Illustrative patterns of scientific exploratory behaviour are shown in Fig. 1. The points in each plot show the settings selected for two of the key variables – volume of polymerase and volume of dNTP – as the search progressed. There was a range of scientific exploratory strategies.
Some scientists chose to fix values of both primers and dNTP and perform experiments in the remaining space (eg. scientist 35, scientist 61 and scientist 63 ran all their experiments at a fixed volume of primer and dNTP).
Others chose to fix one or the other - either the volume of primers was fixed and the volume of dNTP varied, or the volume of dNTP was fixed and the volume of primers explored (eg. 1, 4, 5, 8, 13, 14, 23, 49, 57). Some scientists chose to fix one or other variable before exploring the other (eg. scientists 27, 40, 45). Yet others chose to vary both dNTP and primers at the same time making it impossible to separate out the impact of each of the variables (eg. scientists 11, 26, 33 & 42). The end-result is that the relative importance of the variables can be impossible to determine and very little of the design space is properly explored. While some scientists restrict themselves to relatively small changes from an initial value (eg. scientists 17, 21, 29, 41, 44, 58, 60, 62, 65 & 66). Others make more sweeping changes across a wider range of volumes (eg. scientists 6, 20, 22, 30, 38, 50 & 59).
To visualize the movements of scientists in multi-dimensional space, we used PCA to reduce the twelve-factor design space to just two dimensions and mapped scientist movements in the resulting canonical space – see Fig. 2. Mapping movements in canonical space highlights that there are large regions of the problem space for which we have no information.
Upon closer inspection, larger movements within canonical space, seem to be driven by the success of initial results. In Fig. 3 the search paths of four of the scientists are shown in more detail with elapsed time (in minutes) as indicated. A striking feature of these search paths is their similarity to foraging search strategies in other animals. For example, after one or two initial experiments yielding unpromising results, scientists might make a Lévy flight8 to some other region. They then wander around that region of the design space to fine-tune their results – see Scientists A, C and D. Alternatively, if the initial results are promising some scientists might make an excursion to a choice of more radical settings – see Scientist B. Fingers burnt, they then retreat to their original settings and conduct smaller excursions from that point.