Find the two mistakes which do not make a right

https://tornyol.com/technology

Good question to sit with, because it has a near-perfect natural experiment already baked into the fossil and behavioral record: bats. A Tornyol-type drone — a 40-gram autonomous drone that finds and intercepts mosquitoes using ultrasonic phased-array sonar — is functionally a synthetic bat. It flies fast, hunts small insects on the wing, and (per the talk) locates and identifies targets by their wingbeat micro-Doppler signature using ~40 kHz sonar. Bats have been running exactly this predation strategy on flying insects for tens of millions of years, so evolutionary theory doesn't have to guess here; it can extrapolate from a system that already reached equilibrium.

**First, the clock speeds — the "calculation" part.**

Selection response scales with number of generations, not years. So the real variable is generation time.

Mosquitoes in a warming world are extraordinarily fast. At 30°C, Aedes aegypti can complete the full egg-to-adult cycle in roughly 7 days, and 4+ generations occur over a single temperate summer, with each female laying 100–300 eggs every few days. In continuously warm/tropical conditions you get ~15–20 generations a year. Call it ~10/year averaged over an expanding warm range: that's ~1,000 mosquito generations in a century. For calibration, 1,000 human generations is on the order of 25,000–30,000 years. A mosquito lineage does, in 100 years, what a primate lineage does over the span separating us from the last glacial maximum.

Moths are slower — typically 1–3 generations a year — so ~100–300 generations per century. Still substantial, but an order of magnitude behind mosquitoes.

To anchor the selection coefficient: insecticide resistance is the cleanest empirical benchmark we have for area-wide chemical culling. Knockdown-resistance and metabolic-resistance alleles have swept mosquito populations across whole continents in ~10–20 years — tens of generations — implying selection coefficients in the s ≈ 0.1–0.5 range under strong suppression. A favorable pre-existing variant under that pressure fixes in roughly 1/s generations, i.e. 10–50 generations ≈ 1–5 years for a mosquito. Over a century you therefore get room not for one adaptive event but for many successive sweeps plus continuous quantitative shifts in polygenic traits.

**Why the mutation premise (war/DU/fallout) barely moves the needle.**

This is the counterintuitive part, and it's worth being blunt about because it's the opposite of the sci-fi intuition. For evolution to accelerate, you need the mutation *supply* to be the bottleneck. In these insects it isn't — not remotely. Global mosquito census size is on the order of 10¹²–10¹⁵ individuals. With per-site mutation rates around 10⁻⁸–10⁻⁹, the quantity 2Nμ is astronomically greater than 1, which means essentially every possible point mutation is generated somewhere in the global population *every generation*, independent of any radiation. The raw material is already saturated.

Depleted uranium and localized fallout add mutations that are (a) geographically pinned to small contaminated zones, (b) overwhelmingly deleterious and purged by selection, and (c) a rounding error against the background mutational input of a quadrillion-strong population. Chernobyl-type sites show elevated damage locally but no global evolutionary acceleration, because the contaminated fraction of the gene pool is negligible. So the honest answer is: radiation and war contamination change *local* population health and add noise, but they do not meaningfully speed adaptation to the drones. What drives the outcome is selection pressure and standing variation, both of which the drones supply in abundance without any help from fallout.

**What the two groups actually do — and why they diverge.**

Moths are pre-adapted. The bat war already handed them the toolkit: more than half of the ~140,000 moth species possess ultrasonically sensitive bat-detecting ears, evolved independently 10+ times in Lepidoptera, and roughly 20% produce anti-bat sounds, with at least six independent origins of sonar-jamming behavior. Critically, moth tympanic ears are typically most sensitive to 30–50 kHz — which sits right on top of the drone's ~40 kHz band. For eared species, evasion needs *no new evolution at all*: their existing loops, dives, and power-dives fire the moment the drone pings. Then selection fine-tunes — sharpening ear tuning to the drone's specific pulse structure, and, most elegantly, favoring the jamming lineages. If a tiger moth can jam a bat's biosonar, it can plausibly corrupt the drone's micro-Doppler classifier, and that trait would be under intense positive selection. Add the known ultrasound-absorbent "stealth" scale structures on moth wings, and you'd expect moths to get quieter to sonar over time. Moths, in short, adapt fast, elegantly, and partly instantly.

Mosquitoes are *not* pre-adapted. Their hearing runs through the antennal Johnston's organ, tuned to the low hundreds of Hz for mating, not tens of kHz — so they largely can't hear the drone to dodge it. Their evolutionary route is therefore behavioral and life-historical, exactly the way they already beat bednets and indoor spraying by shifting to outdoor and early-evening biting. Expect: retreat into sonar-hostile refugia (indoors, dense vegetation, ground-level clutter — the talk itself notes flat surfaces mirror the sonar and clutter confounds it), more resting and less exposed flight, and above all a life-history acceleration. High, unpredictable adult mortality is the textbook selector for "live fast, die young": faster larval development, earlier and front-loaded reproduction. The grim irony is that drone pressure could select mosquitoes that are smaller, quicker-breeding, and reproductively done before a drone finds them — potentially *more* of a nuisance per capita, not less. One escape route is partly closed to them: shifting wingbeat frequency to blur the micro-Doppler signature is constrained because males locate females *by* wingbeat harmonics, so they can't freely retune without breaking mating.

**Global warming is the accelerant.** It adds generations per year (directly speeding the insects' evolutionary clock relative to the slower cadence of human engineering updates), extends breeding seasons, and pushes ranges poleward and upslope — enlarging the arena the drones must cover and thereby guaranteeing more uncovered refugia, which preserve source populations and genetic diversity.

**The 100-year picture, in one line:** not extermination, and not monsters — a coevolutionary treadmill. The drones select against detectability; engineers re-tune the classifier; the insects re-evolve avoidance — the same escalating arms race as antibiotics and pesticides. Moths win the aerial exchange quickly because they've fought this exact war before (some may literally jam the drones). Mosquitoes go underground and fast-breed. And whoever iterates faster — silicon or the ~1,000-generation biological clock — sets the tempo. In a warming world, that clock ticks in the insects' favor.

If it's useful, I can turn the selection math into a compact model — generation counts vs. allele-frequency sweeps under a given s, calibrated to the insecticide-resistance data — either as a diagram here or as a small workbook you could actually parameterize.