Intelligence is one of the most consequential subjects in science. Understanding how it works may change what we are able to build.
Aurisus was founded around an interest in intelligence and the mechanisms that make learning possible. Our research began with an investigation into how intelligence arises in nature. That work shaped the questions we pursue today, including how knowledge is acquired, organised and used to make sense of the world.
Our current programme, Project Insight, examines continual causal learning in artificial systems. Drawing on biological learning and memory, it investigates how systems might build an understanding of cause and effect, preserve what they have learned and revise that understanding as experience changes.
The programme is an early step in a much broader effort to understand the principles of intelligence and their potential applications in technology.
Project Insight
Continual Causal Learning.
Learning from experience requires more than recognising patterns. A system must also develop an understanding of how events relate, what produces particular outcomes and when its existing knowledge needs to change. Project Insight investigates how biologically inspired learning mechanisms could support these abilities in artificial systems.
Its central research question is:
How can biologically inspired learning mechanisms enable artificial systems to continuously acquire, consolidate and revise causal knowledge of their environments, while preserving useful prior knowledge and generalising to unfamiliar situations?
The investigation is organised around four connected challenges:
- Causal learning: Discovering cause-and-effect relationships from experience, beyond the recognition of statistical correlations.
- Knowledge consolidation: Integrating new information with existing knowledge so that learning remains useful over time.
- Knowledge revision: Updating existing understanding when new evidence challenges what a system has learned.
- Generalisation: Using acquired causal knowledge to reason about situations the system has not encountered before.
We are examining research in neuroscience, continual learning, causal representation learning and world models. These fields offer different perspectives on how knowledge is formed, stored and applied. Our work will bring relevant findings together to identify limitations and define experiments around them.
Project Insight is currently in its early research stage. We are reviewing existing literature and assessing potential research problems, methods and experimental requirements. The specific mechanisms and methods to be investigated will be determined through this process.
Research Foundations
Scientific foundations and related fields.
Understanding continual causal learning calls for perspectives from several disciplines. Our initial research draws on:
- Neuroscience: How biological systems learn, form memories and change their connections through experience.
- Continual learning: How artificial systems acquire new capabilities while retaining knowledge learned previously.
- Causal learning: How systems discover, represent and reason about cause-and-effect relationships.
- World models: How systems represent their environments and learn the relationships that govern changes within them.
- Memory systems: How information is retained, organised, consolidated and retrieved.
Together, these fields provide the starting point for Project Insight. Their concepts and methods will be evaluated according to the research questions we pursue.
Our Approach
We approach intelligence as a scientific question, one that requires careful study and testing.
Our process begins with the existing body of research. We examine established findings, compare approaches and look for questions that remain unresolved. From there, we develop hypotheses and design experiments that can provide evidence for or against them.
We place importance on documenting how research is conducted, what the results show and where the limitations lie. This includes distinguishing established scientific findings from our own hypotheses and experimental observations.
We intend to build our work through cumulative research, with each investigation contributing to a clearer understanding of intelligence.