Outwitted by the Hidden: Unsure Emotions

Introduction

These days it is not always easy to say whether you are interacting with a male or a female, an adult or a teenager, indeed even a human or machine based solely on text-based interaction (Warwick & Shah, 2014). Being unsure about what or whom you are engaging can have unpleasant consequences, such as a misunderstanding at best, or at worst, risk of cybercrime (Shah & Warwick, 2012). In the Turing test experiments carried out by the authors the Interrogators were allowed to use an ‘unsure’ classification, because control pairs (2machines; 2humans) were embedded among the machine-human pairs (Shah, et al, 2014; 2012), if a judge really could not say whether they had been chatting to a human or a machine (Warwick, 2012). In this paper we present Interrogator-hidden interlocutor questions-answers that occurred during practical Turing tests at Bletchley Park in 2012 (see Warwick et al., 2013, Warwick and Shah 2013; Warwick and Shah, 2014; Warwick and Shah, forthcoming). We examine whether there was any emotional content in the utterances from those tests (see Tables 1, 2 and 3), that might have distinguished the artificial from the natural and revealed the human from the machine. The paper includes post-event feedback from interrogators in three viva voce (Shah, 2013; Shah, 2011) one-to-one Turing tests (see Figure 1). These three human Judge/Interrogators were not able to say whether they had been conversing with a human or a machine, after five minutes of interaction with a hidden entity. The hidden interlocutor received an ‘unsure’ classification from the Judge.

Note: Actual remote hidden entity: male, native English 25-44

Judge, J18, also male, native English, 45-64 with previous experience of chatting with a chatbot – IKEA’s Anna