Mission

The PNW AGI Group seeks to work out the first principles of Artificial Hyper Intelligence: how ideas stand to objects, how universals stand to a world of particulars, how a thing is known by likeness and mereology, how more than one kind of causes operate, how judgment sits on the objective, how the present holds past and future, and how reasoning, understanding, and judging operate altogether.

In 2019, Mike Archbold [et al.] opened what became the Pacific NorthWest AGI Forum.

“The appearance of being is not the same as the being of appearance…”

Mario Bunge’s working definition: metaphysics is the study of the most pervasive features of reality.

Why first principles

Aristotle is often credited as the main reference for the introduction of Term Logic in the West, which lead to a fairly rigorous categorization of reality: a metaphysics. Some contempt themselves with the thought that metaphysics books stand after, on the shelf, the physics ones in most libraries.

Plato recovered the spirit of Socrates’ dialogues and put the Form at the center. The question that followed was: how relevant is logic in a post-symbolic era?

Metaphysics fell out of fashion with the rise of the scientific method. It is not deemed proper science. A more socially acceptable term, which has been popularized by Palantir, is ontology.

We keep investigating.

The gap

Some ordinary statements translate cleanly from an idea into a program, but Physical Reality has mostly particulars. Physical Reality is composed of concretions. Mind appears to form concrete, though immaterial, ideas. Most programming forms abstractions.

Identity is the same kind of trap: we do not know what something is by their identity. We know it by what other things it is like and unlike. Properties describe likenesses and opposites. Deep Learning lives in that neighborhood: similarity, contrast, the space of near and far, yet still does not settle what a thing is.

Time is worse. Augustine’s mind model is always stuck in the present: is recollecting, attending, or expecting. Recognizing efficient causes is fashionable, but hardly enough. Objectivity is the part of the world we are supposed to agree on. Conflict and complexity start in human bias.

Our work

The world of particulars. The world external to mind contains nothing whatsoever other than particular cases. Without generalization, human activity is inconceivable, and little automation as we know it would be possible. What followed during the years reached across psychology, neuroscience, and cognitive architectures; Chollet’s measure of intelligence; compression as a definition of property; pattern discovery as unsupervised generality.

Logic after symbols. Non-axiomatic reasoning with long-term memory and attention: the essential nature of reasoning and rationality, then a survey from GOFAI through NARS to what large language models actually do when they “reason”.

Understanding, judging, reasoning. Kant’s centrality thesis against the climate in which a machine judgment is correct if it matches prior human judgments or hits a reward. Emily Bender’s stochastic parrots: is there room left for a novel judgment beyond the training data? Can a machine make an authentic judgment, or is it forever reckoning, in Brian Cantwell Smith’s sense? How is a judgment related to understanding and reasoning?

Time. When people worry about time in AI, they usually mean how fast the software is. If it is running now, that is taken to cover the present. Hugo Latapie on Korzybski’s time-binding as the human capacity to bind past and future into the present, and what of that we would actually want in a machine.

Then what? Principles and primitives for super-intelligence, natural informatics as natural philosophy. AGI is usually discussed as a capability threshold: reason, learn, plan, act across domains. Before that threshold, a second question. What would make any generally capable intelligence worth trusting? Sane General Intelligence: grounded, corrigible, bounded, reality-attuned. Technorealism: capability is not authority. That is subjectivity as a judgment upon the objective. More intelligence does not automatically route to better outcomes.

How we work

The purpose of the group is to explore all relevant issues related to the development of AGI, encompassing all disciplines. The starting points are many: computer science, neuroscience, psychology, sociology, mathematics, linguistics, electronics, physics, robotics, philosophy, etc.

We meet as the Pacific NorthWest AGI Forum. Members contribute their architectures, papers, and intuitions. Guests bring books and systems. Some nights are a talk and a question. Some nights are a workshop. Some nights are a jam: ideas, sketches, demos, elevator pitches, your consciousness definition, concepts reduced to practice. In person in Seattle, in Vancouver, online, hybrid.