AI Interaction Design

Get Started. It's Free
or sign up with your email address
AI Interaction Design by Mind Map: AI Interaction Design

1. Who is responsible for the personality of your Artificial Intelligence agent aka chatbot, when he/she interacts with your users?

1.1. the role of the design of the chatbot should NOT be given to other stakeholders, as the CTO, the CEO or any AI engineer

1.2. Actually it is highly constructive to have a fight/argument between the engineers and the AI interaction designer

1.3. the AI interaction designer will be requesting more complex features or advanced filtering

1.4. The engineers will tend to want to bring harmony to their code which is going to minimize their work and make it more presentable but usually the real world is much complex to justify this need of the engineers

1.5. This is equivalent as having an Architect Engineer and a Civil Engineer be the same person. Would you expect to have a good building at the end? Probably not, because if you need to design and implement (or do other things) at the same time you will most likely deliver an average at both

2. If​ ​you​ ​are​ ​building​ ​an​ ​AI​ ​that​ ​interacts​ ​directly​ ​with​ ​the​ ​users​ ​you​ ​will​ ​need​ ​someone​ ​to​ ​think​ ​through​ ​each interaction​ ​and​ ​figure​ ​out​ ​how​ ​the​ ​AI​ ​should​ ​respond​ ​to​ ​effectively​ ​do​ ​what​ ​it​ ​is​ ​designed​ ​to​ ​do.

2.1. Unfortunately one model / one AI / one personality is NOT enough

2.2. an AI interaction designer should read through the actual conversations, not all of them

2.2.1. pick the most interesting the length of the convesation is long conversations of a recurrent user

2.3. get inspiration of how the users interact with the AI

2.3.1. a user might like to play with the chatbot

2.3.2. will the chatbot have playful / unnecessary features?

2.3.3. or will the chatbot will give a default answer whenever the user gets out of context

2.4. construct stories and find the pain points

3. The​ ​reason​ ​of​ ​this​ ​initiative​ ​is​ ​to​ ​give​ ​a​ ​little​ ​bit​ ​of​ ​insight​ ​on​ ​this​ ​very​ ​new​ ​and​ ​emerging​ ​job​ ​which​ ​is​ ​called​ ​AI Interaction​ ​Designer​ ​and​ ​to​ ​offer​ ​my​ ​help​ ​by​ ​combining​ ​my​ ​past​ ​UX​ ​skills​ ​and​ ​my​ ​more​ ​recent​ ​AI​ ​skills.

3.1. The AI Interaction Designer will need to be good as a Researcher and building stories, colors are minimal and layouts do not exist

3.2. However there is still a journey map for the user

3.3. What we are calling red routes, the ones that satisfy the goal of the user are now the way the conversation flows

3.4. The AI Interaction Designer should know the architecture of the AI for the chatbot to have an overview of why this works like that and the other like this

3.5. In fact the AI Interaction Designer should be able to do some filtering himself and also be able to play with the hyperparameters and configurations to see how different chatbots can be constructed

4. "For​ ​example,​ ​a​ ​chatbot​ ​is​ ​expected​ ​to​ ​be​ ​really​ ​professional​ ​and​ ​friendly,​ ​polite,​ ​very​ ​detail​ ​oriented,​ ​firm​ ​when​ ​it needs​ ​to​ ​be.​ ​These​ ​are​ ​the​ ​things​ ​you’d​ ​want​ ​in​ ​a​ ​good​ ​human​ ​assistant" + "The​ ​role​ ​of​ ​AI​ ​Interaction​ ​Designer​ ​is​ ​to​ ​create​ ​a​ ​style​ ​guide​ ​for​ ​the​ ​personality​ ​of​ ​the​ ​AI"

4.1. Different people need to be treated differently

4.1.1. a human can adjust his/her attitude according to the person

4.1.2. Construct different Personas Some people will react positively if the chatbot has very serious attitude if it is a doctor for example Other people will like the chatbot to be friendly even if the chatbot play the role of a doctor Others will require as simple language as possible because their English are limited

4.2. People do not know hot to talk to icons

4.2.1. Your chatbot persona should be real, like having a photographs or multiple photographs

4.2.2. If the chatbot is a doctor then you should have a team of doctors

4.2.3. Even let the user choose which kind of doctor we would want to talk to

4.3. New users and Old/Recurrent users should be treated in a different manner

4.3.1. The conversation starter should be much different on new vs old users

5. That’s​ ​the​ ​challenge​ ​of​ ​a​ ​conversational​ ​UI.​ ​Users​ ​can​ ​provide​ ​an​ ​infinite​ ​number​ ​of​ ​responses​ ​with​ ​an​ ​infinite number​ ​of​ ​phrasings,​ ​and​ ​they​ ​expect​ ​the​ ​AI​ ​to​ ​know​ ​exactly​ ​how​ ​to​ ​respond.​ ​Unlike​ ​a​ ​traditional​ ​website​ ​or​ ​app where​ ​you​ ​may​ ​have​ ​at​ ​most​ ​10​ ​options​ ​at​ ​any​ ​given​ ​time,​ ​the​ ​open-ended​ ​nature​ ​of​ ​conversation​ ​means​ ​you​ ​have to​ ​account​ ​for​ ​an​ ​almost​ ​unlimited​ ​number​ ​of​ ​paths​ ​the​ ​user​ ​might​ ​take.

5.1. There are no call to actions. Actually call to actions are implicit on the way a question or statement is phrased by the AI chatbot

5.2. Ideas borrowed from Search Bars

5.2.1. Ideas that have worked for search bars like dropdown could be used here

5.2.2. Dropdown with possible responses to guide the user

5.2.3. Autofilling users responses/questions

5.3. Common Sections

5.3.1. Contact Information

5.3.2. Support Information

5.3.3. Careers

5.3.4. Pricing

5.3.5. There is no reason why any of the above should not be able to be answered by a chatbot But not the primary chatbot If a doctor is your AI chatbot then he/she should transfer you to the assistant chatbot

5.4. Testimonials

5.4.1. Now you can just ask the user in a natural way and the end of the conversation with the chatbot

6. needfinding / research

6.1. interviews

6.1.1. consent form send it via email to bring a professional predisposition keep previous consent forms on the table during interview to show an adequate number of participants 2 copies, 1 copy for you, 1 for him. Designer signs first

6.1.2. after the interview you should have understood mental models people build tools they use terminology they speak workflow underlying goals that people have underlying values

6.1.3. all stakeholders company that is building the product users / patients doctors (indirect stakeholder)

6.2. intercepts

6.2.1. You intervene politely, to ask something specific from close

6.2.2. Obviously you must found a crowd that is related to what you searching for and introduce yourself

6.2.3. e.g. ask brief/quick questions to people on related events

6.2.4. e.g. ask brief/quick questions to people who are visiting the doctor

6.3. contextual inquiry

6.3.1. pay a site visit to understand the context of user's actions

6.3.2. observe what people actually do versus what they say they do

6.3.3. e.g. go to waiting rooms in doctors and observe people interactions, unfortunately you cannot go inside

6.3.4. e.g. if I am using an AI chatbot that is a doctor I could be using it and have friends nearby who I also ask, or have family members taking care of me who I still also ask for extra information

6.4. diary study

6.4.1. give patients diaries to fill as they interact with their doctors and record their experiences

6.5. Research the internet

6.6. Use competitors solutions and compare advantages / disadvantages

6.7. Remote desktop

6.7.1. ask users to share their screen with you and talk out loud as they interact with the AI chatbot

6.8. Interpretation

6.8.1. You had some initial assumptions. Are they still valid?

6.9. user stories

6.9.1. e.g. have users share stories of their interaction with doctors, let them express theirselves

6.9.2. either on interview or in a diary

7. journey maps

7.1. for each persona

7.2. timeline

7.3. emotions / empathy map of the user on every step

7.3.1. how it feels to be that persona in this experience

7.4. touchpoints

7.4.1. possible responses of the user

7.5. sketch the journey

7.5.1. no templates, just sketch it as you imagine it to describe the journey with cohesion

7.5.2. improve its look to make it a deliverable

7.5.3. share and use set a period to re-evaluate the journey of the user (each quarter of year perhaps) stakeholders must be able to view it

8. inspiration / ideation

8.1. build ideas alone

8.2. communicate with the technical team to share ideas and get feedback on technological limits

8.3. create 100 ideas and keep only the one that makes most sense

9. personas

9.1. use stereotypical users

9.1.1. do NOT go with the one and only awesome doctor "key to failure is trying to please everybody"

9.2. name

9.3. photograph

9.4. status

9.4.1. primary persona pick primary if in design you can also satisfy the rest

9.4.2. secondary

9.4.3. lower could have no status

9.5. quote, key objective

9.5.1. e.g. "I have a small issue and I am little bit worried what it is but I do not consider it that important to go to the doctor. Is my case serious or not? I wish there was a cheap doctor to ask such quick questions"

9.6. attitudes/opinions to product

9.6.1. this comes from research

9.6.2. e.g. "I have never interacted with chatbots before but I would be keen to try it out"

9.7. list of key goals

9.7.1. e.g. Understand how likely it is that his issue is severe or could be cured by being patient

9.7.2. e.g. Save money by spending less

9.7.3. e.g. Do it fast while sitting on home sofa

9.8. age

9.8.1. e.g. 30

9.9. job title

9.9.1. e.g. Taxi Driver

9.10. location

9.10.1. e.g. London

9.11. more attributes

9.11.1. fears

9.11.2. motivations

9.11.3. marital status

9.11.4. hobbies

9.11.5. internet connection speed

9.11.6. how technologic is he

9.11.7. computer usage

9.11.8. internet usage

9.11.9. applications used frequently

9.11.10. typical work activities

9.11.11. job goals

9.11.12. job responsibilities

9.11.13. industry

9.11.14. interaction with colleagues

9.11.15. typical day at work, timeline

9.11.16. work skills

9.11.17. spending habits

10. ​Speaking​ ​as​ ​a​ ​user,​ ​I have​ ​already​ ​had​ ​bad​ ​experiences​ ​interacting​ ​with​ ​AI​ ​and​ ​it​ ​is​ ​the​ ​fact​ ​that​ ​most​ ​probably​ ​nobody​ ​thought​ ​of​ ​“Okay, if​ ​I​ ​was​ ​a​ ​personal​ ​assistant,​ ​trying​ ​to​ ​be​ ​the​ ​best​ ​personal​ ​assistant​ ​I​ ​can​ ​be,​ ​how​ ​would​ ​I​ ​write​ ​this​ ​response”? It​ ​is​ ​the​ ​responsibility​ ​of​ ​the​ ​AI​ ​Interaction​ ​Designer​ ​to​ ​get​ ​that​ ​information​ ​from​ ​the​ ​user​ ​as​ ​quickly​ ​and​ ​completely as​ ​possible​ ​without​ ​any​ ​ambiguity​ ​that​ ​could​ ​lead​ ​to​ ​extra​ ​back​ ​and​ ​forth​ ​and​ ​ultimately​ ​think​ ​about​ ​how​ ​simply written​ ​sentences​ ​impact​ ​user​ ​outcomes.

10.1. Big Data guaranteed

10.2. The AI Interaction Designer should have some analytical skills to do a proper A/B testing or null hypothesis

10.3. The AI interaction designer should know what kind of things need to be measured in order to make data analysis possible

10.3.1. For example you need to measure after how many interactions the users abandon the chatbot or have a successful finalized interaction with the doctor

11. Storyboarding

11.1. just build full complete stories

11.2. stories that the users are going through

11.3. stories that we want them to go through

11.4. one story should be drawn in less than one hour

12. Red Routes

12.1. Most common use cases

12.2. Identify them qualitatively

12.2.1. easier to read the chats than analyze them

12.3. Identify them quantitatively

12.4. AI Interaction designer should tune the chatbot to optimize for these use cases or guide AI engineers to

13. Card Sorting

13.1. no need to directly organize content anymore

13.2. still useful to know how to organize knowledge for your chatbot and organize the order of the potential answers or the current context of the chatbot

14. usability evaluation / expert reviews

14.1. Jakob Nielsen's usability guidelines are still useful

14.2. Jakob Nielsen's

14.2.1. visibility of system status when the users ask "what is going on?" immediate feedback the user should be able to ask to be reminded and get a summary of what the user was discussing last time with the AI chatbot

14.2.2. match between system and the real world "What are you talking about?" the user would say people should not go through a long mental process to understand your system use the right metaphors

14.2.3. User control and freedom Have an undo action for the user, we all make mistakes give the user the ability to turn the sound on, instead of suddenly appearing

14.2.4. Consistency and standards "I know what to do with this" the user must think follow standard conventions that users already know

14.2.5. Help users recognise, diagnose and recover from errors "How do I fix this?" the user would ask give constructive help and be polite

14.2.6. Error prevention "Thanks for helping me avoid making that mistake" support undo confirmation for critical actions warning, alarms for critical situations

14.2.7. Recognition rather than recall "Erm, sorry, I've forgotten" the user could say Minimize the user's memory load do not underline a text because people think it is a link

14.2.8. Flexibility and efficiency of use Create a design for both experts and novices do not use enormous dropdown menus Command line calendar to type a phrase and parse it instead of setting up with multiple clicks

14.2.9. Aesthetic and minimalist design Don't make me think puke! Simple and natural lot of text is generally bad BUT craigslist have managed to create a good design

14.2.10. Help and documentation "How do I...?" Write it for frustrated and angry users Solutions to problems

15. analytics and metrics

15.1. total visits and unique users

15.1.1. still relevant for AI Interaction Designer

15.2. page views total

15.2.1. in average how long do the users interact with the AI

15.3. pages per visit

15.3.1. this is now translated to how many interactions the user had with the AI

15.4. bounce rate

15.4.1. People who are abandoning the AI interaction after a couple or very few interactions

15.5. new vs return visitors

15.5.1. still relevant

15.5.2. new high and return low good marketing lots of people start chatting with the AI, the AI has failed to meet their needs, they have not returned

15.5.3. new low and return low your app is dying

15.5.4. new low and return high the AI chatbot is working for a committed community, perhaps niche market (tech-savvy users), not good marketing

15.5.5. new high and return high you are doing good!

15.6. time per visit

15.6.1. This is replaced by time between each interaction

15.6.2. as well as total time spent for interaction (either failed or successful)

16. user testing

16.1. No need for paper prototypes

16.1.1. Just chat with people

16.1.2. could be remote but even better on-site

16.1.3. Again two persons for the prototype Role of computer, chats with users playing the role of the machine Role of moderator. observes and keeps notes


16.2.1. still very useful tool

16.3. field study

16.3.1. the tasks are executed in the office or home or inside a train or a bus where the participant is likely to use the app

16.4. measure

16.4.1. task completion rate how many tasks are being completed successfully how fast the tasks are being completely successfully how fast the tasks are failing

17. extras

17.1. Tutorials

17.1.1. The users should be guided how to chat with the AI

17.1.2. Another agent, another (dummy) chatbot, introduces you to the doctor and how you can communicate with him (the AI chatbot)

17.2. Compare Real Experience with your AI chatbot

17.2.1. for example ask the user to express how many hours we would spend going, waiting, visiting and returning from the doctor

17.2.2. tell the user that he is not alone. Show what others have expressed Make people to be part of something > App-monetization-tips-7

17.2.3. Make the user feel good that his voice has been heard and that other people have a similar experience with him

17.2.4. After each interaction remind him that he has saved so much time and that now he can enjoy his free time and be more healthy

17.2.5. Even tell him beforehand that he is likely to finish his interaction by 17:15 (started at 17:00). If the user can schedule in his head what he is doing for the next hour then he is more likely to commit this (predefined) time to communicate with the chatbot

17.3. After an interaction that ended successfully for the user congratulate him

17.3.1. express to the user that he is a master on talking with AI chatbots

17.4. Chatbot means you can interact with the user even after abandoning the app

17.4.1. In the old days if the user would bounce (abandon the app) then you had only a few options

17.4.2. You could send him an email to continue from where he was left or you could send a relevant notification

17.4.3. However now if the answer to your last question was not effective enough, after updating your chatbot, you can come back and try again by sending a followup message

17.5. tools for AI interaction designer

17.5.1. prototyping is just plain text

17.5.2. mindmap for personal ideation

17.5.3. Tracking track everything a user does

17.5.4. Team Collaboration Slack git

17.5.5. testing User Testing test with real users Optimizely Optimizely focuses on one specific type of test, A/B testing

17.6. homepage

17.6.1. live map and a live counter of AI interactions

17.7. Let the user control the flow

17.7.1. The user might need to reach his goal as soon as possible or he might not

17.7.2. The user for example when interacting with a doctor AI might want to know more of why a process is like this, or what a term means, or what are the worst case or the best case of the progression of a disease etc.

17.7.3. That's why important the AI chatbot to have available more context than necessary to be able to compensate for those questions

17.8. ok for keyboard users but what about mouse users?

17.8.1. You need to have predetermined responses for people who like to click

17.8.2. What about people who like to scroll? Enable scrolling for your chatbot This could be a default response like "Tell me more" on behalf of the user

17.9. Chatbot vs Humans

17.9.1. Have a very distinctive design to separate when people are chatting with an AI and when chatting with a real person

17.9.2. change the colors of the interface if necessary. It must be very obvious