Extended Principles for Human Information Interaction Design_23S2

Principles for Human Information Interaction Design Category Design Principle Brief Description Examples (using a simple assignment bot) Informing Theory Human sensitivity Build rapport through interaction tone When in conversations with other people, we generally do not speak as if we are giving a lecture or talking to a brick wall. We want to connect with the person we are speaking with so that they understand and (often) agree with what we are saying. We do this by showing that we understand them and are interested in what they are saying. Doing this through the tone of an interaction means steering away from short 1 word responses or robotic-sounding speech. If a user asks when an assignment is due, to build rapport, you would have your bot say something like, “that assignment is due on DD/MM/YYY, it’s the reflective journal one, did you want to know what it involves?”, rather than just “DD/MM/YYY”. To increase the rapport building, you could also add some creative phrases of your own that you might notice people saying in their natural conversations, such as “that’s a great question”, or “it must be stressful having all of these assignments”, etc. Bennett (2018), Oinas-Kukkonen & Harjumaa (2009) Conform to social norms This will depend on who your target user is. There will be different cultural and social norms that apply depending on who your bot is designed to converse with. If your bot is designed to talk on behalf of a government organisation to the general public, you will need to build certain manners into its speech. For example, more formal, respectful language rather than colloquial, as the latter may be considered to be rude based on the norms associated with social interactions with government representatives. If, instead, your bot is designed to talk to young uni students, a more colloquial, fun tone could be adopted, as a very formal way of speaking may appear to be distant and unfriendly due to the norms that that social group is used to. Bennett (2018), Oinas-Kukkonen & Harjumaa (2009), Luger & Sellen (2016) Adapt to cognitive factors One way to describe cognition is how our brains process information. You need to be mindful of the nuances and shortcomings of the human information processor when presenting information to your users. You can think about cognitive biases, cognitive load and other restraints when designing your bot. For example, humans are not great at handling large amounts of information at once, so scaffolding information in a logical way is a good way to design an interaction (i.e. think about the information you retain when engaging in a discussion about a topic compared to listening to a one-sided lecture about the same content) Lieberman (1997), Gnewuch et al (2017) Personalise the information Build into your design an ability to customise the information it provides based on what your bot knows/learns about the user. In simpler systems, this could be done by deepening your understanding of your target user and customising the information to provide the best experience for them specifically. In more complex systems, this could mean asking questions or logging behaviour of a more general set of users to determine the best interactions for them. Let’s say you have a user who is very disorganised and never has a good grasp on what they have to do and when – and may be highly stressed because of this. (You determine this either by targeting them as a user type or ascertaining this by their previous interactions with the bot) You would then build in frequent reminders of due dates and assignment criteria into your bot, at intervals in the conversation where this extra information might be appropriate. You would also build your assignment bot to speak to this user in more calming tones, including reassurances and encouragements that they will be able to get everything done. These types of solutions would not be appropriate for a different type of user who is generally on top of everything they need to do and just needs information Oinas-Kukkonen & Harjumaa (2009)

quickly, as the extra encouragement and reminders would probably just annoy these users. In this way, it is personalised to a specific type of user. Utilise human expertise Despite the rise of Artificial Intelligence, your users are always going to be smarter overall than your chatbot. If you assume your users have little capability of knowledge, then your bot won’t work very well. Acknowledge that your users have at least some knowledge – either directly or indirectly related to your topic. You can then leverage that in the conversation to allow them to participate more with your bot rather than assuming your bot will be providing all the information in the interaction. If a user is asking about the due date for a reflection assignment, a bot can ask them – “did you already know what this assignment is about or did you want me to go into some detail?” It could also ask them: “Which assignment do you think is most important, and I’ll prioritise reminders for that one” Lieberman (1997), Gnewuch et al (2017) Communication fluency Minimise information complexity If your topic involves complex information, try to walk your users through it by breaking it down into less complex parts. If a user asks in week 2 when the next assignment is due, the bot might start explaining that there is a formative part of assignment 2 due in the next couple of weeks, but it doesn’t contribute towards your grade but you do have to submit it in time to get feedback…. That explanation could be a little confusing in the context of a chatbot conversation. Instead, what your bot could do is to tell the user that there is a formative feedback part for assignment 1 due on DD/MM/YYYY and ask whether they want to know what that formative part means for their marks. If they want to know this, they can ask, if they want to skip over this, they have the choice too. This not only enables your user to have control over the information that they receive, but also breaks the information into less complex parts. Oinas-Kukkonen & Harjumaa (2009), Luger & Sellen (2016) Maximise interaction predictability One of the main components of usability is predictability. We often run on ‘autopilot’, and if everything acts as we expect it to, we will be happy and won’t have to do much “thinking”. The best way to make an interaction with a chatbot predictable is to ensure it conforms to our expectations about how conversations usually go. Our expectations of how conversations usually go come from our experiences – which are primarily with humans. Therefore, to make an interaction predictable – our bots must act like humans. (note – acting like a human does not mean it has to be deceptive about the fact that it is a bot). This is generally going to be achieved through thinking deeply about each response your bot has and whether or not it feels naturally like a human response. For example, try to avoid things like “say “Assignment 1 Due” for the due date of assignment 1”, or “click for more information”. These sorts of phrases are not normally said in conversation and disrupt the flow of the interaction. In terms of more general predictability – think about consistency. If your bot responds with general information and the due date for assignment 1 if a user asks about assignment 1, but only responds with the due date if the user asks about assignment 2, then this is not consistent. The user will not be able to predict, for example, what type of response they will get if they ask about assignment 3 if the other two responses are inconsistent. Oinas-Kukkonen & Harjumaa (2009)

assignment, or anything about assignment 2 or 3, then this is inflexible and will frustrate users. Artificial agent credibility Create realistic expectations of capability Expectation management is useful in many domains, including in conversations (both human to human and human to bot). Part of enhancing predictability (mentioned in the above principle) is aligning with expectations. There are some expectations that you can’t control (e.g. social norms), but can be aware of, and there are others that you can manage yourself. The expectations that your users have about the capability of your bot is something that you can manage yourself. Your bot will only have a limited capability. For example, our imagined assignment bot will only be able to talk about assignments in this unit. Rather than present this bot as a go-to knowledge bot on all things about this unit (creating the unrealistic expectation that users can ask it questions about the details of workshops, classroom location, assignment results, etc. and leading to frustration when it can’t answer these things), you need to be up-front about what your bot can and can’t do. This way, users will know what they can and can’t talk to your bot about and won’t get annoyed when they encounter something your bot can’t respond to. Luger & Sellen (2016) Provide sufficient capability for information task Having said the above, you also need to ensure that you are able to meet the basic needs of your target users. If your bot is only able to answer questions about one assignment and not the rest, or only knows the due date for assignments 1 & 2, but the marking criteria to all 3, then this is unlikely to be sufficient to meet your users’ goals of knowing what they need to do for assignments and when they need to do it. Gnewuch et al (2017) Provide transparency and accountability Humans are generally quite good at being able to tell whether they are being deceived or lead on. Any technology you provide needs to have a level of transparency and accountability in order for users to trust it. Be honest with your users, and allow your bot to accept mistakes without trying to brush over them. For example, if your bot is unable to answer a question, or speak about a topic, make sure it provides a reason. Something like “I’m sorry – I don’t actually have the information about that”, rather than saying something like “that’s a silly question” or “that’s off topic, try something else”. Being accountable as the creator of your bot is also preferable, so you could add something like “but that’s an interesting question – I will give that feedback to my creator so that I can answer that question better in the future” Oinas-Kukkonen & Harjumaa (2009), Luger & Sellen (2016) Build on user provided information Similarly to utilising human expertise, ensure you integrate what the user says into the interaction. For example, if a user answers early in the conversation that they have an extension for an assignment, if they later ask about the due date for that assignment, the bot should integrate this earlier information into the due date without having to ask for it again. When this integration is done, it would also be wise to have your bot tell the user that this has happened in case they did not expect the due date to be changed to automatically reflect their extension, for example. This last part is an example of providing transparency. Oinas-Kukkonen & Harjumaa (2009) Guide the interaction to fulfil information task You are designing a conversation. You therefore have some control over how that conversation eventually plays out with your users. Thinking about guiding the conversation enables you to steer around potential gaps in your bot’s capability, as well as scaffold a meaningful, personalised conversation. If your target user group are students that are stressed about the unit and are just hoping to pass rather than aiming for a 7, your bot could be designed to focus only on the essential elements of the unit necessary to pass. It might point out places such as the Tier 1 stage for assignment 1 that, if passed, guarantees a pass for the final version of the assignment. Rather than simply waiting for the user to ask the right questions, the bot could begin by suggesting some quick tips to pass the unit in some sort of logical order (e.g. urgency or importance) and allowing the user to explore each in more detail. Oinas-Kukkonen & Harjumaa (2009), Luger & Sellen (2016)

References Bennett, G. A. (2018). Conversational Style: Beyond the Nuts and Bolts of Conversation. In Studies in Conversational UX Design (pp. 161-180). Springer, Cham. https://link.springer.com/chapter/10.1007/978-3-319-95579-7_8 Gnewuch, U., Morana, S., & Maedche, A. (2017). Towards designing cooperative and social conversational agents for customer service. ICIS 2017. http://aisel.aisnet.org/icis2017/HCI/Presentations/1 Lieberman, H. (1997, March). Autonomous interface agents. In CHI (Vol. 97, pp. 67-74). http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.30.8904&rep=rep1&type=pdf Luger, E., & Sellen, A. (2016, May). Like having a really bad PA: the gulf between user expectation and experience of conversational agents. In Proceedings of the 2016 CHI Conference on Human Factors in Computing Systems (pp. 5286-5297). ACM. http://edithlaw.ca/cs889/2018/reading/Asking/Paper2.pdf Oinas-Kukkonen, H., & Harjumaa, M. (2009). Persuasive systems design: Key issues, process model, and system features. Communications of the Association for Information Systems , 24 (1), 28. https://aisel.aisnet.org/cgi/viewcontent.cgi?article=3424&context=cais