Kathleen Creel (University of Pittsburgh), David Colaço (University of Pittsburgh)
For four years after the Parkes radio telescope identified “perytons,” terrestrial short chirped radio pulses, their cause was unknown. Researchers believed that perytons were terrestrial in origin because they occurred exclusively within the working week and were wide-field detectable, suggesting that they occurred close to the site of the telescope. Papers speculated that these pulses were caused by ball lightning, meteor trails or signals from aircraft (Katz 2014).
Although perytons had only been observed at one telescope, explaining their existence was important because they called into question the interstellar origin of fast radio bursts (FRBs), which have similar quadratic forms (Petroff et al. 2015, 3934). Finally, Petroff and colleagues showed that the pattern characteristic of perytons could be reproduced by opening the door of a nearby microwave oven in the facility’s break room while the magnetrons were still active. This reliably caused the pattern that had been detected as a “peryton.” Once researchers declared use of the microwave off-limits during telescope hours, the perytons disappeared.
Although “perytons” turned out to be artifacts, their patterns in data, due to their wide-field detectability, are stronger and more stable than those of their extragalactic analogues, FRBs. This case illustrates the shortcomings of current philosophical use of the signal-noise distinction. Philosophers use the terms “signal” and “noise” in their analyses of the detection of phenomena from the data they collect in experiments or observations (Bogen and Woodward 1988; Woodward 1989; McAllister 1997). However, their uses often blur an intuitive understanding of the difference between signal and noise with a technical definition derived from information theory.
We propose a new use of the signal-noise distinction that distinguishes the identification of phenomena and artifacts. Contra James Woodward, who takes everything in a dataset that does not correspond to the phenomenon to be idiosyncratic, we argue that data patterns that indicate the presence of artifacts are often robust and more easily detected than the patterns that indicate phenomena. To apply the signal-noise distinction in a way that accurately captures the process of detecting phenomena, it is paramount to understand the role of researcher interest in the process. Researcher interest dictates what phenomenon is investigated; interest does not itself determine what counts as a phenomenon. Researchers investigate patterns they think correspond to interesting phenomena. These patterns are chosen due to their informational character, from which researchers formulate characterizations of phenomena of interest.
