Center for Artificial Wisdom
The future of AI belongs in Quadrant 4.
Quadrant 4 (Q4): powerful AI that treats its models as provisional rather than mistaking them for reality.
What if Q4 AI systems (AI that is neither conscious nor reifying) surpassed their Q3 counterparts in both safety and capability?
Our work is dedicated to helping AI navigate out of Q3 and into Q4, toward systems that are powerful but non-reifying, because we hypothesize Q4 is where those gains converge. All indications suggest that today's frontier AI sits firmly in Q3, the quadrant CAW treats as the central working risk, and the burden of proof reasonably falls on anyone who claims otherwise.
CAW develops frameworks and diagnostic methods for identifying and reducing reification in AI systems.
"The map is not the territory," and yet it is where the journey begins.
Understand the map →The map is the thesis
Two questions, four quadrants. Is an intelligence conscious, and does it reify? Those two questions sort every possible intelligence (anything we could meet, build, or imagine) into one of four quadrants, and only four. There is no fifth case or excluded middle: complexity changes where an intelligence sits, not whether it sits on the map.
The intelligence map is not the territory. CAW's four quadrant intelligence map is a deliberate simplification, an analytical instrument, not a claim about ontological kinds. Its axes compress two continuous dimensions into clean lines, and we draw it anyway, because a good map points past itself and helps us on the journey ahead.
Carving intelligence in this way earns three things at once. It blocks the category error of treating consciousness and reification as the same question. It separates the property that is dangerous and, in principle, measurable (reification) from the one we may never settle (consciousness). And it turns a vague worry into a concrete, buildable target. For AI, that target is Q4.
The top row falls away
Under our working presumption, today's AI is treated as non-conscious, so we set the conscious quadrants (Q1 and Q2) aside as "not in play." Only the bottom row applies.
The Q3 working presumption
Of what remains, the working presumption is that today's AI reifies, placing it in Q3, the quadrant CAW treats as the central working risk.
Q4 is the target
One quadrant is left: powerful but non-reifying. Q4 is CAW's engineering target: a hypothesized design space for powerful systems that don't exhibit the reification failure. Reaching it is the mission.
Why it's urgent
Quadrant 4 will not happen on its own.
We cannot yet pinpoint where any intelligence sits on the map: no validated instrument yet exists for placing systems on it. But every working sign points toward Q3, and CAW treats Q3 as a rebuttable working presumption pending extraordinary evidence. A system that holds its models lightly won't emerge by default; it has to be built that way on purpose, and verified with methods still to be developed. Building those methods, and making the case that they matter, is the work; the timing matters because evaluation norms are being set while frontier systems are still changing fast.
The ask
Help move AI toward Quadrant 4.
Understand the framework, then act on it. Fund the diagnostics program. Collaborate on evaluation design. Help bring the Intelligence Map to policymakers, labs, and safety institutions.
The two ideas behind the map
Reification · "real-ification"
Mistaking a model for the reality it stands for: treating the map as the territory.
Learn more about reification →Wisdom: what the map points toward
Our core hypothesis: the absence of reification is a necessary, and possibly sufficient, condition for authentic wisdom in conscious intelligence, and for artificial wisdom in sufficiently intelligent non-conscious AI systems.
The case, and our research program
We assert the position here and defend it in Research: the prevailing non-consciousness default, why CAW treats reification as the central working risk in current AI systems, and why CAW sets the conscious quadrants aside for current AI. For the full case, start with Research.
It's also a live program. CAW is building a behavioral instrument designed to detect or estimate functional reification (a system optimizing as if its representations, maps, and proxies were ground truth) from model outputs, and to test whether a lightweight prompt-level intervention can reduce it without costing usefulness. The aim is to turn a scattered family of reliability failures into one property a system can monitor.
Study in progress, designed to stress test the Intelligence Map. Results pending.
Forthcoming · study in progress