The Cargo Cult Runway

Bamboo Control Towers, Rat Mazes, and the Discipline of Not Fooling Yourself

Across several islands in Melanesia after the Second World War, indigenous villagers cleared dirt airstrips in the jungle and lit signal fires along the edges at dusk. Inside a small bamboo hut beside the strip, a controller sat wearing two wooden blocks over his ears connected to a bamboo pole antenna. During the war, the islanders had watched American military personnel perform those exact routines. Right afterward, four-engine transport planes had descended from the clouds packed with canned food, clothing, and tools. Now the war was over, the runway torches were burning, the controller was sitting in his bamboo tower with wooden headphones, and no airplanes landed.

Standing at the commencement podium at the California Institute of Technology in Pasadena in June 1974, Feynman used that South Pacific scene to name a pervasive flaw in modern research: Cargo Cult Science. In fields ranging from educational curriculum trials to experimental psychology, investigators followed all the outward rituals of scientific inquiry. They wore white laboratory coats, assembled control groups, computed statistical $p$-values, and plotted impressive charts. Yet they left out an essential internal check. They failed to bend over backward to report every possible flaw that could invalidate their own hypothesis.

Feynman warned the graduating class that the first principle of scientific integrity is to avoid fooling oneself, especially since the researcher is always the easiest person to fool.

To show what genuine experimental control demands in practice, Feynman recounted how he had once asked Caltech biologists for literature on rat navigation in T-mazes. In the library stacks, he uncovered a landmark 1937 study by psychologist Paul Thomas Young. Young had tried to train laboratory rats to choose the third door down a corridor regardless of where the food had been placed in prior runs. Instead, he found that the rats kept solving the puzzle by tracking hidden environmental clues.

One after another, Young systematically eliminated every sensory channel he could find. He painted the maze doors uniform white, textured them identically, rotated the apparatus to scramble room sights, masked the odor of the food with chemical vapors, and shifted the overhead lights. Still the rats headed straight for the food box. Finally, Young discovered that the rats were listening to the hollow acoustic resonance of their own footsteps bouncing off the wooden floorboards above the food compartment. Only when he elevated the corridors on concrete blocks set inside beds of damp sand did the acoustic cue vanish. Only then did the experiment measure real learning.

From a physicist's standpoint, Young's 1937 paper was a first-rate experimental control because it exposed the exact instrumental artifacts that ruined every standard rat experiment. Yet when Feynman checked the citation index for subsequent psychology papers on rat learning, he found a troubling silence. Later researchers almost never cited Young and continued building mazes directly on hollow wooden floors.

Even experimental physicists fell into the same trap when they deferred to famous benchmarks. When Robert Millikan measured the elementary charge of the electron in 1913 using falling oil drops, he obtained $4.774 \times 10^{-10}$ electrostatic units. His result was slightly low because he had used an inaccurate viscosity for air. Over the next twenty-five years, succeeding physicists published numbers that crept upward in hesitant steps: 4.780, 4.793, and finally 4.803. Anyone who measured 4.80 right after Millikan assumed his own instrument was faulty and hunted for errors until his number matched the canon. Although Young's 1937 paper proved that laboratory rats navigated mazes by the sound of their own footsteps on hollow floorboards unless the corridors were set in damp sand, subsequent psychology papers cited him almost zero times and kept running uninsulated mazes.