GOOGLE'S Duplex: Pretending to be human

• Date: 01 January 2019
• Author: Daniel E. O'Leary
• DOI: http://doi.org/10.1002/isaf.1443
• Published date: 01 January 2019

RESEARCH ARTICLE

GOOGLE'S Duplex: Pretending to be human*

Daniel E. O'Leary

University of Southern California, Los Angeles,

CA, USA

Correspondence

Daniel E. O'Leary, University of Southern

California, Los Angeles, CA 90089‐0441, USA

Email: oleary@usc.edu

SUMMARY

Google's Duplex is a computer‐based system with natural language capabilities that

provides a human sounding conversation as it performs a set of tasks, such as making

restaurant reservations. This paper analyses Google's Duplex and some of the initial

reaction to the system and its capabilities. The paper does a text analysis and finds that

the system‐generated text creates standardized ratings that suggest the text is

analytical, authentic and possesses a generally positive tone. As would be expected

for the applications for which it is being used, the text is heavily focused on the present.

In addition, this analysis indicates that the text provides evidence of social processes,

cognitive processes, tentativeness and affiliation. Further, this paper examines some

of the characteristics of speech that Duplex uses to sound human. Those capabilities

appear to allow the system pass the Turing test for some well‐structured tasks.

However, this paper investigates some of the ethics of pretending to be human and

suggests that such impersonation is against evolving computer codes of ethics.

KEYWORDS

artificial intelligence, autonomoussystems, computer ethics, imitation technologies, LIWC, natural

language, text analysis, Turing test

1|INTRODUCTION

Google'sDuplex is a computer‐basedsystem with natural languagecapa-

bilities(Fisher, Garnsey,& Hughes 2016) that providesa human sounding

conversationas it performs a set of tasks,such as making restaurantres-

ervations. Recently Google's CEO demonstrated some of Google's

Duplex capabilities in a widely viewed video (https://www.youtube.

com/watch?v=D5VN56jQMWM). In addition, Leviathan and Matias

(2018) providea brief but additional discussionabout Duplex, examining

someof its capabilities. Currently,the system is designedto make restau-

rant reservations and make haircut appointments, but there are other

capabilitiesand applications planned. Duplex is not a research‐only sys-

tem; in particular,recently, Googleannounced that Duplex couldbe used

on Pixel phonesto allow its users to make restaurant reservations.

1,2

The announcement of Duplex brought varied reactions from differ-

ent groups. Developers saw Duplex as an opportunity to put natural

language into practice for real‐world applications, as Google provided

guidance for developers to generate APIs (https://developers.google.

com/assistant/sdk/). Researchers saw Duplex as an opportunity to

further study natural language (https://ai.google/). However, some

commentators saw Duplex as raising a number of ethical issues (e.g.

Meyers, 2018), while still others wondered what ‘laws’robots should

follow (e.g. Heaven, 2018). At the same time, some commentators

suggested that, at some level, Duplex had passed the Turing test.

3

Still

others saw Duplex as just additional applications of the emerging and

increasingly crowded space of personal assistants in the smart home

sector, such as Alexa and Assistant (Wetzel, 2018). Finally, some

seemed just upset at the prospect of computers impersonating people

(Bay, 2018) “nobody likes a fake person”.

The purpose of this paper is to investigate and extend some of

these discussions. In addition, this paper provides a text analysis of

two of the Duplex conversations. That text analysis finds that Duplex's

text is generally more positive than negative, and that Duplex captures

profiles regarded as having social and cognitive processes. In addition,

the system is found to be generally analytic, expert and confident, all

*

An earlier version of this paper was presented in Boston, in July 2018. I would like to thank

the participants for their comments.

DOI: 10.1002/isaf.1443

46 © 2019 John Wiley & Sons, Ltd. Intell Sys Acc Fin Mgmt. 2019;26:46–53.wileyonlinelibrary.com/journal/isaf