
|
Model Type
|
Sub-Symbolic
|
Symbolic (Not GOFAI)
|
|
Architecture
|
Deep Neural Nets
|
Dynamic Knowledge Graphs *
|
|
Algorithm
|
Sequence-2-Sequence with attention
|
Sequence-2-Event and Event-2-Sequence *
|
|
Key Insight
|
Pattern: Relationship between Data
|
Event: Relationship between Concepts *
|
|
Data
|
Pre-trained (billions of parameters)
|
Pre-set (<1 million parameters)
|
|
Main Application
|
Sentence Completion and Language Translation
|
Semantic Search, Reasoning, and Action * (who/what/when/where/how/Why)
|
|
Model Update
|
Non-Real-Time (Retrain)
|
Real-Time *
|
|
Explainability
|
Not Possible
|
Possible
|
|
Scalability
|
Finite
|
Infinite
|
Products
We integrate several cognitive skills into a single entity.
Our agents autonomously apply a series of cognitive skills and instantly transform a linguistic statement into a representation that enables them to understand the information, respond to queries, predict outcomes, or carry actions.
Insights
Our unique approach to natural language, i.e., we extract facts, not patterns, allows our agents to learn instantly and continuously from text, understand content in context, and autonomously generate answers.
The baseline architecture of knowledge in our system is a novel, highly dynamic ontology representing the world as our minds do, a set of concepts and the events that form the relationships between them.
Our cognitive agent infers cause and effect from closely and distantly chained events. The agent can also transfer knowledge to new, unseen domains and apply this knowledge successfully.
Access
We will be releasing an API for a cognitive agent. This general-purpose API provides a “text in, text out” interface for teaching the agent and asking questions about what it knows. You will soon be able to request access to integrate the API into your environments and applications.