From 1966 Rosenblatt turned from perceptrons to testing molecular memory transfer by injecting brain extracts between rats

Having built the perceptron as a model of learning stored in distributed connection weights (claim-rosenblatt-perceptron-inventor-died-on-43rd-birthday), Frank Rosenblatt spent his final years investigating the rival molecular theory of memory — the idea that a learned behavior could be encoded as a transferable biochemical substance rather than as a pattern of synaptic connections. The Cornell Faculty Memorial records: "In 1966 he added an interest in the transfer of learned behavior from trained to naive rats by the injection of brain extracts, and he published extensively in this area."

In practice this meant training rats on Y-maze and operant tasks, preparing extracts from the brains of the trained animals, and injecting them into naive animals to test whether the learned behavior transferred with the substance. The program sat inside a 1960s craze around molecular memory, inspired by James McConnell's planarian experiments (learned responses apparently transferred by feeding trained flatworms to untrained ones).

The juxtaposition is the find: the man who gave AI its distributed-weights account of memory — the connectionist lineage that runs on through claim-hopfield-1982-energy-function-from-spin-glass-physics and claim-dense-associative-memory-exponential-capacity, where memory is a state of a network rather than a molecule — spent his last years testing the opposite, molecular account. What that testing found is in claim-rosenblatt-helped-refute-memory-transfer-effect. See moc-backpropagation-origins.