Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
1
PERSUASION TECHNOLOGIES SHORT NAME: PER TE
Oliviero Stock
January 2012
2
MOTIVATIONS
The traditional goal of human-‐oriented information technology is mostly to offer services. In the present strategic project, instead, the overall goal is to produce an effect on humans, to influence their beliefs, their attitudes and eventually their actions and overall behavior. This goal is ambitious and is in line with a two millennia old sector of study (including communication, rhetorics, argumentation) and with a very large applied interest that spread in the Twentieth Century and developed by the day (advertising, social communication). ICT has the potential to change the picture radically. There is a fundamental factor that makes the situation different from the tradition in applications, which is substantially based on broadcast messages: computer-‐based systems can be flexible. Starting from goals they have to pursue, they can take into account the situation and the specific target, adapt the messages in appropriate ways and assess the outcome. In addition the availability of very large amounts of data which can be exploited also in real time provides unprecedented possibilities.
Theories tend to define two routes to persuading people: one central – direct, rational, argumentative; and one peripheral – indirect, evocative, aesthetic (Petty and Cacioppo 1986). The second one has had recently a growing importance in all situations of idea promotion and in particular a fundamental role in the world of advertising.
For instance, it is easy now to predict these developments for the advertising sector:
a) reduction in time to market and extension of possible occasions for advertisement; b) overall reduction of off target messages, eliminating the less relevant for the individual in a given situation; c) more attention to the wearing out of the message and to the need for planning variants and connected messages across time and space; d) contextual personalization, on the basis of audience profile and dynamic model (emotional state, beliefs, goals, etc.) and situational information; e) interactivity. f) audience reaction monitoring and system feedback on effectiveness.
Even pervasive and ubiquitous computing is less useful without intentionally producing human behavioral change. The possibilities of intervention are extremely wide, and we can think of a new situation where machines can help influence society for the better. Of course it is important to be in control of the involved ethical issues, rather than just see them as an undesired nuisance.
This project aims at contributing to this novel view with the development of original technology and leading to some significant experiments.
3
INTELLIGENT PERSUASION TECHNOLOGY AND ITS RELATION TO CAPTOLOGY
The object of this project is related to Captology, but it emphasizes flexibility and intelligent behaviour. Captology (Fogg 2002, Fogg 2007) is defined as the study of computers as persuasive technologies. This includes the design, research, and analysis of interactive computing products (computers, mobile phones, websites, wireless technologies, mobile applications, video games, etc.) created for the purpose of changing people’s attitudes or behaviors. B.J. Fogg derived the term captology in 1996 from an acronym: Computers As Persuasive Technologies = CAPT.
Captology, and similarly our project, is based on the belief that much like human persuaders, persuasive technologies can bring about positive changes in many domains, including health, business, safety, and education; and even that new advances in technology can help promote world peace in 30 years.
What characterizes the present project is the specific focus on intelligent technologies for persuasion. We look into some original technologies that exploit knowledge and experience in developing intelligent user-‐oriented systems, that are in the background experience of the participating institutions. This is somehow distinct from the larger prospect of Captology, which tends to focus on more quickly achievable but less flexible systems. For instance Captology does not focus specifically on systems automatically adaptable to individuals and small groups behavior. Of course there is much more, more traditional or popular themes that could be ascribed to persuasion technology: we do not aim at all to cover the whole area, but rather to develop some specific creative themes.
BUSINESS AND PUBLIC POTENTIAL
This kind of technologies has a large potential in many business sectors: first of all, of course, advertisement but also communication in big organizations (to support and promote communities of practices or to share mission statements); awareness in dangerous working activities (for example in factories and plants); and last but not least promotion of better quality of life and social awareness.
Specifically, it is well known that some of the biggest industrial ICT actors make their profits with advertising. For instance Google has reported about $23.650 billion in 2009 . This enormous “opportunistic” business, that is really offering the infrastructure for ads, is complementary to a well established advertising industry. It is interesting to note that in the UK, online advertising spending has overtaken television expenditure for the first time in 2009. In the US, according to E-‐Marketer, the online advertising industry is estimated to achieve $25.8 billion in revenue in 2010, for the first time exceeding print advertising revenue at “only” $22.8 billion. Global advertising expenditure will grow to nearly $550 billion in 2011, from about $375 billion in 2003, according to Zenith Optimedia. In 2011, TV and newspapers together are
4
expected to lose about a 10 percent share, while the Internet is up from 3 percent to 16 percent, according to the report, World Digital Media Trends 2009, released by SFN. In substance we are talking of some $90 billion. Yet we are only at the beginning of a novel more pervasive approach, one that at the same time is more efficient and more respectful of the individual. Systems can now be ubiquitous, with a combination of mobile and fixed media to support the task. And novel types of situations and technological means can be introduced. For instance technology can intervene in social situations, like a family context, or in the activity of a group of friends’.
It should be noted that a Google Award (the only one to an Italian in 2011) was just received by Carlo Strapparava, a Trento-‐Rise scientist in creative language persuasion, specifically because of the industrial attractiveness of the theme proposed.
On the public side, there is an extraordinary potential for using intelligent technology for orienting people toward social values, in all areas: health, respect for minorities, environment, energy saving, education etc. In all countries there are prestigious public organizations that coordinate “traditional” promotion campaigns. For instance in the United States the AD Council is such an organization. Leading producer of public service advertisements (PSA’s) since 1942, The Ad Council has been addressing critical social issues for generations of Americans. The Ad Council has helped create some of America’s most memorable slogans, such as “Friends don’t let friends drive drunk” and “A mind is a terrible thing to waste”.
We believe socially oriented technologies and novel context-‐aware techniques will open a new avenue to learning to live together in a better way.
ACTIVE AREAS OF RESEARCH IN TRENTO POTENTIALLY INVOLVED (NOT EXHAUSTIVE)
Potentially involved areas include several ICT fields as well as social and cognitive psychology, sociology and communication studies. Trento RISE and its partners have well established competence and skills in all the relevant areas, and have been playing a pioneering role in bringing the topic of persuasion technology to the attention of the research community. In addition, we are well equipped also for dealing with the ethical aspects that the investigation of these topics inevitably raises.
A table of areas that potentially can be involved is shown below. In this project we shall focus only on some specific areas on which preliminary work on persuasion themes was conducted in the past (and for which we see an important applied potential).
Collaboration within EIT-‐ICT Labs: strong common interests exist in particular with Berlin and DFKI (explicitly for the intelligent transportation theme) and with University of Twente, with SICS (in particular for indirect interfaces) and possibly with Philips (in particular for what concerns behavior analysis). We intend to involve strong industrial partners both locally and internationally.
5
Planning. Example: planning and adapting a personalized promotion campaign Argumentation and negotiation. Example: reason on pros and cons for a situated buying decision Indirect communication technology. Example: use of peripheral devices in everyday form factors (such as table surfaces or mirrors) for promotion and behavioral change Behavioral analysis. Example: assessing high level attitude and affect features out of sensory signals, that can influence message decision, and subsequently assessing message effect Multimodal interaction. Example: make use of appropriate modalities for delivering a message in a flexible situation, engage in an interaction recognizing communicative and affective expressions Group interfaces. Example: assessment of group attitudes, roles and behaviors to produce coordinated messages to group members in a common situation, like a car trip or sitting at a café Natural language generation. Example: flexible message production, taking into account individual characteristics Natural language understanding. Example: understanding their persuasive effect and spreading potential within a community. Computational humor. Example: dynamic production of personalized humorous slogans. Humor is a fundamental resource in advertising (90% of ads in the UK include some form of humor) Personality and user modeling. Example: assessing the individual’s personality and preferences so to choose the appropriate form of humorous strategy Effectiveness evaluation and socio-‐technological studies. Example: techniques for evaluating the effect of individual oriented messages Ethics. Example: define appropriate guidelines; instill in intelligent system fundamental ethical metarules they will respect when deciding messages.
6
GOALS, APPROACH AND TECHNOLOGY
In the present project we focus mainly on what was called the peripheral route to persuasion. So only marginally we consider persuasion tools like argumentation; they are the object of other substantial studies in AI and several important groups are active in the field.
Very rarely has human change been formulated as an explicit goal of a computer system, and when this has been the case the focus has been quite narrow (e.g., recommender systems), targeting specific modifications (e.g., purchase inclination) through a shallow understanding of the relevant parameters (social and individual behaviors, attitudes, etc.). Only recently, attempts at bringing about broader modifications of people attitudes and behaviors have been pursued, as in the case of captology.
Its foundational nature resides in the observation that the capability of affecting other beings is foundational for autonomous agents: there are no prospects for the whole endeavour of pervasive and ubiquitous computing unless machines are made capable of purposely affecting the other entities living in their environment—most notably, humans.
Despite some initial attempts, computer-‐induced human change is still a largely unexplored topic. We do not understand much of the cognitive and psycho-‐social conditions under which it can take place, and miss appropriate computing and interaction paradigms that enable systems to a) sense and understand human behaviors as they are deployed in naturalistic settings; b) reconstruct the attitudes, motivations, beliefs, etc., that underlie them; c) deploy actions to influence and change behaviors and attitudes in accordance with the system’s goals and with contextually-‐determined constraints; d) ensure that system’s goals and actions comply with generally accepted ethical principles concerning human empowerment and human values promotion.
Our attention is devoted to novel scenarios and our view is that intelligent technology can intervene to change attitudes in ways that were not conceivable before. For this, rather than present an overarching approach we shall focus on some specific, appealing issues.
Activity will be organized upon four areas: Persuasion for co-‐located Small Groups (including behavior analysis and modeling), Persuasion based on Natural Language -‐ creative production of persuasive short messages and defense from inappropriate persuasive messages, Persuasion through awareness and collaborative activity, and Computational Ethics.
Assessing the state of groups or individuals and in general modeling them, will be based on a) inertial features like: profile, social roles, personality, as well as b) dynamic aspects such as interests, collaboration and participation.
7
1. BEHAVIOR-‐BASED PERSUASION FOR COLOCATED SMALL GROUPS
This activity is based on novel ways for changing the attitudes, beliefs or behavior of a small group of people. It is based on three main components: a) observation and understanding of the behavior of the persons in the group and of the group dynamics in a real environment; b) appropriate production of targeted multimodal messages on the basis of reasoning and specific persuasive techniques; c) assessment of the impact. It requires sensory system infrastructure, techniques for understanding high level features of the group and its state, study of appropriate techniques for persuasive output in context. The situation can be mobile or stationary and output devices can be large shared displays of various sorts, or coordinated individual devices. The specific output techniques will have to exploit specifically the natural group dynamics to maximize the impact of the message.
A first example we have developed is a prototype tabletop application, ideally to be placed in the museum café. The table is instrumented with sensors and its top surface is used as a medium to display persuasive messages aimed at influencing the conversation of the group, after the visit to the museum. The group's behavior is monitored and reasoning about the overall conversation configuration and the visit to the museum permits to drive the system's actions: the system chooses presentation strategies that lead to specific output on the tabletop. Applied situations that can benefit from this include everyday life scenarios, such as the café, a school cafeteria or a bus stand or sitting in front of the TV set, or possibly a mobile setting like a family out for shopping, or a group of colleagues at a trade fair. For this project, we will evaluate the scenario of internal communication specifically aimed to support communities of practices (see below).
The goals of the applications can be divided in three broad areas: informal learning/edutainment, advertising, internal corporate communication.
We will investigate, prototype and validate a new class of systems aimed at purposely but implicitly inducing change in small groups of users. These systems will
• understand and trace social dynamics and individual traits of informal, non goal-‐oriented small groups (NoGoG); • exploit the assessed dynamic state of the group to plan and deploy minimalist strategies using evocative means; • enforce indirect interaction modalities, leaving the main channel to direct human-‐to-‐human interaction; • use its own influence capability to enforce individual and societal values, this way paving the way for ethical change inducing systems. The project will validate and concept-‐prove its findings in a few scenarios, incorporating the results into a global framework that integrates theoretical (psycho-‐social), technological, and ethical/legal issues bearing on the computer induced change.
The novelty of the proposal is its original mixture of topics and themes (persuasive technologies, monitoring and understanding of people behavior, psycho-‐social theories of
8
behavior and attitudes, etc.) and to their application to a brand new and challenging scenario.
The very rough and coarse grained version of the model for computer-‐induced change is depicted in figure 1 below.
Fig. 1. Conceptual architecture (SPT= Self Perception Theory. TPB=Theory of Planned Behaviour)
The influence from the computer flowing towards the human(s) is the product of a number of computational components composing this kind of system, which are built upon and rooted on concepts and notions from social and cognitive psychology.
From the technological point of view, a sensor network is required to collect information about the environment and model high level behavioral traits, as described above, and a way of realizing communicative strategies. For the former, a minimal infrastructure includes environmental sensors (presence, pressure, etc.; more invasive, wearable sensors may be considered as well), cameras to detect a span of core behaviors, from visual attention to emotions; and possibly speech recognition for simple topic detection or restricted dialogue understanding. For what concerns the communicative strategies: simple visualizations tools (like the one experimented in the café table) may work, but more elaborated techniques for natural language generation (as described below) would increase control of the persuasive effect; other modalities: audio (synthesized, music, environmental audio, lighting) should also be considered.
Our proposal bears some affinity with concepts underlying peripheral displays (Weiser and Brown, 1996) in that our systems are not meant to be central to the attention of the group and people may look at the interface only occasionally. Peripheral displays are normally used as secondary sources of information/stimuli, distinct from the user’s primary and focal one (Matthews et al., 2004); they often have a passive role, just aiming at making users aware of easily graspable information such as weather or stock graphics. In our approach,
TPBSPT
Interfaces and influence
deployment
Planning and scheduling
TPB, SPT,Petty’s and Cacioppo’s
peripheral route to attitude
Behaviour
Attitudes
Sensing the environment
Thin slicesZero acquaintance
Multimodal fusion Interpretation of social behaviour
9
however, the peripherality/centrality of the display is something that is manipulated by the system within the general constraints of our minimalist approach: depending on the strategy chosen (e.g., the peripheral route), the display remains peripheral until the system decides it is time to induce a change and raise the attention of the group towards it. According to the minimalist approach, however, even in this case the amount of direct engagement with the system will be minimal and interaction with it will still remain a secondary task for people (the primary one remaining interaction with the other members of the group). We refer to the new human-‐computer interaction paradigm embodying the minimalist precepts as indirect interaction.
Another specific theme is concerned with mobility. The goal is still influencing a small group, in this case our experience and current goals are mostly in relation to an initial phase of a visit to a the museum, or a visit to a specific territory.
Digital storytelling, which combines adaptivity with drama, offers the possibility of providing different viewpoints of a story at the same time for its members, and to adapt to the behavior of the group as a whole, rather than just individual behavior. In traditional theatre, as in most museums that adopt the drama genre, the audience identically observes the entire scene and together is exposed to the actors’ words, actions and emotions. There are also forms of modern theatre where different parts of the public are exposed to different portions of the scene (e.g., a visual barrier separating parts of the audience). With mobile devices this idea can be extended further: to adopt distinct narrations for different members of a small group while they are observing the same dramatic scene. By presenting group members with slightly different (either complementary or contrastive) versions of the same multimedia dramatic narrative, we believe they will notice that something is missing and then act on that fact by asking their fellow group members, actively leading to a more general conversation about elements of the museum or the territory. Our overall aim is not only to provide an engaging story but also to influence individuals’ behavior, their attitude, and their expectations so they interact more with other members of the group (including also help induce emotions in human participants). The ensuing conversation is thus inherent in the nature of the drama experience itself and the entailed engagement of participants who are standing in front of actual ancient artifacts, rather than being due to some external goal given to the group, like a treasure hunt or a puzzle.
Our group-‐oriented approach integrates group behavior sensors into a coordinated narrative system for mobile devices based on specific techniques for holding narrative tension, in an instrumented cultural heritage environment such as a museum. The primary goals for our system are: (1) to have a positive effect on the group members’ view of their museum visit, leading to a memorable group cultural experience, and (2) to have an impact on the quality and quantity of conversation during the visit. It is taken for granted that the latter is a fundamental aspect of a successful small group visit: conversation has been shown by [Leinhardt and Knutson 2004] to be the key aspect for learning in the museum.
10
2. NATURAL LANGUAGE-‐BASED PERSUASION: PRODUCTION OF CREATIVE MESSAGES AND DEFENSE FROM INAPPROPRIATE PERSUASION
It is obvious that natural language communication is a basic resource in persuasion, even if we leave aside argumentation. The role of specific lexical items, including their affective traits, the role of evoking expressions and the effect of rhetorical structures is only beginning to be studied in automatic production of effective messages [a survey on approaches and themes regarding verbal persuasion can be found in Guerini et al., 2010]. Creative language, oriented toward getting the attention, evoking relevant concepts and helping memorization is a very powerful means for influencing attitude and behavior. From an application point of view we think the world of advertisement has a great potential for the adoption of creative persuasive language. And of course advertisement and promotion can be for a good cause and for social values!
For instance, as things are now, humanly produced humor appears widely used in advertisement. In the UK, advertisements are based on humor in 33% of all cases and in 93% of those cases that show pertinence, according to a 1998 survey [Toncar ,1998]. [Perry et al., 1997] has shown that perception of humor in promotional messages produce higher attention and in general a better recall than non humorous advertisement of the product category, of the specific brand and of the advertisement itself.
A said in the introduction to the project the future will probably include some important factors: a) reduction in time to market and extension of possible occasions for advertisement; b) more attention to the wearing out of the message and for the need for planning variants and connected messages across time and space; c) contextual personalization, on the basis of audience profile and perhaps information about the situation. Leaving alone questions of privacy, all three cases call for a strong role for computer-‐based intelligent technology for producing novel appropriate advertisements.
More than that: if advertisements will be personalized, also creative language will need to follow, given the success it has had in the “pre-‐technological" world of advertisement [see for instance Testa, 2000]. For instance humor is appreciated differently by different individuals. Personality studies regarding this specific theme give important indications [see Ruch, 2002]. So computational humor generators will need to take into account the personality of the receiver and, if possible, his mood and situation to produce well received expressions. There will not be sufficient creative human beings -‐ and anyway they would cost too much -‐ for producing all the necessary quantity of creative expressions in real time. For one of the few initial works on automated humor production, see [Stock and Strapparava , 2003].
Two technological scenarios will be at the focus of attention:
a) Production of creative messages for the specific user state and context. In order for doing that specific techniques will have to be defined, mostly corpus based. Emphasis will be put on automated variation and creative adaptation of existing expressions. The main aspects that will characterize this part of the project are: humor, affect and pragmatic aspects of language. Personality studies will also be taken into account.
11
It has been shown that variations of familiar expressions (quotations, proverbs, movie titles etc.) is particularly well suited for getting across an evoking message provided it follows the aesthetic principle of optimal innovation [Giora 2002]. For this reason, an advertising message must be original but, at the same time, connected to what is familiar. Preliminary work has also shown the feasibility of an approach aiming at automatic production of those creative variations [see Valitutti et al. ,2008].
An important methodological aspect is impact evaluation, which for general assessments can exploit the web (for instance through crowdsourcing [Mason and Suri 2011] or ecological studies [Guerini et al., 2010]), and for the complementary aspect can be oriented toward assessing the impact on the individual. b) Defense from subtle persuasive language in broadcast messages, including social networks, web advertising and specific cases such as political speech. This latter theme will become very important for the progress in democracy – one important case being defending the audience from political bias in news. “Viral” messages have become a very important factor of persuasion and are currently almost entirely out of control. Current approaches tend to focus only on the social aspect, mostly leaving aside contents -‐ and especially communication means. Study will be oriented both at automatically detecting the use of specific techniques aimed at obtaining a persuasive effect through the net, and to develop defenses against subtle persuasive messages. Though different in scope, reference technology include the work developed by Koppel et al [2009] for detecting text authorship, and by Abbasi and Chen [2005] for detecting authorship of dangerous messages. Protection is also needed in competitive commercial situations. For example preventing deceptive consumer reviews on sites like trip-‐advisor is fundamental both for consumers, seeking genuine reviews, and for the reputation of the site itself. Deceptive consumer reviews are fictitious opinions that have been deliberately written to sound authentic. An interesting example of technology concerned with online advertising is [Sculley et al 2011]; it discusses detecting fraudulent ads in the interest of users, of service providers (e.g. Google Adwords system) and other advertisers.
3. PERSUASION AS AN EFFECT OF AWARENESS AND DISCUSSION
This part will deal with persuasion as an effect of situation awareness and discussion. The idea is that the characteristics of the task, the affordances of the interface, and the overall computer-‐facilitated human activity will lead to a change in attitude. The chosen scenario is of high social importance: we focus on conflicts. So called intractable conflicts can become more tractable if people come to know the humanity of the other side. A common approach has been to involve participants to the two sides of the conflict in producing a joint narration. Several studies have shed light on the effectiveness of this approach. Technology can bring in a very important contribution and actually realize on a large scale concepts that are very difficult to implement without it. And in some cases it may even enable the realization of specific techniques for inducing an attitude change in the involved participants, altering the enmity image by a ‘colleague identity’.
The subproject builds on a previous experience that aimed to support a shift to a more
12
positive attitude toward a peer from another culture (Israeli-‐Jewish versus Palestinian-‐Arab) via a technology supported joint narration task. The rationale and design concepts of this novel co-‐located interface for conflict reconciliation emanated from conflict escalation and de-‐escalation theory with a particular focus on the need for visibility and the notion that increased awareness enables a participant to deal with a dyadic cycle of conflict related actions and reactions [Eisikovits and Winstok 2002]. We aimed to channel escalation and de-‐escalation dynamics that could otherwise lead to physical aggression into a process wherein both parties are perceived to be understanding of and getting closer to each other.. We hypothesized and eventually had experimental confirmation that such dynamics could eventually lead to a shift in attitude or to reconciliation.
The current subproject project has the ambition of producing something practical that will have a social impact and will also include a business prospect. The first target area of conflict is the conflict between Israelis and the Palestinians and the suggested model is applicable to many other conflict areas.
We aim at developing a new system whose main characteristics are:
• It will provide two groups in a situation of conflict with the possibility of developing jointly a good quality multimedia product. It is meant that the multimedia product is concerned with narration, documentation or otherwise inspired by the conflict reality. • The two groups will be remote, possibly across the border. The technology will guarantee “presence” of the other group. Recognition of aspects of the behavior and expression of the other group and aids for the interpretation will be continuously granted.
4. COMPUTATIONAL ETHICS
The theme of ethical behavior in automated systems is novel as a serious general challenge. For several years now there has been attention to issues of privacy for computer systems, but privacy, albeit very important in our society, is a rather narrow theme and in practice it is mostly approached with the focus on the designer and without necessarily connecting it to the autonomous behavior of the system. Our main objective is to try to define a reasonable approach to ethics in autonomous artificial agents, with focus on the case of communication and in particular on persuasion, and propose a set of abstract principles.
We believe the best starting point is in the recent cognitive-‐philosophical literature [see for instance Nichols and Mallon 2006, Dwyer et al 2010]. No ultimate word was pronounced but what appears is that different components intervene in our judgment about the ethical standing of an action. For instance experiments with moral dilemmas have shown there is a role for utilitarian, deontic and emotion-‐based decision-‐making. Some limited attempts have also been made to reproduce moral decision making computationally [for instance Dehghani et al 2008].
13
Of course there is a lot of sensitivity when we talk about persuasion by a machine -‐ and rightly so. Actually it is quite peculiar that in our society there is not the same attention when we consider human produced persuasion, as for instance in politics or advertisement. From a pragmatic point of view, one can also observe how for instance the car industry developed without thinking of the environmental impact, and now painfully has to come to terms with this fact. Intelligent persuasive technologies should be aware of those considerations. Therefore it is good to take the initiative and being at the forefront of research aiming at endowing computers of an ethical sense [initial work in this direction in Stock and Guerini in press].
In the long run we aim at building a persuasive system able to decide the morality of a persuasion action of its own in a given context.
14
OPERATIONS
1 A. COFFEE MACHINE IN FBK
The importance of internal communication is today widely recognized [Dawkins, 2004]: in particular for the knowledge and creative industry, an understanding by the employees of the mission and whole culture of the enterprise is crucial to improve job satisfaction. Most often, this kind of knowledge though is not communicated by the traditional internal communication means (newsletter, internal TV, but also more recently web sites, corporate blogs etc.) but rather in the context of informal gatherings by spontaneous aggregation of communities of practices [Wegner, 1998; Wegner et al. 2002] -‐ groups of people that share the same competencies or backgrounds but without formal work relations. In this project, we will merge a newsticker media (a public display presenting information related to the internal life of the company) placed in a place of informal gathering (like the coffee machine) together with persuasive technology to affect group behavior. The aim of the system is to put in place persuasive communication strategies to maximize the information shared among the individuals around the place: different strategies will consider how much the individuals are connected (for example by looking at the patterns of their emails) and how they are interacting the contextual situation (who is talking to whom).
15
One attractive aspect of this scenario is the possibility of experimenting in FBK which can become a sort of living lab for possible future involvement of companies interested in the experimentation or in exploitation. Of course, we expect that the core technology will be portable in several domains. The system will be based and it will progress on the results of the café table experience e (Zancanaro et al 2011). The new scenario will be possibly more attractive for future business opportunities and it will be easier to organized long-‐term evaluations; also, it is recurrent, people tend go back every day to the same site, often in the same group configuration. The system can exploit that and intervene taking into account the progress of the group interaction.
MILESTONES FOR THE FIRST YEAR
During the first year of the project, an initial prototype will be developed starting from the basic components already developed for the café table prototype. The museum cafeteria scenario will be considered as a joint endeavor with a project together with MUSE.
• M2: initial design of the concept : specification of sensors architecture (including visual and acoustical analysis of the environment) and definition of the main communicative strategies to be implemented; • M6: integration of the newsticker architecture and content • M6: integration of sensors and first prototype; • M7 – M10: data collection for sensing and behavior analysis; implementation of communicative strategies. • M11: first prototype developed and tested in FBK for the first year
1 B. MOBILE SMALL GROUP-‐ORIENTED PERSUASION
Narrative Persuasion is an alternate method for persuasion that employs emotional techniques for convincing people as opposed to traditional rational approaches such as structured argumentation [Strothmann, 2009]. It uses the characters and actions in brief stories [de Graaf et al., 2011] to indirectly present arguments in a familiar way to readers or listeners, and by forging identity relationships between the user and the drama's characters. Empathetic reaction to those characters and the situations they find themsleves in are believed by communication and social scientists to both augment the impact of a persuasive message and suppress adverse reactions [Slater and Rouner, 2002].
In this project we will develop a mobile application for small groups, based on behavior assessment and co-‐ordinated system actions across multiple devices, where persuasive content is delivered in the form of short narrated dramas [Yale, 2011]. We shall start from
16
the work we developed for mobile co-‐ordinated dramatic presentations for small groups of museum visitors (see for instance [Callaway et al. 2011]). This work is inspired by the concept of mobile urban drama, and introduces sophisticated technology based on smartphones and sensors embedded in the environment.
In recent activity we have begun experimenting with flexible techniques that induce a desired behavior in the group during a museum visit. We intend to develop this concept into something more flexible, potentially in the MUSE setting, in tourism (e.g. a location-‐based recommendation system), or in a socially oriented scenario, for instance with the goal of having people discuss a social situation (a small group visit to a city quarter) or in a commercial scenario (e.g. small group shopping).
For instance in the city quarter scenario, a small group of friends would visit a part of the city of Trento different from their own, and a persuasive audio drama would be presented piece by piece as they walked together within that area. Each would receive a slightly different narrative about current societal issues in that area, and they would be invited by characters in the drama to compare those issues with current issues where they live (e.g., trash disposal, traffic, or school safety). This type of scenario would involve a wide-‐array localization system such as GPS, and with 5 separate dramas would last around 30 minutes.
In the tourism scenario, a small group of friends would walk around Trento's historic city center. Upon reaching a point of interest, they would hear a drama that is written around both historic and modern themes of the place where they are currently standing. Each person would also hear a second part that was different than what the others heard, e.g. one might hear about options for eating lunch, another about nearby shopping, and a third about a famous person who had lived nearby. This would merge interesting touristic information with more pragmatic “what do we do now” questions, providing both useful and thought-‐provoking themes for group discussion.
Each of these scenarios would require research in the following areas: intelligent adaptivity for persuasion via dynamic presentations, context, user and group modelling especially over a series of interactions, behavior analysis and decision-‐making from streaming group-‐based sensor data, utilizing information from mobile versus fixed sensors, and seamless movement between sensor environments (e.g. from a building into a city street).
It is meant that the mobile persuasion part will potentially involve Telecom Italia and its project WantEAT.
MILESTONES
During the first year of the project, an initial prototype will be developed starting from the basic components already developed for the group smartphone prototype.
M3: Initial design of the concept M6: Completion of mobile corpus acquisition M12: Completion of initial prototype
17
2A. INDIVIDUAL ORIENTED PERSUASION BASED ON CREATIVE TEXT VARIATION
This work will include:
a) Techniques that rely on extraction and modification (e.g. affective loading), for advertising concepts, products, museum exhibits and so on. The rationale is given by the assumption that emotionally loaded messages get the focus on the object more easily than un-‐loaded ones. Persuasive messages have the ability to foster the attention of the recipient, and possibly, the level of his/her emotional arousal. In this sense attention is influenced by the positive (or negative) attitudes toward the object being advertised. Starting point for this work is Valentino, a tool for valence shifting of natural language texts [Guerini et al, 2011]. Valentino can modify existing textual expressions towards more positively or negatively valenced versions.
b) Automatic production of creative variations of familiar expressions (such as proverbs, movie titles, news headlines, etc…), possibly humorous and taking into account the affective content of the produced text. The rationale is based on playing with existing linguistic material and producing novel expressions that catch the attention of the audience, guarantee memorization and of course hint to the product or concept that the message wants to promote. We want our approach to be adaptive, so it is fundamentally not based on lexical or other knowledge bases. Rather, it is corpus-‐based and exploits basic learning techniques at various levels.
c) personalization and contextualization, for the individual in a given situation.
The expected result is a system for contextual individual oriented persuasion, based on variation of familiar expressions.
The main steps required are:
1) introduction of a sophisticated system for deciding lexical variations, learning automatically from a large corpus;
2) some level of functional and rhetorical analysis
3) evaluation of techniques (incremental);
4) introduction of a user personality and context model so to adapt the system to the user situation and taste;
5) an implementation in a mobile setting (to be decided)
6) evaluation.
18
Evaluation experiments and persuasion metrics. In close interaction with the previous point, evaluation experiments and methodologies must be defined in order to assess the effects of persuasive messages on users. We want to focus on novel “cheap and fast” approaches for evaluating: a. Output quality. In case a persuasive message is automatically produce the first step is the evaluation of the output quality (e.g. grammatical and affective consistency). We plan to use Mechanical Turk, to test different NLP metrics on different specialized datasets (to be built) to evaluate Valentino output quality. b. Persuasive impact. We want to refine the methodology presented in [Guerini et al 2010] by conducting a large scale campaign on persuasive messages. Some of these messages will be Valentino outputs. The output will be (i) two experimental design models for evaluating text quality and text impact (ii) a series of evaluation experiments to test the feasibility of the approaches and to validate Valentino output.
MILESTONES
M 6 system for lexical variations
M 16 system with sentence level variations
M 12 conclusion of AdWords experiments
M 12 definition of applied scenario
M 18 user and context model
M 21 conclusion of Woz studies for adaptive scenario
M 27 mobile implementation
M 30 conclusion of evaluation
2B. DEFENSE FROM INAPPROPRIATE PERSUASION ATTEMPTS
This part aims at providing tools for defense from inappropriate (hidden) persuasive linguistic communication, mainly in broadcast messages, and in social networks. “Viral” messages have become a very important factor of persuasion. Differently from most current approaches, which tend to focus only on the social aspect, here the focus is on the investigation of language impact in social networks, and in other forms of potentially persuasive communication. The main themes are:
19
1. Predictability of hidden persuasive and virality effects that texts can have. E.g. predicting phenomena such as: white and black buzz, controversial issues, raising discussion.
Results will be:
a A model of several persuasive phenomena included under the generic term of “virality”. b A collection of resources . c Algorithms to be used in user defense against subtle persuasive messages (e.g. a prototype that makes the user aware of the persuasive dynamics underlying the content being visualized). d Algorithms to be used in applied scenarios such as Buzz prediction. In our view the prediction of viral phenomena can help Marketing Companies, by improving efficiency and reaction times in their activity such as, for example, Brand Protection. Up to now, only Buzz monitoring activities are implemented; they can just recognize viral phenomena that already took place (e.g. identifying positive and negative customer comments), rather than predict them. 2. Political Communication analysis. Work will start from the exploitation of CORPS (a CORpus of tagged Political Speeches) and integrate it with a number of analysis tools, in line with the previous 1, so to yield tools for detecting techniques adopted for subtly and possibly inappropriately influencing the audience in political communication. Analysis of online comments on socio-‐political decision making and prediction of their persuasive impact will be investigated as well. MILESTONES M12 Metrics definition for virality and hidden persuasion
M24 Experimentation results of detection and prediction techniques
20
3A. NEW NARRATIVE NEGOTIATION AND RECONCILIATION TABLE
A technology-‐based negotiation and collaboration environment will provide the means for working together and foster the mutual understanding and eventually a positive attitude. This technology will support two small groups from the two sides of the conflict in working out the task (multimedia production) and overcoming their disagreement in a constructive way.
It will mainly consist of a shared interface for working together both in a co-‐located setting (for the intra-‐group communication) and remotely (for inter-‐group communication). The two teams work together on the same interface while the interface is divided in the two sides of a “remotely” connected table. The two physical tables located in different places (and connected through an internet connection) are programmed to act as a single table. The setting will allow visual objects to be moved from one table to the other by pushing them out of the border of one table while appearing from the border on the other table. This possibility, together with a video-‐communication link, will allow the two team to have face to face meetings even if remotely placed.
Expected results:
Social impact: we expect the project to lead to a real system to be used jointly for instance by Palestinian and Jewish Israeli schools, or in other situations of informal learning and to contribute to an attitude change. We intend the project to be very visible and to show how technology can help in a situation of conflict.
As for the direct results of the use of the technology, high quality and original videos will be produced and published (YouTube, film festivals etc.) that were created by two groups of people with different narratives and point of views. Material from the sessions will be at the basis of novel social research.
MILESTONES:
April 2012: initial design and detailed workplan
November 2012: design and implementation of the graphical interface and the backend system to edit and store videos (release 0)
December 2012-‐June 2013: user studies in Israel
21
4A. COMPUTATIONAL ETHICS
We propose a cognitive approach, specific for the case of persuasion. In the persuasion case the situation is more complex than in the general case of assessing the moral status of an action, because of the involvement of two agents and two different acts (if we mean persuasion to act) -‐ one communicative act by the persuader and one act by the persuadee.
A series of experiments will be conducted with humans, to assess the natural ethical attitude when persuasion is concerned. The goal is of determining the appropriate ethical behavior of an autonomous persuasive system, so that it can take decisions in different situations.
Following the experimental studies a model will then be established as a reference. The model will serve as a reference for subsequently implementing a prototype of a system able to take decisions on a course of action on the basis of moral considerations.
MILESTONES for first 18 months
M 10 results of experiments submitted for publication
M 18 Model definition
22
REFERENCES
Abbasi, A. and Chen, H. Applying authorship analysis to extremist-‐group Web forum messages Intelligent Systems, IEEE 20(5), September 2005
Callaway, C. , Stock, O., Dekoven, E., Noy, K. , Citron, Y. & Dobrin, Y. Mobile Drama in an Instrumented Museum: Inducing Group Conversation via Coordinated Narratives. Proceedings of IUI-‐2011, International Conference on Intelligent User Interfaces, Palo Alto, 2011
Dawkins, J. Corporate responsibility: The communication challenge Journal of Communication Management, 2004
Dehghani M, Tomai E, Forbus K, Klenk M (2008) An integrated reasoning approach to moral decision-‐making. In: Proceedings of the 23rd National Conference on Artificial Intelligence -‐ Volume 3, AAAI Press, pp 1280–1286
de Graaf, A. Hoeken, H., Sanders, J. and Beentjes, W. J. Identification as a Mechanism of Narrative Persuasion. Communication Research, 2011
Dwyer S, Huebner B, Hauser MD The linguistic analogy: motivations, results, and speculations. Topics in Cognitive Science 2(3, 2010 Eisikovits, Z. and Winstok , Z. Reconstructing intimate violence: the structure and content of recollections of violence events. Qualitative Health Research, 12, 2002 Fogg, B. J. Persuasive Technology: Using Computers to Change What We Think and Do. Morgan Kaufman Publishers, 2002.
Fogg, B. J., & Eckles, D. (Eds.). Mobile Persuasion: 20 Perspectives on the Future of Behavior Change. Stanford, California: Stanford Captology Media, 2007.
Giora, R. On Our Mind: Salience, Context and Figurative Language. Oxford University Press, New York, 2003.
Guerini, M.,Strapparava, C. & Stock, O. Evaluation Metrics for Persuasive NLP with Google AdWords. Proceedings of LREC-‐2010 Seventh Language Resources and Evaluation Conference, Malta, 2010
Guerini M., Stock O., Zancanaro M., O'Keefe D.J., Mazzotta I., de Rosis F., Poggi I., Lim M. Y. & Aylett R. Approaches to Verbal Persuasion in Intelligent User Interfaces. In P. Petta, R. Cowie and C. Pelachaud (eds.) The HUMAINE Handbook on Emotion-‐Oriented Systems Technologies. Springer 2011.
Guerini M., Strapparava C. & Stock O. Slanting Existing Text with Valentino. in Proceedings of IUI 2011
23
Koppel, M. Schler, J. and Argamon. S. Computational Methods in Authorship Attribution. Journal of the American Society for Information Science and Technology, Volume 60, Issue 1, pages 9-‐26, 2009
Leinhardt, G., and Knutson, K. Listening in on Museum Conversations. Altamira Press, 2004.
Mason, W. and Suri, S. Conducting behavioral research on Amazon’s Mechanical Turk. Behavioral Research Methods, June, 2011
Matthews, T, Dey, A. K., Mankoff, J., Carter, S. and Rattenbury, T. A toolkit for managing user attention in peripheral displays. In Proceedings of the 17th annual ACM symposium on User interface software and technology (UIST). Santa Fe, NM, 2004
Nichols S, Mallon R Moral dilemmas and moral rules. Cognition 100(3), 2006
Perry, S. Jenzowsky, S. , King, C., Yi, H., Hester, J. and Gartenschlaeger, J. Humorous programs as a vehicle of humorous commercials. Journal of Communication, 47(1):20–39, 1997
Petty, R. E., and Cacioppo, J. T.. Communication and persuasion: Central and peripheral routes to attitude change. New York: Springer, 1986.
Ruch, W. Computers with a personality? Lessons to be learned from studies of the psychology of humor, Proc. the April Fools’ Day Workshop on Computational Humour, Trento 2002.
Sculley, D., Otey M., Pohl, M., Pitznagel, B., Hainsworth, J. and Zhou Y. Detecting Adversarial Advertisements in the Wild. Proceedings of KDD ’11, 17th ACM SIGKDD Int. Conf. on Knowledge Discovery and Data Mining, ACM, New York, 2011
Stock, O. & Guerini, M. Investigating Ethical Issues for Persuasive Systems In F. Paglieri, L. Tummolini, R. Falcone & M. Miceli (Eds.), The goals of cognition. Essays in honor of Cristiano Castelfranchi. College Publications, London, in press.
Stock, O. and Strapparava, C. Getting serious about the development of computational humour, Proc. 8th Int. Joint Conf. Artificial Intelligence (IJCAI-‐03), Acapulco, 2003.
Strothmann, S. Narrative Persuasion: How Emotional Appeal of Narratives and Source Characteristics Influence Belief Change. Master's Thesis, University of Maastricht, 2009.
Testa, A. La parola immaginata. Pratiche Editrice, 2000.
Toncar. M. The use of humour in television advertising: revisiting the US-‐UK comparison. International Journal of Advertising, 20(4):521–540, 2001.
Valitutti, A., Strapparava, C. and Stock, O. Textual Affect Sensing for Computational Advertising Proceedings of AAAI Spring Symposium on Creative Intelligent Systems, pp. 117-‐122, Stanford, 2008
Weiser, M. and Brown, J. S. ‗Designing Calm Technology, PowerGrid Journal, v1.01, July, 1996.
Wenger, E. Communities of Practice: Learning, Meaning, and Identity. Cambridge University Press. 1998
24
Wenger, E.; McDermott, R.; Snyder, W. M. Cultivating Communities of Practice. Harvard Business Press, 2002
Yale, R. The Influence of Narrative Believability on Juror Verdicts and Verdict Confidence: A Test of the Narrative Believability Scale. Robert Yale. National Communication Association Annual Convention, 2011
Zancanaro, M., Stock, O., Tomasini, D. & Pianesi, F. A Socially Aware Persuasive System for Supporting Conversations at the Museum Café. Proceedings of IUI-‐2011, International Conference on Intelligent User Interfaces, Palo Alto, 2011