DIAMOND / RELATIONAL LEARNING
Better judgement for a different situation.
DIAMOND Relational Learning Software is a proposed research direction for AI that can recognise when an approach fits, understand when it does not and adapt more usefully to unfamiliar tasks.
Experience is richer than a right answer.
Two approaches can each be useful in different circumstances. A solution that succeeds in one setting can become the wrong choice when the task, scale or expectations change. We are interested in AI that becomes better at recognising those differences.
Our relational-learning ambition is more flexible, context-sensitive capability: intelligence that can make more useful choices, benefit from both successful and unsuccessful experience, and adapt without treating every past answer as universally applicable.
INTENDED CAPABILITIES
The capabilities we want to advance.
Why this matters.
Greater adaptability could make AI more helpful in research, software development, planning and complex creative work. It could also help reduce repeated mistakes and unnecessary effort. These are research goals, not measured performance claims.
A method that is useful for a small one-off task may be a poor fit for a large repeated workload. We want AI to become more sensitive to that difference and more capable of choosing appropriately. This example describes the intended benefit, not an explanation of the underlying method.
An invitation to investigate.
We are interested in researchers, model developers and evaluation partners who share an appetite for better questions about adaptation, transfer and useful learning.
In development. Relational Learning is a proposed research direction, not a currently available product. Adaptation, transfer and efficiency benefits remain goals to investigate; no ASI outcome is promised.