Figure 4

Capability vs spread for the intrinsic difficulties scenarios. The space is characterised by three normalised generality isometrics shown as dotted curves: the maximum spread in red (complete abstrusity), the minimum spread in green (complete generality) and a middle case where results are independent of difficulty (constant ACC). Top left: Elithorn’s mazes. Results for all 496 participants. Top right: letter series for all participants (48 humans and 12 machines). Middle left: object recognition scenario using the intrinsic difficulty derived from the psychophysical parameters). Middle right 10 rats in the Odour Span Task, with difficulty being the number of scents to remember. Bottom: iris classification problem using 419 classifiers from study 7306 in OpenML (after classifiers with accuracy below 0.35 removed). Left, capability vs spread using KDN difficulty (\(k=10\)). Right: using \(TD_U\) difficulty instead.