clever brute force of NSF

[hyunjimoon]

innovation comes from detecting lifted tradeoff (reversed causality) while AI cleverly brute forces NSF parallel paths

N: possible embodiments of need

• need of vison pro is embodied differently e.g. X, Y, Z, alpha generation
• need of angie's research is embodied differently for [agritech, biotech, healthcare, mobility startup] in [Boston, SG, Korea]

S: possible embodiments of sol

• sol of vision pro is embodied differently with technology e.g. SwiftUI, RealityKit/Unity (3D render/platform), Metal (graphic)
• sol of angie's research is embodied differently with technology e.g. inverse RL, PPL, dynamic simulation (auto-sequencer)

F: possible embodiments of fulfillment

• sol of angie's research is embodied differently with fulfillment strategy (collaborate/compete process and product with Tom, Scott, Vikash, m3s etc)

There are some downside of having theory (e.g. Theory-Driven Reasoning about Plausible Pasts and Probable Futures in World Politics: Are We Prisoners of Our Preconceptions?), but no 🔍magnifying glass #276!

Cognitive theories predict that even experts cope with the complexities and ambiguities
of world politics by resorting to theory-driven heuristics that allow them: (a) to make
confident counterfactual inferences about what would have happened had history gone
down a different path (plausible pasts); (b) to generate predictions about what might yet
happen (probable futures); (c) to defend both counterfactual beliefs and conditional fore-
casts from potentially disconfirming data. An interrelated series of studies test these pre-
dictions by assessing correlations between ideological world view and beliefs about
counterfactual histories (Studies 1 and 2), experimentally manipulating the results of hy-
pothetical archival discoveries bearing on those counterfactual beliefs (Studies 3-5), and
by exploring experts' reactions to the confirmation or disconfirmation of conditional
forecasts (Studies 6-12). The results revealed that experts neutralize dissonant data and
preserve confidence in their prior assessments by resorting to a complex battery of belief-
system defenses that, epistemologically defensible or not, make learning from history a
slow process and defections from theoretical camps a rarity.

four category to align agents across time and space, but "continuity from bayesian update" itself may be bottleneck AI can help improve

[hyunjimoon]

Wolfram’s notion of “pockets of computational reducibility” from computation of biological evolution, in an otherwise irreducible system parallels the logic of simulation-based calibration (SBC): in a vast landscape where you cannot globally predict outcomes, SBC carves out local checks that ensure your Bayesian computations match the generative process in a well-defined test space. Even though the overall phenomenon—be it entrepreneurship, a large hierarchical model, or biological evolution—remains too complex to fully skip ahead, systematic measurement or theory-building (like these repeated fake-data checks) can isolate manageable sub-problems where you can test coherence and partially validate your approach. This is effectively “clever brute force”: brute force in that you iterate multiple simulations, and clever in that you focus on a narrower component (calibration) that can be reliably probed in a sea of potential unpredictability.

---

Above synthesizes three inputs:

1. Andrew's high evaluation of my interpretation of test statistics of SBC:

“Cleverly brute force”
Posted on March 20, 2022 9:30 AM by Andrew
In a meeting with Paul and me, Angie proposed an idea that she described as allowing us to be “cleverly brute force.”
I like that. In statistics we want everything to be brute force. Statistics is the science of defaults. A clever solution is a kind of cheat, and if we have a clever solution we want to figure out how it works, so we can make it brute force.
But . . . we can use cleverness to come up with a brute force solution. Hence, cleverly brute force.

2. JB's pointer to Wolfram's computational approach (link above) and his interpretation that context embedding reduces complexity
3. @jeff-dotson's high evaluation on Wolfram's writing on pockets of computational reducibility among the sea of computational irreducibility. From 1's clever brute force, Clever = reducibility (closable pockets), brute force = irreducibility.

[hyunjimoon]

From https://statmodeling.stat.columbia.edu/2021/01/18/simulation-based-calibration-two-theorems/, I asked gpt which of the theorems Andrew conjectures is more relevant. It answered: they both illustrate “partial checks” rather than a full global proof, but Theorem 2 (the hierarchical-model case) more directly captures that idea of a large, irreducible system where you still isolate a manageable sub-problem (the local parameters (\xi)) and approximately calibrate that piece. It’s basically Wolfram’s “pockets of reducibility” in action: you can’t solve the entire system outright, but by dividing the parameter space and focusing on the (\xi)-part, you get a locally valid check even if the full joint distribution remains too big or complex to handle all at once.