You need to show that you are not a robot

Abstract

Given that today 60% of Internet traffic is generated by bots, ‘CAPTCHA’ (Completely Automated Public Turing Test to tell Computers and Humans Apart) tests that are supposedly impossible to be done by robots have been introduced. What are the cognitive and emotional effects of these tests on Internet users? Does this request to demonstrate they are not a robot affect users’ identity as human beings? To answer these questions, we selected two groups (117 and 116 respondents, respectively). An online questionnaire that differed only in the task was proposed: we asked the first group to complete some CAPTCHA tests, and the second group to complete some logic tests. In addition to other questions in both versions, we introduced the TLX scale (NASA). Preliminary results show that CAPTCHA execution is associated with feelings of alienation and that the user’s self-perception of humanity is influenced by the execution of the two different types of test.

Keywords

Bots CAPTCHA human identity TLX scale Turing test

Introduction

Human-computer communication lacks a series of cues to the identity of the interlocutor and to the characteristics of the context in which the interlocutor is located (Suchman, 1987; Walther, 1996). It is because of this lack of visibility and awareness that today the majority (60%) of the traffic on the Internet can be generated by bots. These are software designed to perform various functions, without creating serious social outcry (Zeifman, 2017). They act as spiders of search engines that monitor the Internet to discover content and websites, or they act as programmes that scour the net looking for information to be used for spam campaigns and so on. But, as Bakardjieva (2015) and Gehl and Bakardjieva (2017) have argued, bots also pollute information online both at political level (astroturfing) and the commercial level, distorting the information on how many visits of potential clients a business site receives because bots do not buy anything.

Quite understandably, a strategy has been developed to combat these kinds of programmes which provide some types of tests that are supposedly impossible for robots to perform. These tests are named ‘CAPTCHA’ (Completely Automated Public Turing Test to tell Computers and Humans Apart). The classical tests consist of those almost illegible words, which are proposed by many websites, and which one has to decipher and rewrite correctly in order to demonstrate that one is not a robot (Abrich et al., 2011; Gafni and Nagar, 2016). According to Brodić et al., (2016), the reasons for using CAPTCHA are the following: they prevent spams on forums, they avoid having a large number of orders opened by users on sites that offer free services like Gmail, they protect user accounts from attacks through which bots discover users’ passwords and they concur to defend the validity of online surveys by detecting whether humans or bots answer the questionnaires. These tests, which have evolved and have been diversified over time, should be seen as magic helpers, guardians of the truth in online environments, because they enable website owners to defend themselves from automation and robotization, by distinguishing human traffic from robot traffic. However, as we will see later, they may present some problems for the well-known reason that there is no solution to a problem that does not present some secondary negative effects.

The CAPTCHA was coined in the 2000s by Louis Von Ahn, Manuel Blum and Nicholas J. Hopper, professors at the Carnegie Mellon University, and became immediately popular (Singh and Pal, 2014). Over time, other types of CAPTCHA tests have been invented, such as those based on image recognition, vocal CAPTCHA and audio CAPTCHA, whose functional interpretation is supposed to be analysed only by a human user. The aim of these tests is to stop the attacks of bots by obtaining the correct answer from the user and thus classifying the user as a human. These tests, of course, can work as long as no machine will ever pass the Turing test (Turing, 1950). Recently, according to Dehaene et al. (2017), a new type of artificial intelligence was able to solve certain types of CAPTCHA with an accuracy of up to 66.6%. To put this in perspective, a CAPTCHA is considered broken if a bot can pass it 1% of the time. This means that bots are learning very quickly how to perform this kind of test correctly.

Neither are humans perfect as they can perform the same type of CAPTCHA with different degrees of accuracy due to multiple interpretations of some examples (Fidas et al., 2011). Thus, up to now, humans continue to be better than bots. So far, the scientific literature on CAPTCHA has investigated mainly their functional effectiveness, the administrator experience and even users’ experience (Abrich et al., 2011) in the attempt to establish what is the ‘ideal’ CAPTCHA (Brodić et al., 2016). The ‘ideal’ CAPTCHA is designed to be solved in a reasonable time (less than 30 seconds), and the solution time should not depend on the age, education and gender differentiation, nor on the specific device used to access the Internet (desktop computer, laptop, smartphone and tablet) nor on the specific culture to which the user belongs (since users pushed web designers to use national languages in CAPTCHA tests) (Lin et al., 2018).

Specifically, users’ experience is analysed mainly in terms of response time for each test and rate of success (Gafni and Nagar, 2016) or in terms of limited website usability by respondents with all kinds of different disabilities (Pieper, 2012). In these type of studies, generally users are conceptualized in an instrumental way to understand better the functioning of the CAPTCHA tests. It remains necessary to understand the use and performance of these tests more broadly and to investigate the consequences of their use on Internet users. This is the challenge that we have taken up in this study.

The present article is organized in the following way: in the next section, we will provide some reflections on the current processes of diffusion of automation and robotization in society and on human identity. Next, we discuss the methods and the measures applied in this research, reporting also the statistics we applied. The following section illustrates the main findings, and finally, a discussion of the results and closing remarks on the strong and weak points of the study, as well as on the future direction which this research needs to take, will conclude this discourse.

CAPTCHA, bots, robots and human identity

As we mentioned in the introduction, we need to situate the issue of CAPTCHA tests inside a framework larger than that in which the literature about CAPTCHA usually addresses this issue. So far, we have seen how computer-mediated communication lends itself to accepting increasingly automated processes. The acceptance of this increase in automated processes, however, has involved not only computer-mediated communication but the entire mediated communication sphere from electronic mass media to print newspapers (O’Sullivan et al., 2017) to mobile communication (Taipale et al., in press) in general. Even more generally, it has involved all spheres of society, especially the spheres of individual and social reproduction (Barile and Sugiyama, 2015; Esposito et al., 2014; Höflich, 2013; Pfadenhauer et al., 2015; Sugiyama, 2013; Sugiyama and Vincent, 2013; Taipale et al., 2015). Automation is a precious resource in all productive sectors, but it brings with it specific problems, especially when it addresses the immaterial sphere of labour-force reproduction, which is made up of several dimensions: education, affects, gender, information, interpersonal communication, emotions, learning, sociability and so on (Fortunati, 2018). This literature describes and maps various processes of automation that are developing in society, outside of the industrial sectors where, traditionally, automation and robotization were first developed. It has also stimulated a series of empirical research on robotization in the reproduction sphere and has advanced the debate about the social meaning of these processes.

The main point put forward by the field of social sciences on the topic of automation and robotization is that these processes extract value from sectors, such as the individual and social reproduction of the labour force, that had so far escaped a direct extraction of value. In this globalized and informatized society, individuals as prod-users are enveloped in a network of complex and unobvious relationships that expropriates them of their humanity and treats them like any commodity. This is especially true in the immaterial sphere, constituted by elements such as emotions, affects, communication, information, entertainment, learning, teaching and so on. The effects on the labour force are readable in terms of an increase in the commodification of individuals in the specific dimensions that are included in the sphere of immaterial labour and in the spreading of new forms of alienation. Behind this commodification, there is an exploitation of individuals that takes place in forms never experienced before (Ekbia and Nardi, 2017). Our interest here is to understand whether these huge processes reverberate even on a small but significant problem, such as that represented by CAPTCHA.

Applying this approach to CAPTCHA tests, a series of questions arise. How can we read from a social science approach the consequences that the CAPTCHA execution produces on Internet users? How do human beings react physically, cognitively and emotionally to this request coming from the online environment? Does the request, made to Internet users to demonstrate that they are not a robot before presenting themselves in their own individuality, affect their identity as human beings?

The point is that the advantage that website owners have in using the CAPTCHA tests seems to be gained at the expense of Internet users. In fact, the request to perform these tests requires energy and time from users for reasons that are not in their interests but in those of website owners.

Moreover, this request creates the paradoxical situation of changing radically the presentation of the self on the part of humans in computer-mediated communication. Usually, human beings perform the presentation of the self in order to express the uniqueness of one specific individual compared with other individuals. During their presentation, humans convey the rich data of their personal identity that are different from the identities of the other human beings. Now, after the advent and the increased diffusion of bots, in computer-mediated communication, people are asked to express their identities in terms of not being a robot. The shift to an intergroup dimension, which forces users to self-categorize themselves as humans in opposition to robots, eliminates the issue of uniqueness in personal identity. Personal identity is no longer relevant and leaves room only for an identity defined as generically human. This paradoxical request is advanced to put users in the situation of having to demonstrate they are humans. It is a sort of reversal of the situation in which automata and androids were accused in the early days of cheating on their identity because they imitated human beings so well to the point of being erroneously mistaken as such (Fortunati, 2013).

This situation is intriguing and opens up the necessity of investigating the question of identity in humanity. Human identity has been explored over time in terms of similarity or dissimilarity. Anthropologists, for example, have elaborated on the notion of universals, which are ‘those features of culture, society, language, behaviour, and mind, so far as the record has been examined, are found among all peoples known to ethnography and history’ (Brown, 2004: 47) and which thus represent what human beings have in common beyond the different societies in which they live. In contrast, the ‘naturae scala’, the classical model of the hierarchical structure of the natural world, which has been widely accepted by most European scholars from the time of Lucretius to the final development of the Renaissance (Lovejoy and James, 1936), has explored the notion of humanity through dissimilarity. This model consists of putting God at the top, under God the angels, then human beings, animals, plants and the inanimate world. Although it has been superseded by modern taxonomies based on an evolutionary approach, this model is still rooted in our culture.

However, over time the perception and the investigation of the identification of humankind has been silenced, as the scientific debate ended, and the focus changed to elaborating on constructs such as personal and social identities (Capozza and Brown, 2000), community identity (Puddifoot, 1995) and national identity (Smith, 1991).

Interesting exceptions to this perspective have been, for example, Adler ([1929] 1964) and Maslow (1954). Adler ([1929] 1964) advanced the idea that the most mature meaning of ‘social interest’ is the sense of ‘oneness with humanity’, and Maslow (1954) elaborated the theory of ‘self-actualized individuals’ in which individuals with human kinship ‘have a deep feeling of identification, sympathy and affection for human beings in general’ (p. 138). More recently, the socio-psychologists Turner et al. (1987) elaborated self-categorization theory with three levels, where the highest level corresponds to one who categorizes oneself as a part of all humanity. In the same vein, Gaertner et al. (1993) and Dovidio et al. (1998) elaborated on the common in-group identity model, which states that individuals ‘who regard all humanity as one ingroup should be low in bias against groups whom others would regard as outgroups (other races, nationalities, religion)’.

Drawing on these theorists and on other kinds of works by Edward Steichen (1955), who organized the exhibition The Family of Man, and the book A Life in Photography (1963), by Monroe (1996) and Oliner and Oliner (1988) on individuals who rescued Jews during the Holocaust, McFarland et al. (2012) took up the notion of human identity again. They acknowledged that very little research has been done on this construct and proposed a measure of identification with all humanity, which has been further developed by Reese et al. (2015) with the purpose of capturing how much this construct accounts for the human identity. The issue of the identity with humankind has been illuminated paradoxically also by the debate on dehumanization (Bar-Tal, 1989; Volpato, 2011). Dehumanizing means denying humanity to the other, individual or group, which is openly accused of being without, or lacking in, the essential qualities that define a human being as such. Dehumanization can be considered the most radical of the intergroup delegitimization strategies, which permanently excludes the delegitimized group from the circle of groups considered as civilians (Bar-Tal, 1989). Dehumanization is expressed through metaphors such as animalization, biologization or demonization. One of the most recent metaphors, from an historical point of view, is the mechanistic metaphor that employs comparisons drawn from the world of machines and automata and refers to the negation of the aspects of emotionality, warmth and vitality, which characterize human beings. The individuals and groups who are the objects of this form of dehumanization are considered machines, automata or robots, things that induce indifference and lack of empathy. This metaphor (also known as objectification) finds its most important application in the industrial world, where workers can be seen as instruments, lacking intellectual autonomy and always replaceable (Andrighetto et al., 2017). Being subjects of objectification has important psychological consequences, including the tendency towards self-objectification, which is expressed, among other things, in a diminished experience of some typically human mental states such as reasoning, planning, desiring (Baldissarri et al., 2014) and then in a diminished attribution of humanity to one’s own self.

In effect, other social disciplines also contribute to this question in a series of discourses and research. A constellation of processes and phenomena that occurred in the last half century have been identified by social sciences as crucial factors in supposedly strengthening the common recognition by individuals of being part of humankind as a whole. At least five main elements should be listed: globalization (Hall, 2018); diffusion of digital technologies (smartphones, Internet and computer) all over the world (Balbi and Magaudda, 2017); the rise of networked individualism and the parallel weakening of family and community ties as well as of national identity (Sciolla, 1997; Wellman, 2001); increase in geographical mobility (Fortunati and Taipale, 2017); and, finally, the sustainability debate (Harte, 1995; Reysen and Katzarska-Miller, 2013). One of the relevant effects of these intertwined processes and phenomena is the weakening of identity at personal, community and national levels and its becoming dynamic, unstable and fragmented and thus ‘liquid’ (Bauman, 2003).

Does this configuration of identity open up new room for individuals’ identification in humankind? To answer this question and those posed above, we formulated two hypotheses:

H1: CAPTCHA execution is associated with forms of users’ disturb and/or alienation, since users are asked to lose time and produce a cognitive effort for reasons that are extraneous to their interests.

H2: Execution of the CAPTCHA tests asking immediately for proof of not being a robot would diminish an individual’s self-perception as a human being.

Procedure, measures and participants

The data analysed in this chapter come from an online research in which two different groups were approached through personal contacts in social networks and ‘snowball’ technique (see, for example, Baltar and Brunet, 2012). They were invited to fill out one of the two questionnaires, uploaded on a private server. The respondents were given online questionnaires including open and closed questions. We prepared two versions of the questionnaire that differed only in the task, which was randomly assigned to the respondents: in the first version, we asked our respondents to complete a series of CAPTCHA tests, and in the second version, we asked respondents to complete a series of logic and basic culture tests. We illustrate in Table 1 the socio-demographic variables of the two groups activated in this study.

Table 1.

Socio-demographic features of the two groups.

Variable	CAPTCHA group	Test group	Total
Gender	Absolute numbers and %	Absolute numbers and %	Absolute numbers and %
Males	54 (46.2)	50 (43.1)	104 (44.6)
Females	63 (53.8)	66 (56.9)	129 (55.4)
Total	117 (100.0)	116 (100.0)	233 (100.0)
Age
Less than 18	25 (21.4)	27 (23.3)	52 (22.3)
18–24	40 (34.2)	38 (32.8)	78 (33.5)
25–44	37 (31.6)	32 (27.6)	69 (29.6)
45–65 and older	15 (12.8)	19 (16.4)	34 (14.2)
Total	117 (100.0)	116 (100.0)	233 (100.0)
Education
Primary school diploma	38 (32.5)	42 (35.9)	77 (33.0)
High school diploma	42 (35.9)	53 (45.7)	95 (40.8)
College degree and above	37 (31.6)	24 (20.7)	61 (26.2)
Total	117 (100.0)	116 (100.0)	233 (100.0)
Activity
Student	57 (48.7)	59 (50.9)	116 (49.8)
Workers	56 (47.9)	50 (43.1)	106 (45.5)
Unemployed, Pensioner, House-person	4 (3.4)	7 (6.0)	11 (4.7)
Total	117 (100.0)	116 (100.0)	233 (100.0)

CAPTCHA: Completely Automated Public Turing Test to tell Computers and Humans Apart.

The percentages are calculated by column.

The application of the χ² test shows that there is no significant difference regarding the socio-demographic characteristics of these two groups. It is worth noting that all respondents, except for six, use the Internet every day, without differences between the two groups. Furthermore, respondents think that the Internet is an important tool nowadays, with an average score of 4.56 (standard deviation [SD] = .66) on a five-point Likert-type scale, again without any differences between the two groups (t₂₃₁ = 1,334, p = n.s.).¹

To the first group, which we will call CAPTCHA group, eight different types of CAPTCHA were proposed: the first was the classical test with distorted letters or numbers to identify between lines and additional signs to disturb any bots. The second was a test with rotated tri-dimensional letters and with resized fonts. The third was an audio CAPTCHA; the fourth was a CAPTCHA based on the interpretation of a series of images; and the fifth was a CAPTCHA made up of a sentence lacking a word, and the respondents were required to fill the empty space with a word that could give meaning to the sentence. The sixth test was a drawing CAPTCHA, which required respondents to draw, with the aid of the mouse, a form of some kind (usually geometric or at least simple to reproduce for any subject). The seventh was a reCAPTCHA test in which the respondents had to check the box corresponding to the sentence: ‘I am not a robot’. The final test was an insertion CAPTCHA with some specific features: the text to be inserted was a sequence of words with complete sense (in this case in English). There was the possibility of refreshing, or the possibility of changing the words to be written, in case of lack of clarity or excessive insecurity in completing the test.

In contrast, to the second group, which we call ‘Test group’, ten logical tests or questions on basic culture were proposed. The first test consisted of the request to choose from a list of five animals which of them was less similar to the other four. The second test was a simple logic problem: a series of numbers was proposed according to a precise pattern, and the respondents simply had to guess what the next number to be inserted in the series would be. The third test was a slightly more complex logic problem than the previous two: Anna, who is 16 years old, has four times her brother’s years. How old will Anna be when she will have twice her brother’s years? The fourth was the question: Where does the polar bear live? Choose only one of the following answers: North Pole, South Pole. The fifth test was as follows: complete the sequence: 3, 5, 8, 13, … Choose only one of the following numbers: 21, 24, 18, 20. The sixth test was a question: The novel The Promised Spouses was written by? Choose only one of the following answers: Dante Alighieri, Giovanni Pascoli, Giacomo Leopardi, Alessandro Manzoni. The seventh test was another question: Is Google Chrome a …? Choose only one of the following answers: browser, search engine, social network, text editor. The eighth was the question: If I am in Cairo I’m in … Choose only one of the following answers: Libya, Egypt, North Korea, Tunisia. The ninth was the logic test: 2 – 5 + 6 + 3 − 2 × 2 = … Choose only one of the following answers: 3, 2, 1, 0. Finally, the 10th was the logic test: BESTIA stands at BST as ARCAICA stands at … Choose only one of the following answers: RCIC, ARCIC, ARCC.²

In both versions, we introduced the TLX scale (NASA) (Hart and Staveland, 1988), which was used to measure the cognitive efforts of our respondents in completing the various types of CAPTCHA (in the first group) and in completing logic and basic culture tests (in the second group) (Milyavskaya et al., 2018).

This scale had specific purposes: usually used by NASA, it aims to calculate the cognitive effort necessary to carry out any task in general. This scale is constituted by the following six dimensions, as reported in Hart and Staveland (1988):

1) Mental Demand (How much mental and perceptual activity was required (e.g. thinking, deciding, calculating, remembering, looking, searching, etc.)? Was the task easy or demanding, simple or complex, exacting or forgiving?); 2) Physical Demand (How much physical activity was required (e.g. pushing, pulling, turning, controlling, activating, etc.)? Was the task easy or demanding, slow or brisk, slack or strenuous, restful or laborious?); 3) Temporal Demand (How much time pressure did you feel due to the rate or pace at which the tasks or task elements occurred? Was the pace slow and leisurely or rapid and frantic?); 4) Performance (How successful do you think you were in accomplishing the goals of the task set by the experimenter [or yourself]? How satisfied were you with your performance in accomplishing these goals?); 5) Effort (How hard did you have to work (mentally and physically) to accomplish your level of performance?); and 6) Frustration Level (How insecure, discouraged, irritated, stressed and annoyed versus secure, gratified, content, relaxed and complacent did you feel during the task?). (p. 170)

For each question, the score ranged from 1 to 100; the scores of each dimension corresponded to the sum of the scores attributed to each individual question included in the dimension.

In addition to this scale, we asked the CAPTCHA and Test groups a series of questions. The first aimed to explore whether it was easy to resolve the tests proposed (with a yes/no modality of answers). The second wanted to investigate whether respondents were amused by doing the tests (with a five-point Likert-type scale in which 1 was ‘a little amused’ and 5 was ‘very much amused’). In addition, we then proposed a question that explored the strongest emotion felt while doing these tests (with only one answer possible). The list of emotions suggested, from which to choose an answer, were the following: Irritation, Anxiety, Indifference, Agitation, Boredom, Curiosity, Safety, Fun, Interest and Pleasure. Finally, to investigate the issue of human identity, we introduced the question: ‘To what extent have you felt diminished in terms of your human identity?’ The answer was based on a five-point Likert-type scale, in which 1 meant ‘not at all’ and 5 meant ‘very much’.

We asked the CAPTCHA group, two further questions with the aim of investigating more deeply, participants’ experience in undertaking the various types of CAPTCHA tests. The first was as follows, ‘Which of the CAPTCHA tests did you take more willingly?’ (more than one answer was possible). The second question was, ‘Which of the CAPTCHA tests did you find most difficult to take?’ Choose only one test. The answer was on a five-point Likert-type scale, in which 1 meant ‘not at all’ and 5 meant ‘very much’.

The socio-demographic variables investigated are gender, age (in four age groups: under 18 years old, young people from 18 to 24, young adults from 25 to 44, adults from 45 to 65 and older), education, consisting of three groups: secondary school, high school and college degree and above. Given the objective of our study – the CAPTCHA tests – we decided that it was important to also investigate the practices of use of the Internet and the importance attributed by respondents to this tool. The frequency of Internet use thus was investigated by means of a four-point Likert-type scale (daily, weekly, monthly and occasionally), and the evaluation of the importance of the Internet with a five-point Likert-type scale, where 1 corresponded to ‘few important’ and 5 to ‘very important’. The two groups were the following: the CAPTCHA group of 117 respondents and the Test group of 116 respondents.

Data analysis

Analyses were performed via Statistical Package for Social Sciences (SPSS, Version 22.0) and included reliability estimates by means of Cronbach’s alpha (TLX scale NASA), contingency tables with Pearson’s chi-square test and the standardized residuals to identify the cells of the contingency tables, which are responsible for a significant overall chi-square (Field, 2009: 698–700). The differences between the mean values of the two groups were assessed by using a t-test for independent samples, univariate analysis of variance (ANOVA) and multivariate analysis of variance (MANOVA).

Results

The concrete experience of the tests

How was the concrete experience, on the part of our respondents, of doing CAPTCHA and logic and basic culture tests? Table 2 summarizes the correct and wrong answers collected for each CAPTCHA, as well as how many attempts were necessary for the respondents to complete the CAPTCHA we proposed. The tests are presented in descending order of correct executions.

Table 2.

Numbers of correct and wrong answers and number of attempts necessary for respondents to complete the CAPTCHA tests (N = 117).

	Correct executionsAbsolute numbers and %	Wrong executionsAbsolute numbers and %	AttemptsAbsolute numbers
CAPTCHA 7 (reCAPTCHA)	110 (94.0)	7 (6.0)	2
CAPTCHA 8 (textual insert contained in the image)	106 (90.6)	11 (9.4)	4
CAPTCHA 2 (3D image)	102 (87.2)	15 (12.8)	Only an attempt was possible
CAPTCHA 5 (textual insert)	100 (85.5)	17 (14.5)	Only an attempt was possible
CAPTCHA 4 (selection of an image)	97 (82.9)	20 (17.1)	Only an attempt was possible
CAPTCHA 6 (drawing CAPTCHA)	84 (71.8)	33 (28.2)	4
CAPTCHA 3 (audio CAPTCHA)	65 (55.6)	52 (44.4)	Only an attempt was possible
CAPTCHA 1(2D image)	37 (31.6)	80 (68.4)	Only an attempt was possible

CAPTCHA: Completely Automated Public Turing Test to tell Computers and Humans Apart; 2D: two-dimensional; 3D: three-dimensional.

Among the eight types of CAPTCHA, the reCAPTCHA is without doubt the preferred test by our respondents, with 94% correct executions, while the classical CAPTCHA with distorted letters and numbers had the lowest percentage of correct executions (31.6%). The average percentage of correct executions of these tests is 74.9%.

With the purpose of acquiring other important elements for understanding how respondents experienced the execution of the CAPTCHA tests, we investigated other two variables. These have the same structure, and for this reason we illustrate them in Table 3. The first is which of the CAPTCHA tests respondents did more willingly and the second is which of the CAPTCHA tests respondents found more difficult. Table 3 shows that there seems to be a certain correspondence between the sense of pleasure respondents felt when doing these tests and their perception of the degree of ease. In fact, the CAPTCHA 4, which requires the selection of an image among several, presents the lowest degree of difficulty for respondents, and vice versa, CAPTCHA 1 which requires rewriting a two-dimensional image is the test that presented the highest degree of difficulty for the respondents. At the same time, the former is also considered difficult by a low percentage of participants (2.6%), while the second is the one that is judged the most difficult by the highest percentage of participants (41.0%).

Table 3.

The CAPTCHA tests that respondents did more willingly and those that they found more difficult to do.

	More willinglyAbsolute numbers and %	More difficultAbsolute numbers and %
CAPTCHA 1 (2D image)	16 (13.7)	48 (41.0%)
CAPTCHA 2 (3D image)	22 (18.8)	10 (8.5)
CAPTCHA 3 (audio CAPTCHA)	23 (19.7)	12 (10.3)
CAPTCHA 4 (selection of an image)	65 (55.6)	3 (2.6)
CAPTCHA 5 (textual insert)	29 (24.8)	8 (6.8)
CAPTCHA 6 (drawing CAPTCHA)	51 (43.6)	20 (17.1)
CAPTCHA 7 (reCAPTCHA)	45 (38.5)	4 (3.4)
CAPTCHA 8 (textual insert contained in the image)	18 (15.4)	12 (10.3)
Total	177 (100.0)	117 (100.0)

CAPTCHA: Completely Automated Public Turing Test to tell Computers and Humans Apart; 2D: two-dimensional; 3D: three-dimensional.

Furthermore, these results are in line also with the objective data emerging from the execution task and reported in Table 2. At this point, we compared the CAPTCHA group with the Test group to verify whether the two tasks we assigned were perceived differently regarding ease and fun. The χ² test applied to the contingency table between perceived easiness and groups shows that there is no significant difference between the two groups (χ² = 1.037, df = 1, p = n.s.). A total of 91.5% of CAPTCHA group respondents and 94.8% of Test group respondents found it easy to complete the CAPTCHA and the logic tests. The t-test applied to the average scores with regard to the fun felt by the two groups shows instead that there is a significant difference between the two groups (t₂₃₁ = −4.788, p < 0.001): the group that did the logic tests had more fun than the CAPTCHA group (M = 3.74 vs M = 3.01).

Finally, we investigated the emotions that the two groups felt when doing these tests. The most chosen emotions for the CAPTCHA group are curiosity (N = 41), indifference (N= 29) and boredom (N= 14), while for the TEST group they are curiosity (N= 43), interest (N= 22), indifference (N= 16) and fun (N= 12). Just as these frequencies illustrate well the different emotional states in which respondents did the CAPTCHA tests and the logic tests. The χ² test applied to the contingency table between emotions and groups confirms that the differences between the two groups are significant (χ² = 23.728, df = 9, p < 0.006) and that in particular the difference concerns the interest (st.res. = −2.0) and the boredom (st.res. = 1.9). Interest for the Test group was much higher (75.9% vs 24.1%), while the boredom affected the CAPTCHA group more (82.4% vs 17.6%).

Before proceeding with the analysis of the differences between the CAPTCHA group and the Test group regarding the TLX NASA scale, we proceeded to verify its reliability by calculating Cronbach’s α (Table 4). Moreover, in Table 4 we report the average scores of the two groups regarding the question: How much have you felt diminished in your human identity? (with a five-point Likert-type scale where 1 is ‘not at all’ and 5 ‘very much’). This question, which is immediate and bears directly on the issue we investigate in this study, helps us to capture precisely to what degree respondents feel disturbed about their own identity as human beings after completing CAPTCHA tests.

Table 4.

Reliability of the measures and comparison of the average means of the two groups.

	Cronbach’s α	Average meansCAPTCHA group	Average meansTest group	F_univariate (1,231)	p <
Mental demand	0.84	95.03 (SD = 69.03)	123.07 (SD = 69.07)	9.60	.003
Physical demand	0.68	119.35 (SD = 67.81)	129.00 (SD = 67.76)	1.18	n.s.
Temporal demand	0.68	51.44 (SD = 40.87)	46.88 (SD = 36.86)	.80	n.s.
Performance	0.65	51.44 (SD = 40.87)	46.88 (SD = 36.86)	.80	n.s.
Effort^a	/	22.10 (SD = 19.89)	30.34 (SD = 20.69)	9.58	.003
Frustration^a	/	22.26 (SD = 23.70)	21.14 (SD = 20.48)	.15	n.s.
Satisfaction^a	/	50.08 (SD = 31.66)	57.86 (SD = 29.49)	3.77	.05
Identification with Humanity	0.80	2.48 (SD = 0.91)	2.51 (SD = 0.86)	.08 n.s.
How much have you felt diminished in your human identity?	–	1.95 (SD = 1.19)	1.50 (SD = 0.57)	11.54	.002

CAPTCHA: Completely Automated Public Turing Test to tell Computers and Humans Apart.

In these cases, the dimensions were assessed by a single question.

The scores of the two groups related to the seven³ TLX NASA scales were submitted to MANOVA analysis, which showed a significant multivariate effect due to the factor ‘group’ (F[6, 226] = 4.62, p < 0.0001). The analysis of univariate effects (Table 4) shows that the two groups differ in respect to the dimensions of mental demand, effort and satisfaction. Examination of the average scores shows that the execution of the logic tests required from the subjects more cognitive engagement and more effort and gave rise to more satisfaction in the performance of the task in comparison to the execution of the CAPTCHA tests.

The two groups of respondents felt themselves differently diminished in their human identity after completing their respective tasks. Those belonging to the CAPTCHA group (M= 1.95, SD = 1.19) felt more diminished than those belonging to the Test group (M= 1.50, SD = 0.57), t₂₃₁ = 3.40, p < .002.

Using MANOVA analysis, we evaluated whether the socio-demographic variables we considered interact with the execution of the two tasks (CAPTCHA or logic test) in determining the answers to the questions on cognitive effort (TLX NASA scales) and in the perception of being diminished in their human identity. From these analyses emerged a significant multivariate effect of education (F[10, 448] = 1.93, p < 0.04): the analysis of univariate effects shows that the most educated respondents feel less diminished in their humanity when carrying out the two assigned tasks (F[2, 227] = 3.99, p < 0.03). No interaction effect was significant.

Discussion and final remarks

Among the eight types of CAPTCHA proposed in this study, the reCAPTCHA is without doubt the test preferred by our respondents, with 94% correct executions. In the second place is the textual insert contained in an image with 90.6% and the three-dimensional letters and numbers with 87.2%. In contrast, the classical CAPTCHA with distorted letters and numbers had the lowest percentage of correct executions (31.6%). The average percentage of correct executions of these tests in this research is 87.3%.

Regarding our first hypothesis that the CAPTCHA execution is associated with forms of users’ disturb and/or alienation, since users are asked to lose time and produce a cognitive effort for reasons that are extraneous to their interests, the summary of the main results we obtained are:

The scores of the six TLX NASA scales showed that the performance of logic and basic culture tests required more cognitive engagement and more effort but brought more satisfaction than the performance of CAPTCHA tests.

CAPTCHA and Test groups did not evaluate differently the easiness of the tasks.

The Test group found the performance of the task significantly more amusing.

As to the emotional reaction to the performance of the tasks, the CAPTCHA group suffered more boredom, while the Test group felt more interest in their task.

This is in line with the results obtained by the research carried out by Milyavskaya et al. (2018), who demonstrated that people in a state of boredom report greater subjective fatigue than those in cognitive effort condition.

Thus, our first hypothesis is confirmed. The results of our study show that the execution of the CAPTCHA tests is more alienating than the experience with the logic and basic culture tests. These last tests aim to test the logical reasoning of respondents with the sole purpose of making them entertained and to provoke a self-assessment of their abilities. On the other hand, the CAPTCHA tests, first of all, are not chosen by Internet users but are imposed by website owners; and second, they are part of the parasite activities that Internet users have to undertake to properly access many websites and that take a lot of time from them (Ekbia and Nardi, 2017; Terranova, 2004). On the shoulders of Internet users, an increasing amount of work of learning and continuous updating is being reversed, which makes them more and more immaterial workers. CAPTCHA tests are only a sign of the burdensome conditions experienced by Internet prod-users (Bruns, 2008).

With respect to our second hypothesis that the execution of the CAPTCHA tests, by asking immediately for proof of not being a robot, would diminish a person’s self-perception as a human being, we have obtained a clear result.

The scores on the question of whether the execution of the CAPTCHA tests, by asking immediately for proof of not being a robot, would diminish a person’s self-perception as a human being show that those belonging to the CAPTCHA group felt more diminished in their human identity than the respondents of the Test group. We also found an effect of education: with increasing educational levels, the entire sample felt less diminished in their humanity. This probably means that users with robust cultural tools are more equipped to defend themselves from attempts to devalue their human identity.

The execution of CAPTCHA tests is not able to reach the core of the human identity, because the latter contains social dimensions, which refer to bonds that individuals share within their human group. By contrast, it affects the perception of the self as human being because, as we mentioned before, one of the strategies of dehumanization is ‘to equate humans to machines, automata or robots’ (Volpato, 2011). In this case, the situation is even more dehumanizing, since humans are not merely equated to robots but have to demonstrate that they are not robots before presenting themselves as humans with a personal identity.

Let us conclude with some final remarks on the strengths and weaknesses of the study, as well as on the future direction, which this research needs to take. The strong point of this study is its attempt to explore an issue scarcely studied up to now. The weak point is that the non-statistical representativeness of the two groups does not enable us to generalize the results of this study.

However, in spite of the inevitable limitations that this study presents, it has the merit of travelling along paths that have been unexplored so far and to open new directions of research both in the new forms of alienation that Internet users experience online and on measuring the identification with all humanity that surely deserves to be further investigated. Furthermore, it makes the recommendation to design the formal procedures that constitute part of users’ daily experience online not only with the interest and convenience of website owners in mind but also with an eye on the feelings and preferences of users themselves. In other words, the design of such procedures should be user-centric, looking for ways to advance valuable human skills and stimulate self-development and confidence as opposed to causing frustration and alienation. In this regard, this article also launches an inquiry into possible ways of designing positively experienced procedures.

Footnotes

Acknowledgements

We thank Andrea Bellotto for helping to collect the data. This paper was presented and discussed at the ICA preconference ‘Communicating with Machines: Theory and Practice’, 24 May 2018, Prague.

Funding

The author(s) received no financial support for the research, authorship and/or publication of this article.

Notes

ORCID iDs

Leopoldina Fortunati

Filippo Cavallo

Author biographies

Leopoldina Fortunati is a full professor of Sociology of Communication and Culture at the Department of Mathematics, Computer Science and Physics of the University of Udine and is the director of the laboratory of research on new media NuMe. For some years she has been studying and doing research on the social and communicative aspects of the robotics diffusion and adoption in the sphere of individual and social reproduction.

Anna Maria Manganelli is a full professor of Social Psychology at University of Padua, Department of Department of Philosophy, Sociology, Education and Applied Psychology. She carried out researches projects in the fields of social psychology (theories of prejudice and measurement of “modern” forms of prejudice and sexism), communication and information technologies, methodology of research. She published in journals like Journal of Economic Psychology, Personality and Individual Differences, Leadership, Health Care Management Review, International Journal of Nursing Studies, Information & Society.

Filippo Cavallo, MScEE, PhD in Bioengineering, is assistant professor at the BioRobotics Institute of Scuola Superiore Sant’Anna, Pisa, Italy, focusing on cloud and social robotics, wireless and wearable sensor systems, biomedical processing, acceptability and ICT roadmapping. He participated in various National and European projects and is author of various papers on conferences and ISI journals.

Furio Honsell got his degrees at the University of Pisa in Mathematics and Logic. In the 80s he worked as research fellow at the University of Torino and Edinburgh. He was among the developers of the Edinburgh Logical Framework, a proof development environment for proving software reliability, which received a IEEE Test-of-time award. He is full professor in Automata Theory at the University of Udine since 1990. He served as president of the University of Udine from 2001 to 2008 and then as Mayor of Udine from 2008-2018, position in which he promoted the European digital agenda.

References

Abrich

Berbenetz

Thorpe

(2011) Distinguishing between humans and robots on the web. Available at: http://richardabrich.com/abrich-berbenetz-thorpe_thecaptchaexperiment.pdf

Adler

([1929] 1964) Problems of Neurosis. New York: Harper Torchbooks.

Andrighetto

Baldissarri

Volpato

(2017) (Still) modern times: objectification at work. European Journal of Social Psychology 47(1): 25–35.

Bakardjieva

(2015) Rationalizing sociality: an unfinished script for socialbots. The Information Society 31(3): 244–256.

Balbi

Magaudda

(2017) Storia dei media digitali.Rivoluzioni e continuità. Roma: Laterza.

Baldissarri

Andrighetto

Volpato

(2014) When work does not ennoble man: psychological consequences of working objectification. Testing, Psychometrics, Methodology in Applied Psychology 21(3): 1–13.

Baltar

Brunet

(2012) Social research 2.0: virtual snowball sampling method using Facebook. Internet Research 22(1): 55–74.

Barile

Sugiyama

(2015) The automation of taste: a theoretical exploration of mobile ICTs and social robots in the context of music consumption. International Journal of Social Robotics 7(3): 407–416.

Bar-Tal

(1989) Delegitimation: the extreme case of stereotyping and prejudice. In: Bar-Tal

Grumann

Kruglanski

et al. (eds) Stereotyping and Prejudice, Changing Conception. New York: Spinger-Verlag, pp. 169–182.

10.

Bauman

(2003) Intervista sull’identità. Roma; Bari: Editori Laterza.

11.

Brodić

Petrovska

Jevtić

et al. (2016) The influence of the CAPTCHA types to its solving times. In: 39th International Convention on Information and Communication Technology, Electronics and Microelectronics (MIPRO). Available at: http://ieeexplore.ieee.org/abstract/document/7522335/

12.

Brown

(2004) Human universals, human nature & human culture. Daedalus 133(4): 47–54.

13.

Bruns

(2008) Blogs, Wikipedia, Second Life, and Beyond: From Production to Produsage (Digital Formations). New York: Peter Lang.

14.

Capozza

Brown

(2000) Social Identity Processes: Trends in Theory and Research. London: SAGE.

15.

Dehaene

Lau

Kouider

(2017) What is consciousness, and could machines have it? Science 358(6362): 486–492.

16.

Dovidio

Gaertner

Validzic

(1998) Intergroup bias: status, differentiation, and a common in-group identity. Journal of Personality and Social Psychology 75(1): 109–120.

17.

Ekbia

Nardi

(2017) Heteromation, and Other Stories of Computing and Capitalism. Cambridge, MA: The MIT Press.

18.

Esposito

Fortunati

Lugano

(2014) Modeling emotion, behavior and context in socially believable robots and ICT interfaces. Introduction to the Special Issue of Cognitive Computation 6(4): 623–627.

19.

Fidas

Voyiatzis

Avouris

(2011) On the necessity of user-friendly CAPTCHA. In: Proceedings of the international conference on human factors in computing systems (CHI 2011), Vancouver, BC, Canada, 7–12 May, pp. 2623–2626. New York: ACM Press.

20.

Field

(2009) Discovering Statistics Using SPSS. 3rd ed. London: SAGE.

21.

Fortunati

(2013) Afterword: Robot conceptualizations between continuity and innovation. Intervalla: Platform for Intellectual Exchange 1(1): 116–129.

22.

Fortunati

(2018) Robotization and the domestic sphere. New Media & Society 20(8): 2673–2690.

23.

Fortunati

Taipale

(2017) A different glimpse into mobilities: On the interrelations between daily spatial mobility and social mobility. The Information Society 33(5): 261–270.

24.

Gaertner

Dovidio

Anastasio

et al. (1993) The common ingroup identity model: recategorization and the reduction of intergroup bias. In: Strobe

Hewstone

(eds) European Review of Social Psychology, vol. 4. London: Wile, pp. 1–26.

25.

Gafni

Nagar

(2016) CAPTCHA – Security affecting user experience. Issues in Informing Science and Information Technology 13: 63–77.

26.

Gehl

Bakardjieva

(eds) (2017) Socialbots and Their Friends: Digital Media and the Automation of Sociality. New York: Routledge.

27.

Hall

(2018) Comparing globalizations: conclusions, questions, speculations. In: Hall

(ed.) Comparing Globalizations: Historical and World-Systems Approches. Cham: Springer, pp. 1–12.

28.

Hart

Staveland

(1988) Development of NASA-TLX (Task Load Index): results of empirical and theoretical research. Advances in Psychology 52: 139–183.

29.

Harte

(1995) Ecology, sustainability, and environment as capital. Ecological Economics 15(2): 157–164.

30.

Höflich

(2013) Relationships to social robots. Intervalla: Platform for Intellectual Exchange 1(1): 35–48.

31.

Lin

et al. (2018) Chinese Character CAPTCHA Recognition and performance estimation via deep neural network. Neurocomputing 288: 11–19.

32.

Lovejoy

James

(1936) The Great Chain of Being: A Study of the History of an Idea. Cambridge, MA: Harvard University Press.

33.

McFarland

Webb

Brown

(2012) All humanity is my ingroup: a measure and studies of ‘identification with all humanity’. Journal of Personality and Social Psychology 103: 830–853.

34.

Maslow

(1954) Motivation and Personality. New York: Harper & Row.

35.

Milyavskaya

Inzlicht

Johnson

et al. (2018) Reward sensitivity following boredom and cognitive effort: a high-powered neurophysiological investigation. Neuropsychologia 123: 159–168.

36.

Monroe

(1996) The Heart of Altruism: Perception of a Common Humanity. Princeton, NJ: Princeton University Press.

37.

O’Sullivan

Fortunati

Taipale

et al. (2017) Innovators and Innovated: newspapers and the post digital future beyond the ‘death of print’. The Information Society 33(2): 86–95.

38.

Oliner

(1988) The Altruistic Personality: Rescuers of Jews in Nazi Europe. New York: Free Press.

39.

Pfadenhauer

Sugiyama

Ess

(eds) (2015) Social robots: form, content, critique. International Journal of Social Robotics 7(3): 333–334.

40.

Pieper

(2012) Sociological issues of inclusive web design. In: Miesenberger

Karshmer

Penaz

et al. (eds) Computers Helping People with Special Needs (ICCHP 2012: Lecture Notes in Computer Science), vol. 7382. Berlin; Heidelberg: Springer, pp. 371–377.

41.

Puddifoot

(1995) Dimensions of community identity. Journal of Community & Applied Social Psychology 5(5): 357–370.

42.

Reese

Proch

Finn

(2015) With all humanity: The role of self-definition and self-investment. European Journal of social Psychology 45(4): 426–440.

43.

Reysen

Katzarska-Miller

(2013) A model of global citizenship: antecedents and outcomes. International Journal of Psychology 48(5): 858–870.

44.

Sciolla

(1997) Italiani.Stereotipi di casa nostra [Italians and Self-stereotypes]. Bologna: Il Mulino.

45.

Singh

Pal

(2014) Survey of different types of CAPTCHA. International Journal of Computer Science and Information Technologies 5(2): 2242–2245.

46.

Smith

(1991) National Identity. Reno, NV: University of Nevada Press.

47.

Steichen

(1955) The Family of Man: A Photographic Exhibition Created by Edward Steichen for the Museum of Modern Art. New York: Simon & Schuster.

48.

Steichen

(1963) A Life in Photography. Garden City, NY: Doubleday.

49.

Suchman

(1987) Plans and Situated Actions: The Problem of Human-Machine Communication. Cambridge, MA: Cambridge University Press.

50.

Sugiyama

(2013) Melding with the self, melding with relational partners, and turning into a quasi-social robot: a Japanese case study of people’s experiences of emotion and mobile devices. Intervalla: Platform for Intellectual Exchange 1(1): 71–84.

51.

Sugiyama

Vincent

(eds) (2013) Social robots and emotion: transcending the boundaries between humans and ICTs. Intervalla: Platform for Intellectual Exchange 1(1): 2–6.

52.

Taipale

de Luca

Sarrica

et al. (2015) Europeans’ perception of robots implications for social policies. In: Vincent

Taipale

Sapio

al.

(eds) Social Robots from a Human Perspective. Berlin: Springer, pp.11–24.

53.

Taipale

Turja

Van Aerschot

(in press) Robotization of mobile communication. In: Ling

Fortunati

Goggin

al.

(eds) Handbook on Mobile Communication, Culture, and Information. New York: OUP.

54.

Terranova

(2004) Politics for the Information Age. London: Pluto Press.

55.

Turing

(1950) Computing machinery and intelligence. Mind 49: 433–460.

56.

Turner

Hogg

Oakes

et al. (1987) Rediscovering the Social Group: A Self-Categorization Theory. Cambridge, MA: Blackwell.

57.

Volpato

(2011)Deumanizzazione.Come si legittima la violenza [Dehumanization: How to legitimize violence]. Roma; Bari: Editori Laterza.

58.

Walther

(1996) Computer-mediated communication: impersonal, interpersonal, and hyperpersonal interaction. Communication Research 23(3): 3–43.

59.

Wellman

(2001) Physical place and cyberplace: the rise of personalized networking. International Journal of Urban and Regional Research 25: 227–252.

60.

Zeifman

(2017) Bot traffic report 2016. Incapsula. Available at: https://www.incapsula.com/blog/bot-traffic-report-2016.html (accessed 1 February 2017).