Law V
The Targeting Problem
Misaimed capability always wastes the aimed-at effort, but harms the system only when the misfire is hard to detect or re-aim — so treat a suspected wrong target as a hypothesis to test, not a verdict.
The claim in full
When a system underperforms, the failure is frequently misdirection rather than insufficient capability — the system has the capacity but aims it at the wrong level of abstraction. AMENDED 2026-09-02: this is a diagnostic HYPOTHESIS, not a certainty. Misaiming always wastes the aimed-at effort, but whether it harms the SYSTEM is conditional on whether the misfire is cheaply detectable and re-aimable: detectable and re-aimable, misaiming becomes search and the system improves (drug repurposing); undetectable within the aimed-at frame, it becomes harm (leaded petrol); detectable only late and not re-aimable, it becomes expensive neutrality (microfinance) — a third outcome the original "worse, not better" dichotomy omits. And because a crude-looking target may be a sufficient statistic for the objective, which you usually cannot know in advance (LLM scaling on next-token prediction), the right response to a suspected misaim is to test it with a cheap independent signal, not to assume adding capability will harm.
What would falsify it
A law that cannot fail is not a law. Each of these is the observation that would break this one, written before the evidence was looked for, so the frame can lose.
A system where capability aimed at a target that was a poor proxy for the objective (not a sufficient statistic), with the misfire undetectable or un-re-aimable, nonetheless produced sustained improvement rather than waste or harm. Supersedes the prior falsifier ("adding raw capability to a misidentified target led to sustained improvement rather than waste or harm"), which was met by sildenafil (re-aimed into search) and by LLM scaling (a sufficient-statistic target) and could not distinguish a genuinely-wrong target from one that only looked wrong.
How it is standing
Every candidate counter-example logged against the falsifier above, counted. Dismissed means examined and found not to bear on the claim; watching means open.
- Challenges logged
- 0
- Currently watching
- 0
- Examined, dismissed
- 1
- Last examined
- Not yet challenged
A low challenge count is not strength. It can equally mean the law has not been aimed at anything hard enough to break it, which is why the date matters as much as the number. Record compiled 14 Sept 2026.
Amendments
The statement above is not the original. Each row is a time the law was rewritten because it failed a test, with the reason it failed.
-
2 Sept 2026
Two-facet scope clause, from TWO independent pre-registered probes (2026-08-18 S229 and 2026-09-02), each returning ONE CONTESTED falsifier of 5 — NOT clean, so neither moved standing on its own. Law V's falsifier conflates the aim with the system and the actor's targeting with a critic's. Facet (a) 2026-08-18: capability aimed at a misidentified target always wastes that effort, but the SYSTEM is harmed only when the misfire is undetectable/un-re-aimable (tetraethyl lead = harm; microfinance = expensive neutrality, a third state the law has no word for; sildenafil = detected+re-aimed into search). Facet (b) 2026-09-02: when the crude target is a SUFFICIENT STATISTIC for the objective it is not a misfire at all and adding capability succeeds (LLM scaling on next-token prediction; primary-grounded, Bender & Koller ACL 2020 for the named gap, Brown et al. arXiv 2005.14165 for transfer). Both facets reduce to the same primitive as Law II's amendment: whether an independent route exists to tell a real misfire from an apparent one and correct it. Confirmers behaved as Law V predicts (NICE-SUGAR intensive glucose control raised 90-day mortality 24.9%->27.5%, p=0.02; fire suppression worsened megafire severity, Nature Communications 2024). Blocking checks passed 2026-09-02: the clause contradicts none of Law V's seven historical supporters (the Diagnostic Gap is evidence FOR facet a; the Sophistication Trap is refined by facet b), facet (a) is a usable pre-commitment test, and facet (b) is not knowable in advance so it demotes Law V from a confident prescription to a diagnostic hypothesis. This is Law V's FIRST amendment; standing moves to `load-bearing` EARNED (narrowed under genuine pressure), no longer asserted-on-use.
Essays that stress-test it
- Your Verification Budget Is Backwards You can't check everything your agents produce, so you're already triaging. Most of us spend the effort where checking is easy, which is rarely where a mistake is expensive.
- You Built a Number You Will Not Trust Some things you build get graded by the world. Some only ever hand you back your own guess wearing a decimal. Telling the two apart before you start is the cheapest hour you will spend.
- You Reach Before You Think What leaning on AI for every small decision quietly does to your own judgement.
- Your AI Looks Best Where You Can Check It Least When the first failure is terminal, you cannot iterate your way back.