AI + Human Judgment: The Three Exceptions
AI does almost everything better, faster, cheaper — except Taste, Responsibility, and Novelty Judgement. This paper develops the framework that underpins every Thoth product and methodology decision.
Thoth ARKives
Research, development,
and publications.
The ARKives is where we document how we work, what we're learning, and what the edges of AI capability actually look like when you push them rigorously. Part methodology lab, part case archive, part publication channel.
The method
Hemispheric research.
Most AI research follows a single pass: prompt, generate, review, publish. Our approach is different. We call it hemispheric research — a structured process where the creative pass goes deliberately wide (unconstrained synthesis, aggressive conjecture, cross-domain pattern recognition) and the analytical pass goes deliberately strict (rigorous validation, stress-testing against known results, systematic red-teaming).
Think of it as: the right-brain function goes chaotic on purpose — exploring edge cases, unexpected connections, and speculative hypotheses that a conventional analysis would self-censor. Then the left-brain function orders, filters, and checks truths — applying scientific method with full rigour to every surviving thread. The tension between the two passes is what produces results that are both novel and defensible.
This is the same polyspheric AI architecture that powers the ARK platform and every MbM consulting engagement. The ARKives is where we document the methodology itself and publish what it produces.
Current topics
What we're working on.
None of these are published yet — they are in draft or outline, developed through live engagements and internal research sprints. We publish them here as they are finished.
Polyspheric Architecture: How Multi-Pass Reasoning Works
A technical overview of the polyspheric AI architecture that powers the ARK platform. How synthesis, validation, and counter-synthesis passes interact. Where AI excels and where human judgment intervenes.
The Flipped Firm: AI-Led Consulting
Why AI-first consulting inverts the traditional consulting pyramid. How the compounding flywheel works in practice. Revenue sharing, skills capture, and the economics of expertise preservation.
Sprint Methodology: 5–10 Days to Defensible Conclusions
How MbM structures a consulting sprint. Scoping, execution, delivery. What a deliverable looks like. How we name uncertainty. Anonymised case notes from early engagements.
Expert Skill Monetisation via MCP Server Infrastructure
How captured expertise gets distributed through the ARK platform. The MCP (Model Context Protocol) server architecture that routes skills to engagements. Revenue attribution and specialist royalty mechanics.
Consciousness, AI, and the Limits of Pattern-Matching
Where AI pattern-matching ends and genuine understanding begins — if it does at all. A rigorous exploration of what "judgment" means in the context of machine intelligence, framed through the lens of working with AI systems daily.
Film series
The flipped economy.
Short essays on what changes when intelligence stops being scarce. Written, voiced and rendered in-house, on our own hardware — which is the same claim we make about everything else we build.
Watch the series as a playlist on YouTube. Videos here load from YouTube only when you press play — nothing is requested from Google until then.
Contribute
Early access and submissions.
The ARKives is currently produced internally. If you'd like early access to draft papers, want to contribute a case study, or have a topic suggestion, the contact form is the right starting point.