Artificial Evolution in the Physical World

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Concrete Problems in AI Safety (2016, background): “More sophisticated exploration strategies that adopt a coherent exploration policy over extended temporal scales [112] could actually have even greater potential for harm, since a coherently chosen bad policy may be more insidious than mere random actions.”
Quantilizers: A Safer Alternative to Maximizers for Limited Optimization (2016, methodology): “Curiously, five of those cells were not connected to either the input or the output—the evolved circuit was making use of them via the physical features of that specific chip and machine (Thompson suggest that the circuit was taking advantage of electromagnetic coupling or power-supply loading).”
The evolved radio and its implications for modelling the evolution of novel sensors (2002, background): “In conventional electronic design the interaction of the cells and the overall behaviour of FPGAs is viewed at a functional, logic gate level, rather than in terms of transistor dynamics [16].”
Structure-Adaptable Digital Neural Networks (1999, methodology): “Thompson [207] developed an FPGA-based extension turret to run evolvable hardware experiments (e.”