Needfinding

Human-AI Interaction
Ali Saudi
Academic Year 2022/2023
Goals

▪ Understanding the system requirements and user needs

o “Needfinding”
▪ Knowing methods for Needfinding
▪ Knowing steps for Needfinding
▪ Knowing Needfinding in AI-powered systems

2
Human-AI Interaction
Human-Centered Design Process

Needfinding

3
Human-AI Interaction
Main Needfinding questions

▪ Needfinding = Finding Potential User Needs

o What do users need?
o What do users want?

▪ That also requires

o Who are the users?
o How are they doing it, now?
o What is the context in which they are doing it?
o Can’t we just ask them?

4
Human-AI Interaction
 «Know Your Users»

Know Your Users (1)

▪ Who are the users of the system?
o Uniform, or different categories/groups?
o Young/old? Novice/experienced?
o Do not think of “generic” users, split the categories
▪ *You* are not a [representative] user
o Designers and developers’ skills, knowledge, attitude, background, interests, …
are totally unlike those of your users
o Except by chance (e.g., you are also students, developers, …)
▪ The client is not a [representative] user
o Bosses, managers, directors, … believe they know their employees and their jobs.
Actually, they don’t
o Always seek the actual users that will use the system

5
Human-AI Interaction
Know Your Users (2)
▪ Talking to users
o Surveys
o Interviews
o Direct involvement (participatory design)
o Bypass corporate policies
o Understand real current behavior, pain points, workarounds, …
▪ Watching users
o Observation sessions
o Video recording (and analysis)
o Diaries
o Analyze their work (artifact, processes, action sequences)
o Discuss with users the findings of the observation (may discover the “why”)

6
Human-AI Interaction
Know Your Users (3)

▪ Imagining users
o When real users are not available
o Imagine how a real user would behave (very difficult)
o Building “imaginary” users: personas
• Detailed description of hypothetical persons in a given role
• Imagine them as they were a real person

7
Human-AI Interaction
Ways of Knowing Your Users
▪ Knowing by asking
o Interviews, Surveys
▪ Knowing by experimenting
o Lab experiment, statistical analysis
▪ Knowing by designing and building
o Ideation
▪ Knowing by deploying
o Field of study, Diary studies, log data analysis
▪ Knowing by sensing
o Remote user monitoring
▪ Knowing by Observing
o Case studies

8
Needfinding methods
Techniques adopted in Needfinding and Requirement Analysis

9
Human-AI Interaction
Needfinding methods

▪ Qualitative Approach
- Goal: understand the “how”, the “why” and meanings
- Discover concepts, flows and relationships
- “being there”, get immersed
text visuals
- Methods
- Interviews Conclusion (the why)
- Observations
- Visual, audio methods

10
Human-AI Interaction
Needfinding methods (Cont’d)
▪ Quantitative Approach
▪ Goal: answer the question of how many…?
▪ Discover trends in a large group of people
▪ Methods
o Online Questionnaires
o Log-File Analysis of Webserver
o Sentiment Analysis (automatic analysis of e.g. blogs, Twitter, Facebook
to determine attitudes and moods)

11
Human-AI Interaction
Needfinding methods (Cont’d)

Qualitative questions Quantitative questions

•Tell me about the experience with…? •How would you rate the quality of the
•Could you describe how you…? image on a scale from 1 to 5 (1: poor, 5:
very good)
•What did you feel when…?
•Have you ever written a book (y/n)?
•How was it possible that…?
•What is your age?
•What are your thoughts about…?
•How many coffees do you drink during a
•Can you give me an example of…? normal working day (none, 1, 2, 3-4,
•What do you mean by…? more than 4)?
•Can you tell me more about…?

12
Human-AI Interaction
Methods outline

▪ Observation
▪ Diaries
▪ Interviews
▪ Surveys

13
Human-AI Interaction
Observation
“You can observe a lot, just by watching” – Yogi Berra

14
Human-AI Interaction
Observation
▪ Embed in the users’ environment, culture, behavior
▪ Goal: to obtain the necessary data to influence interface (re)design
▪ Learn the language of users and their environment
▪ Listen and observe carefully
o Sometimes ask questions and clarifications
▪ Audio-Video Record / Take Notes
▪ Risks:
o Misinterpret observations
o Disrupting normal practice
o Overlooking important information

15
Human-AI Interaction
What should learn by observation? (1)

1. What do people do now?

2. What values and goals do people have?
3. How are these activities embedded in a larger ecology?
4. Similarities and differences across people
5. Other types of context, like time of day

16
Human-AI Interaction
What should learn by observation? (2)

▪ Process vs Practice
▪ Process: how things are officially supposed to happen, and are officially part
of the training
▪ Practice: set of workarounds, practical tricks, information learnt from the field
and from experience, etc., that are part of daily activities

17
Human-AI Interaction
Diaries
Move the observation to the daily routine, with the help of the users

18
Human-AI Interaction
Diaries for Longer observation periods

▪ Observing users for long periods of time in many locations (“in the wild”) is
not possible by an observer
▪ Diaries are tools (paper-based or computer-based) that require users to take
note of their actions
o When they perform a specific action
o At predefined time intervals
▪ Stronger motivation should be ensured (incentives?)
▪ Analysis of the diaries may be done off-line (by researchers) or in the context
of an interview

19
Human-AI Interaction
Interviews
Asking users about their needs and desires…
…what could possibly go wrong?

20
Human-AI Interaction
Main forms of interviews

▪ Surveys
o Fast, more superficial
o Sets of questions with predefined possible answers
o Paper-based or on-line
▪ In-person interviews
o Time-demanding, in-depth knowledge
o Structured vs Unstructured
o One-to-one vs Focus groups

21
Human-AI Interaction
Choosing participants for interviews

▪ Representative of target users

o [All] Stakeholders
▪ May be current users of a similar system
▪ Might also be the non-users (for a new product)
▪ Approximate, if necessary (with similar users)

▪ Incentives, motivations, small gifts

22
Human-AI Interaction
Executing interviews

▪ Schedule time&place comfortable for users

▪ Introduce yourself, explain your purpose
o You are not testing them; they are helping you
▪ Begin with open-ended, unbiased, non-leading questions
▪ Ask the question and let them answer
o Give enough time. The 2nd reply is often more interesting than the 1st.
▪ Follow-up with related questions. Deep dive into interesting points

23
Human-AI Interaction
Surveys
Requirements gathering through (on-line) surveys and questionnaires

24
Human-AI Interaction
On-line surveys

▪ Familiar, cost-effective, potentially reaching a very wide audience

▪ Results can be easily visualized and analyzed with statistical methods

▪ Should be very careful with setting the goal of the survey

o First define what statistics and/or charts you need, then design the survey
structure and questions

25
Human-AI Interaction
Risks in surveys

▪ Good for a shallow view over a large base of respondent, but not for a “deep”
analysis
▪ Impossible to ask follow-up questions
▪ Biased data if questions rely on user’s memory or on “sensitive” issues
(money, emotions, …)

▪ Finding a representative target population

o Stratification

26
Human-AI Interaction
Survey structure

▪ Declare the purpose of the survey and the expected time

▪ One or more sections with targeted questions
▪ Background information about the users
o Limit “mandatory” fields to a minimum

27
Human-AI Interaction
Needfinding steps
Steps to Identify User Needs

28
Human-AI Interaction
Needfinding steps
▪ User Segmentation - Personas
o A way of understanding users’ needs by breaking the audience down into
segments based on some shared characteristics.

▪ Scenarios
o A short story about a specific user with a specific goal at your system.
o Give you a deeper understanding of users. Therefore, you can develop the
right content at the right level.

29
Human-AI Interaction
User Segmentation - Personas
▪ What is it?
o Generalized representation of user group

▪ How is it done?
o Interview representative number of members
of user group
o Extract most important characteristics
o Describe Persona

30
Human-AI Interaction
User Segmentation - Personas
▪ Personas
o A character sketch based on the user research.
o Extrapolate from a general set of characters to a specific case.
o An imaginary person that represents one of the users' segments.
o Contains a character sketch of:
1. Name, Age, Location
2. Photo
3. Quote
4. Goals
5. Motivations
6. Lifestyle

31
Human-AI Interaction
User segmentation - Personas

32
Human-AI Interaction
Example: “Around the block”
Imagine that we are designing an application named “Around the block” that aims
to provide location-based services like locating the nearby restaurants. Also, it
enables people to claim businesses by providing basic info about the business place
like the opening hours, offered services, and contact info.
1. Determine the following points by applying the user segmentation approach on
“Around the block”:-
i. Target users
ii. Main user groups (segments)
iii. Construct the users’ groups profile table
2. Based on your answers in 1, create personas to represent your target users.
3. Determine the business objectives for “Around the block”.

33
Human-AI Interaction
Example: “Around the block”
1. Determine the following points by applying the user segmentation approach on
“Around the block”:-
i. Target users:
o Smartphones users – Business owners
ii. Main user groups (segments):
o Travelers - Small Business Owners – Companies Managers

34
Human-AI Interaction
Example: “Around the block” Characteristic Traveler Small Business Companies Managers
Owner
Age 16-35 21-50 40-60
iii. Construct the users’ groups Education Mid Grades / Vary from Most of them have
High School illiterate till College Diploma / BSc
profile table Degrees /BSc college diploma Degree
Degree
Language English/Arabic Arabic as mother- At least: English/Arabic
tongue
Language
Mobile/Web High None-Low Med-High
Experience

Domain Med-High Low Med-High

Expertise
Task Initial: Med Initial: Very Low Initial: Med
Knowledge After two days : After three weeks After one week : High
High : Low-Mid
Expectations - Speed of task - Ease of use - Preserve company
- Real-time data - Increase Profit image
about places and Reputation - Get real-time feedback
- Reviews and - Straight Forward -Summarized user stats
ranking system Tasks - Visual aids to
- Friends near by - Visual aids to represent info like
- Accurate describe tasks using graphs
navigation

35
Human-AI Interaction
Example: “Around the block”
2. Based on your answers in 1, create personas to represent your target users.

Sample Persona for small business

owners
Sample Persona for travelers

Background: Background: Things he Wants To Do:

Name: Beth Greene Name: Mr. Smoke • Free Advertisement/Marketing For
Age: 16-Year-old Things She Wants To Do: Gender: Male his Workshop
Gender: Female • Discovering new Entertainment Age: 42-Year-old • Receive Online Recommendations
Education: 7th Grade places Education: Expelled From And gain reputation
Student • Share Awesome Locations with her 7th Grade Class
Tech-Skills: Web Surfer Friends Tech-Skills: None
Occupation: Small Business
Motivations: Owner (Cars Workshop)
• Keeping Touch with her
Friends Motivations:
• Social Gathering in • Make More Profit
cheerful places

36
Human-AI Interaction
Example: “Around the block”
3. Determine the business objectives for “Around the block”.
o Reduce the look-up time for specific services in a given location.
o Make users discover hidden places around them

37
Human-AI Interaction
Scenarios
■ Formal Definition: “a narrative or story that describes the activities of one
or more persons, including information about goals, expectations, actions,
and reactions” (Rosson & Carroll)

■ Informally; a scenario consist of

– An Abstract User
– Expected Outcome
– Computer Facilities
– Constraints
– Emotional Context
– Design Questions

38
Human-AI Interaction
Example: “Flight Booking System”
State the scenario components and design questions/issues for the following
scenario:
Jamal wants to plan a trip from Alex to Aswan to see his family. He has 4-5 days
free but should be back to work by Tuesday at the latest. Since he’s paying for it,
he wants the cheapest fare he can get. Once in Aswan, he’ll need a car, he prefers
Taxi-Cab since it has Premium Service with his company. He’s staying with family,
so no hotel is needed. He has frequent flyer miles with Egypt Air, so would prefer
to use them, if the price is right. It’s lunchtime and he has a meeting in 30
minutes, so he’s in a hurry. At least he’d like to look at some flight options and
rental car prices. If it looks great and he can do it quickly, he’ll make the
reservations with a credit card.

39
Human-AI Interaction
Example: “Flight Booking System”
Scenario Components
Abstract user Name = Jamal, Location = Alex
Expected outcome Book the cheapest flight and rental car
Computer facilities Web browser on PC/Cellphone + Internet access + (valid credit card)
Constraints Operation to be done in 3o mins max – Needs the cheapest option – Online Payment –
Preferences on the flight carrier and the rental car company
Emotional context Visiting his family
Design questions − Bring flight search functionality to the front to be the first thing seen in the UI
− Sorting search results from cheapest to highest
− Add search filters like adding the preferable flight carrier to refine the search results
− Do I need to handle rental car & hotel reservations?
− Online payments in house vs getting a third-party company (ex. PayPal)
− Support one-hand operations (Jamal will be eating during performing the task as it’s
lunchbreak)

40
Human-AI Interaction
Example: “Bank Customer Service Agent”
State the scenario components and design questions/issues for the following
scenario:
On this particular Sunday morning, two of Aly’s colleagues haven’t come to work.
He has a long line of customers. He has clear instructions from his boss not to
spend too much time with every customer, and to resolve every question as
quickly as possible. One customer approaches him with multiple requests. The
customer wants to check his outstanding loan balance and make a partial
payment for the loan. He wants travelers’ cheque issued for his foreign travel, and
also has inquiries about the status of a particular cheque that he had deposited
last week.

41
Human-AI Interaction
Example: “Bank Customer Service Agent”
Scenario Components
Abstract user Name = Aly, Location = Bank
Expected outcome Serve customers’ needs/inquires
Computer facilities Banking system on PC
Constraints − Getting the job done + min duration
− Colleagues are absent → increased workload → error rate is high
Emotional context Visiting his family
Design questions − No need for unnecessary visual load (avoid direct manipulation + colors)
− Provide direct access to all system functionalities (either by conversational menu or
commands)
− Min number of clicks/steps needed to perform an action → will make user spend the min time
in doing a certain task.
− Prevent users from making errors by supporting selection over typing, recognition rather than
recalling, and add last-step confirmation for all actions
− Group related features together
− Make system suggest actions for users based on the current object in hand (ex. suggest loan
control options like payment, withdraw…etc. while viewing a loan page)

42
Human-AI Interaction
Needfinding in AI-powered
systems
43
Human-AI Interaction
What’s unique in designing AI-powered systems?

● Data matters
○ Big data preferred, Cost & time associated in collecting and handling data,
Bias and fairness issues

● Model matters
○ Probabilistic, Hard to predict, Possibly a black box, Transparency,
Interpretability, Resources

● Interaction matters
○ Mental model, Explanation & feedback, Intelligent UI, Helping users in the
task, Human-AI collaboration

44
Human-AI Interaction
Our Three-Stage Framework
DATA MODEL INTERACTION

● Collecting, ● Train + Test + ● Model results

cleaning, Parameter (and
processing, tuning explanations) are
labeling, ● Apply to a new shown to the
verifying problem for user through UI.
data prediction. ● User feedback
● Crowdsourcing, ● Picking the can be fed back
Logs, User input, right model is to the system.
... challenging.
45
Human-AI Interaction
Steps in applying ML to practice

https://www.youtube.com/watch?v=nKW8Ndu7Mjw&vl=en
46
Human-AI Interaction
AI Design Process

●
● Gather data
● Prepare the data
DATA
● Choose a model
● Train the model MODEL
● Evaluate the model
● Tune hyperparameters
● Predict
●

47
(Human-Centered) AI Design Process

● Set a goal & identify user needs PLANNING

● Gather data
● Prepare the data
DATA
● Choose a model
● Train the model MODEL
● Evaluate the model
● Tune hyperparameters
● Predict
● Design human-AI interaction
INTERACTION
● Apply to UI
● Evaluate & Improve
48
Set a goal & identify user needs

▪ User-centered design process is helpful here.

▪ Building AI systems should also consider its users.
▪ Without this stage, it’s likely to yield technology that kind of works but
nobody uses.
▪ Netflix: "The user either finds something of interest [within the first 60 or 90
seconds] or the risk of the user abandoning our service increases substantially,"

Gomez-Uribe, Carlos A., and Neil Hunt. "The netflix recommender system: Algorithms, business value, and innovation." ACM
Transactions on Management Information Systems (TMIS) 6.4 (2015): 1-19.
http://www.spring2innovation.com/2019/04/design-thinking-vs-user-centred-design/

49
References
▪ David Benyon: Designing Interactive Systems, Chapter 7
▪ Alan Dix, Janet Finlay, Gregory Abowd, Russell Beale: Human Computer Interaction, 3
rd Edition, Chapter 5
▪ Shneiderman, Plaisant, Cohen, Jacobs, Elmqvist & Diakopoulos: Designing the User In
terface: Strategies for Effective Human-Computer Interaction, 6th Edition, Chapter 4,
5.4
▪ COGS120/CSE170: Human-Computer Interaction Design -
http://courses.pgbovine.net/cogs120-cse170/schedule.htm - Week 1
▪ Lazar, Feng, Hochheiser: Research Methods in HCI, 2010, Chapter 9 (Observation),
Chapter 5 (Surveys)

50
Human-AI Interaction