In the Turing test a computer model is deemed to “think intelligently” if it can generate answers that are indistinguishable from those of a human. We developed an analogous Turing-like handshake test to determine if a machine can produce similarly indistinguishable movements. The test is administered through a telerobotic system in which an interrogator holds a robotic stylus and interacts with another party—artificial or human with varying levels of noise (figure 1). The interrogator is asked which party seems to be more human. Here, we compare the human-likeness levels of three different models for handshake: 1) Tit-for-Tat model, 2) λ model, and 3) Machine Learning model (figure 2). The Tit-for-Tat and the Machine Learning models generated handshakes that were perceived as the most human-like among the three models that were tested (figure 3). Combining the best aspects of each of the three models into a single robotic handshake algorithm might allow us to advance our understanding of the way the nervous system controls sensorimotor interactions and further improve the human-likeness of robotic handshakes.
Figure 1. The human and the interrogator held the stylus of a haptic device. Position information was transmitted between the two devices and forces were applied on each of the devices according to the particular handshake.
Figure 2. In each trial, the interrogator performed two handshakes and he had to determine which handshake was perceived as more similar to a human handshake. In one of these handshakes, the interaction was with one of three models for handshake: tit-for-tat, lambda or iML-shake; the other handshake was generated from a linear combination of human handshake and noise with different weights. This setup enables comparison between a handshake model and a human handshake that is disturbed with different levels of noise. The basic idea is that the bigger the weight of the noise required disturbing the human likeness of the human handshake so that it will be indistinguishable from human handshake, the smaller the human likeness level of the handshake model.
Figure 3. Human likeness for the models. Panels (a) and (b) present examples of psychometric curves that were fitted to the answers of interrogator #2 and interrogator #5, respectively. Dots are data points, and the horizontal bars are 95 percent confidence intervals for the estimation of PSE. A model with higher MHLG (Model Human Likeness Grade) yields a curve that is shifted further to the left. (c) The MHLGs that were calculated according to each of the 10 interrogators who performed the test. Symbols are estimations of MHLG, and vertical bars are 95 percent confidence intervals. (d) Means with bootstrap 95 percent confidence intervals (error bars) that were estimated for each of the models using the MHLG of all the interrogators (*-Tukey's honestly significant difference criterion, p<0.05).
