Intelligent Sofware Applications Tutorial 1 Examples for intelligent applications: Introduction Artificial Intelligence Recommender systems, involving abstraction and ofen Trust learning. Trust, Confidence, Configuration systems, being able to plan new products. Familiarity Diagnostic systems, exhibiting reasoning capabilities. Explanations Spam filters , which ofen have to learn from experience. Context Tutorial 2 Usually involves delegation of responsibilities. Example “How do I know that this product recommendation is relevant?” “I don’t trust automatic bayesian spam filtering!” SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 14 / 96
Ambient Intelligent Systems Definition Tutorial 1 Introduction At the core of an ambient intelligent system lies the ability to Artificial Intelligence Trust appreciate the system’s environment , be aware of persons in Trust, this environment, and respond intelligently to their needs Confidence, Familiarity (Ducatel et al. (2001), ISTAG Scenarios for AmI in 2010 ). Explanations Context Tutorial 2 Perception: The initial act of perceiving the world that the system inhabits Context Awareness: Being aware of the environment and reasoning about ongoing situations Context Sensitivity: Exhibit appropriate behaviour in ongoing situations Action: Changing the environment according to context SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 15 / 96
Tutorial 1 Introduction Artificial Intelligence Trust Trust, Confidence, Familiarity Trust Explanations Context Tutorial 2 SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 16 / 96
What is Trust? Example Tutorial 1 “Trust is not a new research topic in computer science, spanning Introduction Artificial Intelligence areas as diverse as security and access control in computer Trust networks, reliability in distributed systems, game theory and Trust, Confidence, agent systems, and policies for decision making under Familiarity uncertainty. The concept of trust in these different communities Explanations varies in how it is represented, computed, and used.” (Artz and Context Gil, 2007) Tutorial 2 Trust can be something to be avoided (in security). Can be a desirable feature (electronic voting). Can be computationally modeled (multi agent systems). Can be understood as a mental attitude cognitive agents have towards each other (Falcone and Castelfranchi, 2001). It can focus on different aspects in cognitive agents. SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 17 / 96
Aspects of Trust Let us look at different aspects of trust Tutorial 1 Introduction Accordance with Mental Models Artificial Intelligence Trust Relying on past performance Trust, Confidence, Providing explanations for (changed) behavior Familiarity Explanations Context Tutorial 2 ☞ xkcd 364: responsible behavior SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 18 / 96
Mental Models Tutorial 1 Example Introduction Artificial Intelligence “This trust comes from an ability to predict the system’s Trust Trust, behavior through observation. To predict and explain the Confidence, Familiarity behavior of a system, people construct mental models that may Explanations be more or less complete and accurate. Therefore, designers Context must create intelligent applications that enable the formation of Tutorial 2 mental models that are predictable enough to merit their trust.” (Tullio et al., 2007) The authors strengthen the role of mental models. They also highlight the fact that the user should be able to explain and predict the system’s behaviour. SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 19 / 96
Change Over Time, Explanation Example Tutorial 1 “Results [...] indicate that trust is an important factor in Introduction Artificial Intelligence understanding automation reliance decisions. Participants Trust initially considered the automated decision aid trustworthy and Trust, Confidence, reliable. Afer observing the automated aid make errors, Familiarity participants distrusted even reliable aids, unless an explanation Explanations Context was provided regarding why the aid might err. Knowing why the Tutorial 2 aid might err increased trust in the decision aid and increased automation reliance, even when the trust was unwarranted.” (Dzindolet et al., 2003) The author point out that initial positive trust can decrease and increase based on performance. They indicate that explanations can help building trust. Trusted systems are less disused or misused. SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 20 / 96
Experience Tutorial 1 Example Introduction Artificial Intelligence “[...] it can be shown that positive experiences can be identified Trust Trust, that (usually) have an increasing or at least nondecreasing Confidence, Familiarity effect on trust, and negative experiences that have a decreasing Explanations or at least non-increasing effect. Here it appears easier to Context destroy trust than to build trust: the designed negative Tutorial 2 experiences show a stronger negative effect on trust than the positive effect shown by the designed positive experiences.” (Jonker et al., 2004) The role of experience with a system is highlighted. Flexibility over time is discussed, as are countermeasures. SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 21 / 96
Trust: Typology McKnight and Chervany (2001) develop a typology of trust Tutorial 1 based on literature survey and identify core characteristics: Introduction Artificial Intelligence benevolence, integrity, competence, and predictability. Trust “ Benevolence means caring and being motivated to act in Trust, Confidence, one’s interest rather than acting opportunistically. Familiarity Integrity means making good faith agreements, telling the Explanations truth, and fulfilling promises. Context Competence means having the ability or power to do for Tutorial 2 one what one needs done. Predictability means trustee actions (good or bad) that are consistent enough to be forecasted in a given situation.” They also organise trust by conceptual type, “such as attitude, intention, belief, expectancy, behavior, disposition, and institutional/structural.” SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 22 / 96
Trust: Typology McKnight and Chervany (2001) develop a typology of trust Tutorial 1 based on literature survey and identify core characteristics: Introduction Artificial Intelligence benevolence, integrity, competence, and predictability. Trust “ Benevolence means caring and being motivated to act in Trust, Confidence, one’s interest rather than acting opportunistically. Familiarity Integrity means making good faith agreements, telling the Explanations truth, and fulfilling promises. Context Competence means having the ability or power to do for Tutorial 2 one what one needs done. Predictability means trustee actions (good or bad) that are consistent enough to be forecasted in a given situation.” They also organise trust by conceptual type, “such as attitude, intention, belief, expectancy, behavior, disposition, and institutional/structural.” There seem to be huge disagreement about what trust is! SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 22 / 96
Tutorial 1 Introduction Trust, Confidence, Familiarity Overview Examples Trust, Confidence, Familiarity Explanations Context Tutorial 2 SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 23 / 96
Tutorial 1 Introduction Trust, Confidence, Familiarity Overview Examples Overview Explanations Context Tutorial 2 SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 24 / 96
Problems with the Definitions Partly problems between disciplines are to blame: Tutorial 1 Introduction Example Trust, Confidence, “A disciplinary lens sheds significant light on a topic like trust, Familiarity Overview but can also blind the researcher to possibilities outside the Examples paradigm the discipline pursues. Based on the differences Explanations among their definitions of trust, it appears that psychologists Context analyzed the personality side, sociologists interviewed the Tutorial 2 social structural side, and economists calculated the rational choice [...]. Few researchers [...] have developed trust typologies that define a set of trust constructs, and fewer still [...] have tried to reconcile interdisciplinary sets of constructs. More typically, trust typologies have stubbornly retained an intra-disciplinary flavor.” McKnight and Chervany (2001) SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 25 / 96
Problems with the Definitions Partly problems between disciplines are to blame: Tutorial 1 Introduction Example Trust, Confidence, “A disciplinary lens sheds significant light on a topic like trust, Familiarity Overview but can also blind the researcher to possibilities outside the Examples paradigm the discipline pursues. Based on the differences Explanations among their definitions of trust, it appears that psychologists Context analyzed the personality side, sociologists interviewed the Tutorial 2 social structural side, and economists calculated the rational choice [...]. Few researchers [...] have developed trust typologies that define a set of trust constructs, and fewer still [...] have tried to reconcile interdisciplinary sets of constructs. More typically, trust typologies have stubbornly retained an intra-disciplinary flavor.” McKnight and Chervany (2001) But there might also be an epistemic problem . SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 25 / 96
Danger vs. Risk Let’s step back a bit and look at some basic properties, as Tutorial 1 defined by sociologist Niklas Luhmann. Introduction Trust, He looks at the risk or dangers (of not reaching a goal) Confidence, Familiarity involved when taking certain decisions: Overview Examples Explanations Definition Context “[...] uncertainty exists in relation to future loss. There are then Tutorial 2 two possibilities. The potential loss is either regarded as a consequence of the decision, that is to say, it is attributed to the decision. We then speak of risk [Risiko] – to be more exact of the risk of decision. Or the possible loss is considered to have been caused externally, that is to say, it is attributed to the environment. In this case we speak of danger [Gefahr].” (Luhmann, 1993) SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 26 / 96
Choice and Alternatives Definition Tutorial 1 “...an attribution can be made to a decision only if a choice Introduction between alternatives is conceivable and appears to be Trust, Confidence, reasonable, regardless of whether the decision maker has, in Familiarity Overview any individual instance, perceived the risk and the alternative, Examples or whether he has overlooked them.” (Luhmann, 1993) Explanations Context Tutorial 2 Luhmann thinks it is essential for regarding something as a risk that there are alternatives to be considered, whether considered in practice or not. If a user chooses to use a system, he deliberately takes the risk of failure. Using the system is the result of an (potential) analysis . If he is bound to use it, he has the object of danger. Using the system is grounded in habit . SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 27 / 96
Familiarity and Trust Luhmann (1979) distinguishes several types of trust relations. Tutorial 1 Introduction First of all, he distinguishes between familiarity Trust, [Vertrautheit] and trust [Vertrauen]: Confidence, Familiarity Overview Definition Examples Explanations “ Familiarity reduces complexity by an orientation towards the Context past. Things that we see as familiar, because ‘it has always been Tutorial 2 like that’, are accepted – we do engage in relations with those – and things that we see as unfamiliar are rejected – we do not engage in relations with those.” Pieters (2008) For example, especially elderly people ofen refuse to use ATM’s, precisely because they are not used to them. Trust, on the contrary, has an orientation towards the future: it involves expectations. We trust in something because we expect something. SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 28 / 96
Trust and Confidence Luhmann (1988) also distinguishes trust [Vertrauen] and Tutorial 1 confidence [Zutrauen]. Introduction Trust, Both involve expectations with respect to future events. Confidence, Familiarity Overview Examples Definition Explanations “According to Luhmann, trust is always based on assessment of Context risks, and a decision whether or not to accept those. Confidence Tutorial 2 differs from trust in the sense that it does not presuppose a situation of risk. Confidence, instead, neglects the possibility of disappointment, not only because this case is rare, but also because there is not really a choice. This is a situation of danger, not risk.” Pieters (2008) Only when we chose to use a system, we talk about trust. SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 29 / 96
Tutorial 1 Introduction Trust, Confidence, Familiarity Overview Examples Examples Explanations Context Tutorial 2 SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 30 / 96
Familiarity Tutorial 1 Introduction We can use this distinction to clear the muddy waters around Trust, Confidence, different definitions of trust prevailing in the computer science Familiarity literature: Overview Examples Example Explanations Context “[Trust is] a subjective expectation an agent has about another’s Tutorial 2 future behavior based on the history of their encounters.” (Mui et al., 2002) We can classify this as being familiar with a system. SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 31 / 96
Confidence Tutorial 1 Introduction Trust, Confidence, Familiarity Example Overview Examples “[Trust is] the firm belief in the competence of an entity to act Explanations dependably, securely, and reliably within a specified context.” Context (Grandison and Sloman, 2000) Tutorial 2 We can classify this as having confidence in a system. SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 32 / 96
Trust Tutorial 1 Example Introduction Trust, “Trust of a party A to a party B for a service X is the measurable Confidence, Familiarity belief of A in that B behaves dependably for a specified period Overview Examples within a specified context (in relation to service X).” (Olmedilla Explanations et al., 2005) Context Tutorial 2 We can classify this as having trust towards a system. SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 33 / 96
Trust Tutorial 1 Example Introduction Trust, “Trust of a party A to a party B for a service X is the measurable Confidence, Familiarity belief of A in that B behaves dependably for a specified period Overview Examples within a specified context (in relation to service X).” (Olmedilla Explanations et al., 2005) Context Tutorial 2 We can classify this as having trust towards a system. In the following, we will look at two aspects that can increase or decrease trust or confidence into the systems: The aspect of explanations, and the aspect of contextuality. SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 33 / 96
Tutorial 1 Introduction Trust, Confidence, Familiarity Explanations Explanations Overview Black Boxing Possibility of Failure Context Tutorial 2 SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 34 / 96
Tutorial 1 Introduction Trust, Confidence, Familiarity Explanations Overview Overview Black Boxing Possibility of Failure Context Tutorial 2 SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 35 / 96
Why bother to explain? Tutorial 1 Introduction Important vehicle to convey information between Trust, communicating people in everyday human to human Confidence, Familiarity interaction. Explanations Enhance the knowledge of the participants in such a way Overview Black Boxing that they accept certain statements and gain a better Possibility of Failure Context understanding of the actions of the other persons involved Tutorial 2 and their motivations. They understand more, allowing them to make better informed decisions themselves. Explanations are the most common method used by humans to support their decision making (Schank, 1986). SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 36 / 96
Use of Explanations Tutorial 1 Introduction Trust, Confidence, If we cannot follow a conversation, Familiarity we ask our conversation partner about concepts that we Explanations did not understand, Overview Black Boxing we request justifications for some fact or we ask for the Possibility of Failure cause of an event, Context we want to know about functions of concepts, Tutorial 2 we want to know about purposes of concepts, and we ask questions about his or her behaviour and how he or she reached a conclusion. SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 37 / 96
Explanations in Intelligent Systems Tutorial 1 System Centric View Introduction Explanation as part of the reasoning process itself. Trust, Confidence, Familiarity Example: a knowledge intensive case-based reasoning Explanations Overview system can use its domain knowledge to explain the Black Boxing Possibility of Failure absence or variation of feature values. Context Tutorial 2 User Centric View Giving explanations of the found solution, its application, or the reasoning process to the user . Example: in an engine failure diagnosis system, the user gets an explanation on why a particular case was matched. SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 38 / 96
Explanations for {Trust|Confidence} Tutorial 1 Introduction Systems being able to explain their behaviour and Trust, reasoning increase the user’s perception of the system’s Confidence, Familiarity competence and integrity . Explanations This in turns support building up trust and confidence Overview Black Boxing (McKnight and Chervany, 2001). Possibility of Failure Context Looking for a model describing the relation between Tutorial 2 explanation and {trust|confidence} as well as possible points of failure. Taking a actor network perspective: looking at the translation and delegations processes involving system and user as actors. SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 39 / 96
Tutorial 1 Introduction Trust, Confidence, Familiarity Explanations Black Boxing Overview Black Boxing Possibility of Failure Context Tutorial 2 SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 40 / 96
Black Boxes Tutorial 1 Pieters (2011) introduces the concept of black boxing with Introduction regard to explanations: Trust, Confidence, In different IT settings, the black box character of systems Familiarity lacking explanations is ofen mentioned. Explanations Overview This concept can mean very different things. Black Boxing Possibility of Failure In the common sense meaning, a black box is something Context that outputs something based on certain inputs, but that Tutorial 2 we do not know the inner workings of. In a more philosophical sense, a black box is something that has been “blackboxed”; a theory or technology of which the supporting network of actants has become invisible. (Latour, 1999) SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 41 / 96
Explanation Programs Latour associates the process of blackboxing with three other phenomena: translation, composition and Tutorial 1 delegation. Introduction Composition means that actants in a network form a Trust, Confidence, composite actant to which actions can be attributed. Familiarity Translation denotes that the “action program”, the Explanations intentions and possibilities for action, change when actants Overview Black Boxing join forces. A man plus a gun has different action Possibility of Failure possibilities than a man or a gun alone. Context Delegation is the the process in which parts of an action Tutorial 2 program are delegated to different actants. The responsibility of delivering hotel keys at the reception can be delegated to large pieces of metal. We “translate” these concepts to explanation and trust. Actants have an explanation program: when they are asked to explain something about a theory or system, they have certain intentions and possibilities for explaining in a certain way. SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 42 / 96
Explanation for {Trust|Confidence} Tutorial 1 Introduction Explanation may serve different purposes. Trust, It can either aim at acquiring confidence or at acquiring Confidence, Familiarity trust. Explanations Explanation-for-trust is contrasted with Overview Black Boxing explanation-for-confidence Possibility of Failure Context Tutorial 2 Definition Explanation-for-trust is explanation of how a system works: the black box of the system is opened. Explanation-for-confidence is explanation that makes the user feel comfortable in using the system: the black box is not opened. SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 43 / 96
Black Boxes and Trust Tutorial 1 A black box cannot acquire trust, but only confidence. Introduction Black boxes can explain things to their environment, but Trust, Confidence, only as an explanation-for-confidence . Familiarity Black boxes can be opened when trust is required instead Explanations Overview of confidence; this opening produces an Black Boxing Possibility of Failure explanation-for-trust of how the system or network does Context what it is supposed to do. Tutorial 2 It reveals part of the inner workings, thereby reveals part of the risks, and thereby trades confidence for (possible) trust. If the explanation program of the network around a technology is strong enough, the black box of the inner mechanisms of the technology itself may not need to be opened. SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 44 / 96
Tutorial 1 Introduction Trust, Confidence, Familiarity Explanations Possibility of Failure Overview Black Boxing Possibility of Failure Context Tutorial 2 SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 45 / 96
Failure to Create Trust Tutorial 1 A bad explanation-for-trust may fail to create trust. Introduction Trust, Too little detail does not explain-for- trust : it fails to open Confidence, the black box, by only providing superficial reasons. Familiarity Explanations For example, the spam filter is explained to be working Overview within acceptable limits because it has been tested. Black Boxing Possibility of Failure Such explanations may contribute to confidence, but fail Context when trust is required, because the black box is not being Tutorial 2 opened. Too much detail, on the contrary, does not explain -for-trust . It fails to make the system comprehensible, because the user is not capable of processing the information at this level of detail. SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 46 / 96
Failure to Create Confidence Tutorial 1 Introduction A too detailed explanation may fail to reach its goal, Trust, because it does not explain-for- confidence . Confidence, Familiarity It aims for trust instead of confidence, by opening the black Explanations box of the system. Overview For example, a system may provide a complete reasoning Black Boxing Possibility of Failure trace when only some indications are required by the user Context in order to provide her with confidence. Tutorial 2 In that case, it may even decrease confidence. On the other hand, too little detail will not explain -for-confidence . “Because I said so” might help to deal with unruly kids, but is not likely to increase their confidence. SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 47 / 96
Levels of Detail Tutorial 1 We can map levels of detail to different results of explanations: Introduction Trust, Confidence, Familiarity level of detail result Explanations too low explanation fails Overview low explanation-for-confidence, justification Black Boxing Possibility of Failure high explanation-for-trust, transparency Context too high explanation fails Tutorial 2 Please note that level of detail is a simplification ignoring the qualitative aspects (what kind of explanations are needed to open the black box, and are they different from those not opening it?). SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 48 / 96
Outlook Tutorial 1 If intelligent systems can reach a level of explanation that Introduction creates as much confidence in these systems as we have in Trust, Confidence, people, they may become increasingly blackboxed Familiarity phenomena in our society. Explanations Overview We know more about how they work than we know about Black Boxing Possibility of Failure how people work, because we designed intelligent systems Context ourselves. Tutorial 2 Even so, the need for knowing precisely how they work may become less pronounced. In a complex society, there is still a need for experts who trust the systems because they know about their inner working. SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 49 / 96
Challenges for Ambient Systems Tutorial 1 Introduction An ambient systems that works in accordance with the Trust, mental model of the user can probably remain blackboxed Confidence, Familiarity No surprises Explanations High weaviness Overview Black Boxing If something goes wrong, i.e. not according to user Possibility of Failure expectations, the system will fall out of ambience Context Tutorial 2 System becomes visible again as a technological system Unboxing is required, at lest to a degree Problem: How, if there are no explicit system outputs available? Do we always need a “Herczeg-display” or “Herczeg-voice”? SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 50 / 96
Tutorial 1 Introduction Trust, Confidence, Familiarity Explanations Context Context Overview Systemic-Functional Theory of Language SFL in Context Context and Explanations Abstract Concepts Tutorial 2 SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 51 / 96
Tutorial 1 Introduction Trust, Confidence, Familiarity Explanations Overview Context Overview Systemic-Functional Theory of Language SFL in Context Context and Explanations Abstract Concepts Tutorial 2 SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 52 / 96
Contextualisation for {Trust|Confidence} Tutorial 1 Introduction Trust, Contextually adequate behaviour increases the user’s Confidence, Familiarity perception of the system’s competence and predictability . Explanations This in turns supports building up trust and confidence Context Overview (McKnight and Chervany, 2001). Systemic-Functional Theory of Language Looking for a model describing contextually adequate SFL in Context Context and behaviour and possible points of failure. Explanations Abstract Concepts Taking a semiotic perspective: looking at the meaning Tutorial 2 making processes involving system and user as actors. SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 53 / 96
Semiotics Tutorial 1 Semiotics is the science of signs or the study of sign Introduction systems (Fawcett, 1992). Trust, Semiotics, or semeion, was originally peculiar to medicine, Confidence, Familiarity referring to inference on the basis of some outward Explanations manifestation of state (or sign) (Eco, 1984). Context Overview We can think of semiotics as a perspective, as a means of Systemic-Functional Theory of Language looking at anything from the point of view of how it SFL in Context Context and generates meaning (Halliday, 1992). Explanations Abstract Concepts Semiotics deals with understanding sense making Tutorial 2 processes and sense making systems. Interaction is a process of exchanging and interpreting signs, symbols referring to and standing for something else. The users of a computer system see their interaction with the system against this background. SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 54 / 96
Tutorial 1 Introduction Trust, Confidence, Familiarity Explanations Systemic-Functional Theory of Language Context Overview Systemic-Functional Theory of Language SFL in Context Context and Explanations Abstract Concepts Tutorial 2 SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 55 / 96
Systemic Functional Theory of Language Tutorial 1 Introduction Systemic Functional Linguistics (SFL) is a social semiotic Trust, Confidence, theory that sets out from the assumption that humans are Familiarity social beings that are inclined to interact (Halliday, 1978). Explanations In addition, Halliday states that human communication is Context Overview inherently multimodal. Systemic-Functional Theory of Language Halliday combines the strengths of the approaches of SFL in Context Context and Saussure (1966), Peirce (1904) and Voloshinov (1973) Explanations Abstract Concepts (Cassens and Wegener, 2008). Tutorial 2 Saussure: the tradition of relational thinking Pierce: the understanding of meaning across modalities Voloshinov: the insistence that the sign is social SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 56 / 96
SFL: Stratification Stratification: A stratified model of language systems including: Tutorial 1 Introduction Sound Systems – phonetics, phonology, gesture, pixels etc. Trust, Lexicogrammar – lexis/grammar; or wording and structure Confidence, Semantics – the meaning system Familiarity Context – culture and situation; elements of the social Explanations structure as they pertain to meaning Context Overview Systemic-Functional Theory of Language Example SFL in Context Context and Explanations Abstract Concepts Context: the situation we are in is a lecture Tutorial 2 Semantics: a lecturer standing in front of students and talking constitutes knowledge transfer Lexicogrammar: from the worked examples down to the sentences used Sound Systems: the phonemes said and gestures used SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 57 / 96
SFL: Register Register: Dialectic relation of system and instance Tutorial 1 System – at the level of context the culture Introduction Instance – at the level of context the situation that we are in Trust, Confidence, Register – dialectic relation Familiarity Abstraction of instances which typically share a similar Explanations structure Context Concretisation of parts of the system Overview Systemic-Functional Theory of Language SFL in Context Example Context and Explanations Abstract Concepts System: the computational or linguistic system Tutorial 2 Instance: the concrete situation Register: the instantiation/generalization that allows the system to work in different concrete situations This is a relation, not an entity SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 58 / 96
SFL: Metafunction Metafunction: What function do representations have: Ideational – structure, relation of linguistic elements Tutorial 1 Logical Introduction Experiential Trust, Confidence, Interpersonal – relation of actors Familiarity Textual – content of discourse Explanations Together, these concepts span a space of exploration and Context description Overview Systemic-Functional Theory of Language SFL in Context Example Context and Explanations Abstract Concepts Ideational – using the field of discourse Tutorial 2 what is it about? Interpersonal – using the tenor of discourse how do the actants interact? Textual – using the mode of discourse what is being said and how? SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 59 / 96
Dimensions of Language Tutorial 1 Introduction Trust, Confidence, Familiarity Explanations Context Overview Systemic-Functional Theory of Language SFL in Context Context and Explanations Abstract Concepts Tutorial 2 The dimensions of language – Halliday and Matthiessen (2004) SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 60 / 96
Language as ... Tutorial 1 Halliday uses a tripartite representation of language, which Introduction has language as system, language as behaviour and Trust, language as knowledge. Confidence, Familiarity Language as system encapsulates the abstract structure of Explanations language, regularised (though changeable) patternings. Context Language as behaviour looks at the activity of language. Overview Systemic-Functional Language as knowledge looks at the way in which we know Theory of Language SFL in Context language. Context and Explanations But we do not do these things independently; we do not Abstract Concepts Tutorial 2 know language as a set of abstract rules. When we try to build a device, it is behaviour and knowledge that we face, yet it is the seemingly inaccessible system that we need to encode. SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 61 / 96
Tutorial 1 Introduction Trust, Confidence, Familiarity Explanations SFL in Context Context Overview Systemic-Functional Theory of Language SFL in Context Context and Explanations Abstract Concepts Tutorial 2 SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 62 / 96
Field Tutorial 1 Introduction Definition Trust, Confidence, Familiarity “The FIELD OF DISCOURSE refers to what is happening, to the Explanations nature of the social action that is taking place: what is it that the Context participants are engaged in, in which the language figures as Overview Systemic-Functional some essential component?” (Halliday and Hasan, 1985) Theory of Language SFL in Context Context and Explanations We are talking about ideational aspects. Abstract Concepts Tutorial 2 What is the domain? What are the long term or short term goals? The experiential domain? What is the structure, what are the networks of interaction? SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 63 / 96
Tenor Tutorial 1 Definition Introduction Trust, “The TENOR OF DISCOURSE refers to who is taking part, to the Confidence, Familiarity nature of the participants, their status and roles: What kinds of Explanations role relationship obtain among the participants [...], both the Context types of speech role that they are taking on in the dialogue and Overview Systemic-Functional the whole cluster of socially significant relationships in which Theory of Language SFL in Context they are involved?” (Halliday and Hasan, 1985) Context and Explanations Abstract Concepts Tutorial 2 We are talking about interpersonal aspects. What is the power structure between actors involved? What is the agentive role? What is the competence of the actors? SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 64 / 96
Mode Tutorial 1 Definition Introduction Trust, “The MODE OF DISCOURSE refers to what part the language is Confidence, playing, what is it that the participants are expecting to do for Familiarity Explanations them in that situation: the symbolic organisation of the text, the Context status that it has, and its function in the context ...and also the Overview rhetorical mode, what is being achieved by the text in terms of Systemic-Functional Theory of Language SFL in Context such categories as persuasive, expository, didactic, and the Context and Explanations like.” (Halliday and Hasan, 1985) Abstract Concepts Tutorial 2 We are talking about textual aspects. What medium is used? What is the type of interaction (dialogic, monologic)? What is the rhetorical thrust? SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 65 / 96
Tutorial 1 Introduction Trust, Confidence, Familiarity Explanations Context and Explanations Context Overview Systemic-Functional Theory of Language SFL in Context Context and Explanations Abstract Concepts Tutorial 2 SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 66 / 96
Failure to Create {Trust|Confidence} The different actors being aligned in their perception of Tutorial 1 context will usually have an increasing or at least Introduction Trust, non-decreasing effect on trust and confidence. Confidence, Familiarity The different actors being misaligned in their perception of Explanations context will usually have an decreasing or at least Context non-increasing effect on trust and confidence. Overview Systemic-Functional Theory of Language SFL in Context Example Context and Explanations Abstract Concepts If the intelligent system misjudges the competence of the Tutorial 2 human user (misalignment in the TENOR ), it might adjust the rhetorical thrust (leading to a misaligned MODE ) and for example deliver an explanation-for-trust instead of an explanation-for-confidence, thereby risking to decrease confidence. SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 67 / 96
Challenges for Ambient Systems Depending on the type of system, Field, Tenor and Mode Tutorial 1 may not be easy to define Introduction Field Trust, Smart Meeting rooms have, given “normal” usage, a Confidence, Familiarity restricted set of activities to support Explanations Smart Homes may need to support a wide variety of Context activities Overview Tenor Systemic-Functional Theory of Language Ambient Systems in closed settings (universities, SFL in Context Context and companies) have to deal with a limited amount of different Explanations users Abstract Concepts Tutorial 2 Public systems can make very little assumptions about their users (walk up and use) Mode Available modes of discourse are themselves restrained by context (no Klingon opera aria in a lecture) Modality/Mediality/Codality might be limited, and therewith the expressiveness SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 68 / 96
Tutorial 1 Introduction Trust, Confidence, Familiarity Explanations Abstract Concepts Context Overview Systemic-Functional Theory of Language SFL in Context Context and Explanations Abstract Concepts Tutorial 2 SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 69 / 96
Abstract Concepts Tutorial 1 Introduction Definition Trust, Confidence, Abstract concepts: concepts which have no grounding in the Familiarity material setting of the activity Explanations Context Overview Value: abstraction, or the ability to create a more general Systemic-Functional Theory of Language category from a set of specifics, by whatever principle, is SFL in Context Context and one of the most useful mental tools that humans possess. Explanations Abstract Concepts Challenge: to function intelligently in context artifacts Tutorial 2 must be able to recognise and understand abstract concepts and respond appropriately but the meaning of such concepts is not grounded in the material setting. SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 70 / 96
Example Example Tutorial 1 Emergency in the hospital domain has meanings that are Introduction distinct from meanings in other domains. These might be: Trust, Confidence, Hospital specific meanings (cultural specific) Familiarity Explanations Activity specific meanings (situation specific) Context Overview Systemic-Functional Theory of Language Concrete: Having a direct material referent of place, using SFL in Context Context and the specific deictic (e.g. ‘the emergency department’) and Explanations Abstract Concepts having the potential to be used as a circumstance location Tutorial 2 spatial (e.g. ‘in the emergency department’). Abstract: Having no clear referent in the material setting but referring rather to a state, using the no specific deictic (e.g. ‘an emergency’) and having the potential to take the specific deictic in past tense (e.g. ‘the emergency’) (e.g. ‘the emergency this morning’). SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 71 / 96
Emergency Tutorial 1 Introduction Trust, Definition Confidence, Familiarity Emergency: a complex set of actions and relations that Explanations constitute an interruption to the normal flow of a social process. Context Overview Systemic-Functional Theory of Language Culture based: deriving from the function of the broader SFL in Context Context and hospital culture, or, Explanations Abstract Concepts Context based: deriving from variation within the Tutorial 2 structure of the social process itself. SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 72 / 96
Application: Culture-Based Tutorial 1 Introduction Example Trust, Confidence, Familiarity Culture based emergency (e.g. the doctor is called away from Explanations the ward round because of pressures from the wider hospital). Context Overview Systemic-Functional Response from artifact: provide new information Theory of Language SFL in Context Why: a culture based emergency constitutes a change in Context and Explanations Abstract Concepts context because the field (topic), tenor (relations) and Tutorial 2 possibly the mode (interactional features) have changed; this means that new information will be needed by the doctor. SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 73 / 96
Application: Context-Based Tutorial 1 Introduction Example Trust, Confidence, Context based emergency (e.g. the doctor is required to Familiarity Explanations resusitate a patient during ward round) Context Overview Systemic-Functional Response from artifact: be quiet and await query – Theory of Language SFL in Context alternant modes may be needed Context and Explanations Why: a context based emergency is a sequence shif and Abstract Concepts Tutorial 2 not a new context. There are only minor changes to the field, if any. This situation requires material action from the doctor but the device needs to be ready for queries. SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 74 / 96
Semiotic Profile Tutorial 1 If the system acts contextually appropriate, user Introduction confidence in the system can be increased Trust, Confidence, If the user understands why, it can also increase trust Familiarity How can we model such abstract concepts? Explanations Not having a material grounding does not mean that there Context Overview are no observable features Systemic-Functional Theory of Language In particular, contextual appropriate behaviour follows SFL in Context certain “scripts” Context and Explanations Diversion from these scripts can be a sign for a change in Abstract Concepts Tutorial 2 context that is due to abstract concepts Here: Semiotic profiles and Generic Structure Potential We can try to model abstract concepts as unexpected context shifs SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 75 / 96
Scripts Example Tutorial 1 Introduction “When you go to buy something in a convenience store you can Trust, Confidence, be reasonably certain of what’s going to happen in that Familiarity situation. First, you’ll walk in and you might say ‘hello’. Then Explanations you’ll ask for some batteries and then pay. We can guess this Context Overview sequence due to our previous experience with these kinds of Systemic-Functional Theory of Language situations and the fact that they are nearly always the same. SFL in Context Context and Some parts may change (you might not say hello) but you Explanations Abstract Concepts always have to pay.” (David Didau) Tutorial 2 Related (but not identical) concepts: Scripts (Silvan Tomkins, Roger Schank, Robert Abelson) Generative grammar (Noam Chomsky) Frame (Marvin Minsky) SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 76 / 96
Generic Structure Potential Tutorial 1 Introduction Within certain recurring sets of texts then, coherence of Trust, structure is formed through obligatory and optional Confidence, Familiarity elements, the totality of which forms the Generic Structure Explanations Potential (GSP) for that set (Halliday and Hasan, 1985) Context In other words, there are certain obligatory elements that Overview Systemic-Functional Theory of Language characterize the genre and other optional ones that add SFL in Context elaboration but are not necessary Context and Explanations Abstract Concepts There is thus a structure to social interactions Tutorial 2 We can call it potential because it has a predictive quality that allows us to navigate these social situations almost unconsciously SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 77 / 96
Semiotic Profile Tutorial 1 Introduction Trust, Confidence, Familiarity Explanations Context Overview Systemic-Functional Theory of Language SFL in Context Context and Explanations Abstract Concepts Tutorial 2 SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 78 / 96
Challenges for Ambient Systems Tutorial 1 Structural view: Introduction Conceptual descriptions of context parameters such as the Trust, notion of tenor, field and mode can help model parameters Confidence, Familiarity of concept important for the activities to be supported Explanations Conceptual models have to be transformed into computational models and filled with experiential data Context Overview to sense shifs in tenor, field or mode Systemic-Functional Theory of Language to act appropriately with regard to tenor, field or mode SFL in Context Context and Procedural view: Explanations Abstract Concepts Descriptions of unfolding activities like Generic Structure Tutorial 2 Potential can help model activities to be supported Conceptual models have to be transformed into computational models and filled with experiential data to “go along” with a possible instantiation to detect deviations that could indicate context changes SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 79 / 96
Tutorial 1 Introduction Trust, Confidence, Familiarity Explanations Tutorial 2 Context Tutorial 2 SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 80 / 96
Descriptive Framework Version 4 Tutorial 1 Introduction Contextualisation Trust, Confidence, Contextual Parameter Familiarity Environment – things, services, people Explanations Personal – mental & physical information about user Context Social – roles & relations Tutorial 2 Task – what is the user doing Spatio-Temporal – when & where are we Other Process of Contextualisation Awareness – what aspects are taken into account? Sensitivity – what aspects are changed? SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 81 / 96
Descriptive Framework Version 4 Tutorial 1 Introduction Intelligence Trust, Confidence, Familiarity System Intelligence Explanations Personalized – tailored to individual needs Context Adaptive – changing in response to user needs Tutorial 2 Anticipatory – can act on its own on user’s behalf Social Intelligence Socialized – compliant to social conventions Empathic – take user’s inner states into account Conscious – introspection, has inner state SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 82 / 96
Descriptive Framework Version 4 Tutorial 1 Ambience Introduction Trust, Perception Confidence, Familiarity Mediality – media types Codality – semantic representation Explanations Modality – human senses Context Tutorial 2 Reasoning Context Awareness Context Sensitivity Other Action Mediality – media types Codality – semantic representation Modality – human senses SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 83 / 96
Descriptive Framework Version 4 Interaction Tutorial 1 Introduction Implicit vs. Explicit Trust, Confidence, Implicit input – through behaviour not primarily aimed at Familiarity interacting with the computerised system (walking through Explanations a door, using a whiteboard...) Context Explicit input – primarily aimed at interacting with the Tutorial 2 computerised system (voice or gesture commands...) Explicit output – designed to get the users’ attention (voice output...) Implicit output – change of material setting where the users’ attention is not the primary goal (opening doors...) Emotion Does the system sense emotions? Does the system show emotions? SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 84 / 96
Descriptive Framework Version 4 Embeddedness Tutorial 1 Introduction Weaviness Trust, Confidence, Is the system woven into the background? Familiarity Is the interaction naturally/culturally sound? Explanations Enhancement Context Does the system enhance or replace current solutions? Tutorial 2 Current “technical” solutions – using (computerized) artefacts Current “non-technical” solutions – not using (computerized) artefacts Social Interaction Does the system enable/enhance social interaction amongst humans? Is the system targetting at supporting individual users? SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 85 / 96
Architecture Version 1 Tutorial 1 Introduction Trust, Confidence, Context Context Sensing Acting Familiarity Sensitivity Awareness Explanations Context Tutorial 2 Sensors Actuators World Sensable Actable World World General, simplified architecture SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 86 / 96
Assignment 4.6: Course Context Group Work Tutorial 1 Introduction Form groups of 3-6 Trust, Confidence, Position the required reading within the course context Familiarity Course roadmap given Explanations Which categories to use? Context Granularity Tutorial 2 Show links Between texts Between lectures Between texts and lectures Find a ways to show those links Iconic Hypermedia Present your idea in the course/learnweb SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 87 / 96
Assignment 4.6: Roadmap Group Work Tutorial 1 � Definitions & Theory Introduction Context, Ambient Intelligence Trust, Confidence, � Descriptive Framework & Examples Familiarity Facets, Architectures, Examples Explanations � Implementation & Evaluation Context Tutorial 2 Challenges, Prototyping, Deployment, Evaluation � Human & Computer Interaction, Privacy, Emotion Trust Explanations, Context � Computing & Culture Arts & Games � Specific Issues Uncertainty, Privacy-respecting technologies SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 88 / 96
Assignment 4.6: Required Reading I Required reading for week 1 Tutorial 1 Weiser, M. (1991). The computer for the 21st century . Introduction Scientific American, pages 94–104. Trust, Confidence, Required reading for week 2 Familiarity Aarts, E., R. Harwig, and M. Schuurmans. 2001. Ambient Explanations Intelligence. In The Invisible Future: The Seamless Context Integration of Technology into Everyday Life , ed. P. J. Tutorial 2 Denning, pp 235-250. New York: McGraw-Hill Companies. Required reading for week 3 Dourish, Paul, and Ken Anderson. Collective information practice: exploring privacy and security as social and cultural phenomena . Human-computer interaction 21.3 (2006): 319-342. Required reading for week 4 Dourish, Paul. “What we talk about when we talk about context.” Personal and ubiquitous computing 8, no. 1 (2004): 19-30. SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 89 / 96
Assignment 4.6: Required Reading II Required reading for week 5 Tutorial 1 Tom Geller: “How Do You Feel? Your Computer Knows.” Introduction Communications of the ACM Vol. 57(1), pp. 24-26. Jan. 2014 Trust, Rosalind W. Picard: “Affective Computing”. MIT Technical Confidence, Familiarity Reports – TR 321. Nov. 1995 Explanations Required reading for week 6 Context Davies, N., & Gellersen, H. W. (2002). “Beyond prototypes: Tutorial 2 Challenges in deploying ubiquitous systems.” IEEE Pervasive computing, 1(1), 26-35. Hansen, T. R., Bardram, J. E., & Soegaard, M. (2006). “Moving out of the lab: Deploying pervasive technologies in a hospital.” IEEE Pervasive Computing, 5(3), 24-31. Required reading for week 7 Abowd, Gregory D., Elizabeth D. Mynatt, and Tom Rodden. “The human experience” IEEE pervasive computing 1.1 (2002): 48-57. SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 90 / 96
Assignment 3.5: New Lab Room Group Work Tutorial 1 Introduction Form groups of 3-6 Trust, Confidence, Develop the outline of a project idea to change A120 into a Familiarity room you would like to use: Explanations Today, traditional computer lab Context How to change it? Tutorial 2 Interior decor Furniture Technology Possible technologies: Tab, Pads & Boards Behavioural interfaces Natural language processing Pitch your idea in the course SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 91 / 96
Assignment 3.5: Old Lab Room Tutorial 1 Introduction Trust, Confidence, Familiarity Explanations Context Tutorial 2 Samelsonplatz, A 120 SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 92 / 96
Assignment 3.5: Old Lab Room Measurements Tutorial 1 Introduction Trust, Confidence, Familiarity Explanations Context Tutorial 2 15 computer workstations (9+6) 16 group work seats (8+8) SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 93 / 96
Video 3.1: Universität 2025 Tutorial 1 Introduction Trust, Confidence, Familiarity Explanations Context Tutorial 2 ☞ Where VR in 2025 (6:45) SoSe 2018 Jörg Cassens – Building and Losing Trust in Ambient Intelligent Sofware Applications 94 / 96
Recommend
More recommend
