DATATOMIC / CONCEPTUAL LAB NOTE
Are we building intelligence on the right foundations?
A starting framework for separating better algorithms, different mathematics, and different physical machines.
EXPLORATORY NOTE · PROPOSED INQUIRY
An introductory framing of the question and possible next steps. No completed study or experimental result is reported.
The question beneath the benchmark
The central question at Datatomic Labs is whether our machines use the right physical and mathematical foundations for intelligence. To investigate it, we need to distinguish three proposals: changing an algorithm within the same representation, changing the representation itself, and changing the physical substrate that implements it. Each proposal has a different burden of evidence.
Make a small claim first
A useful starting experiment would choose one task where context changes a judgment. We could implement two models, specify the same observations and evaluation budget, and compare their predictions on examples withheld during development. The point would be to identify what one representation makes easier to express, and whether that difference survives a fair comparison.
What would count as progress?
A more compact explanation, a reproducible improvement, or a clearer account of failure could each be progress. A new vocabulary by itself would not establish a better model. Nor would a successful mathematical model demonstrate that the brain uses the same physical mechanism.
Next notebook artifact
The proposed next step is a task specification, two explicit models, and a comparison protocol. This essay frames a research question; it reports no completed experiment or validated result.
CONTINUE THE INQUIRY
Read the research direction