DePiction
-

AI Agents

DePiction
-

AI Agents

DePiction
-

AI Agents

PROBLEM

In this study, we explore the behavior of AI in cooperative gameplay (social games)–Pictionary, with an intention to understanding atypical activity (cheating).

SOLUTION

To test our hypothesis, we developed an online social game–Pictionary. Concurrent test sessions enabled us to collect data, leading to interesting findings.

IMPACT

I optimized the design to be fluidic adopting to diverse screen sizes leading to widespread adoption and optimized the code to support 100+ concurrent sessions.

KEY DECISIONS

The major decisions were of understanding the system (AI) complexity, helping me to implement relevant design principles for the same.

TOOLS USED

Introduction

Introduction

Human-AI Interaction

People use digital whiteboards to share ideas and brainstorm, human-AI interaction will improve this by removing aberrant activities (profanity, cheating), therefore, improving user experience.

Atypical behavior

text

No text

123

No numbers

No symbols

The above mentioned actions are classified as atypical behavior. Gameplay only allows the use of sketches.

Identified Problems

Identified Problems

Design Systems

The application did NOT have a dedicated design system to guide its future iterations.

AI Complexity

In order to design for an AI application the correct level of AI complexity was to be identified.

UI Design

The UI design was lacking in basic UI principles such form, hierarchy, scale and harmony.

The Game - Pictionary

The Game - Pictionary

Pictionary is a social game. But for the study it was restricted between two players. The drawer - person who sketches the prompted word, and the guesser - person who guesses the word based on the sketch.

Why Gamification?

Remote Gameplay

We used gamification to understand the difference in collaboration between human-human and human-AI.

Playing against the unknown

Gamification will help in understanding the behavior of users when they play against an AI-agent unknown.

Key Focus

Key Focus

How to design to optimize for AI complexity and partially observable cooperative games?

How to design to provide a delightful gaming experience, engaging players in a ‘flow state’?

AI Complexity

AI Complexity

Identified Complexity

Level 1

Current Complexity

Level 3

Evolving Complexity

Current level of complexity is at 1 but it may increase to level 3 once the database becomes dynamic.

Design Guidelines

System

Outputs

Fixed outputs - Hit, Miss,

False Alarm, Correct Rejection

System

Capability

Current database is fixed,

therefore probabilistic.

Re

Design

Level 1 is benefited by user testing and iterative design.

Design

Prototype

The prototype ensures the code is human-centric.

AI and Gamification

The broader impact is to make training datasets on atypical behavior (cheating, profanity) more accessible.

User behaviors

Turing Test!

The game encourages the players to guess whether they are playing against other players or AI agents. Which helps in identifying humanness of the agents.

Deviant Behaviors

The game also serves as a tool to identify the player’s tendency to behave atypically (writing words), testing the AI agent’s ability to identify the same.

Our Players

Our Players

User Profile

479

Participants

328

328 male and

98 female

388

388 right-handed

38 left-handed

19-60

Diverse age range

Jobs to be Done

1

Our users need a clean design implementation to make it easier for them to engage in the gameplay.

2

Our users did not know if they are playing against AI, therefore design implementation increasing transparency is imperative.

3

For optimal gameplay, the gameplay should be engaging - neither difficult nor simple helping our users in achieving ‘flow state’.

Goals

AI
Inclusion

Designing for AI ethics, ensuring user autonomy.

Responsive

UI

Diverse age group makes responsive design imperative.

Minimal

Design

WCAG guidelines ensures ease-of-access and accessibility.

Application
Development

The prototype was developed in a Django-based platform.

Pain Points–Legacy Design

Pain Points–Legacy Design

Navigation

The old design did not feature clear directions to the user for navigating through the website, as evident in the image shown above - there is no clear way to exit the ‘Terms and Conditions’ screen.

Color Palette

The initial design was using a lot of colors like - yellow, red, green, blue and many more, which made it confusing for our users, as they were unable to locate the CTA at times.

Legibility and Readability

It was difficult to establish hierarchy as the previous design was using a noisy background image, along with that the noisy background also made it difficult to locate the icons located at the top right corner of the screen.

Redesign Systems

Redesign Systems

Atoms

These atoms include basic HTML like forms, labels, inputs, buttons and others that cannot be broken down any further without ceasing to be functional.

Pencil

Button

NEXT

False Alarm

Molecules

UI elements functioning together as a unit. For example, a form label, search input, and button can join together to create a search form molecule.

Pencil

Eraser

Highlighter

Guesser on right track

Guesser on wrong track

Correct

Organisms

Complex UI components composed of group and molecules and/or atoms and/or other organisms.

Templates

Templates are page-level objects that place components into a layout and articulate the design’s underlying content structure.

tower

tree

house

treehouse

Type your guess h...

Pages

Pages are specific instances of templates that show what a UI looks like with real representative content in place.

Color Palette

Color Palette

Primary

Used across all interactive elements such as CTA’s, links, active states.

#F8FAFC

50

#F1F5F9

100

#E2E8F0

200

#CBD5E1

300

#94A3B8

400

#64748B

500

#475569

600

#334155

700

#1E293B

800

#0F172A

900

Secondary

Used for secondary button and to establish hierarchy among non-interactive UI elements.

#F8FAFC

50

#F1F5F9

100

#E2E8F0

200

#CBD5E1

300

#94A3B8

400

#64748B

500

#475569

600

#334155

700

#1E293B

800

#0F172A

900

Error

Used to draw users attention towards faults and errors to varying degrees.

#F8FAFC

50

#F1F5F9

100

#E2E8F0

200

#CBD5E1

300

#94A3B8

400

#64748B

500

#475569

600

#334155

700

#1E293B

800

#0F172A

900

Typography

Typography

Permanent Marker

This was the primary font for the application, but it was later substituted with Inter, to increase the readability and legibility of the application, overall.

Aa Bb Cc Dd Ee Ff Gg Hh Ii Jj Kk Ll Mm Nn Oo Pp Qq Rr Ss Tt Uu Vv Xx Yy Zz

1234567890

Inter - Redesign Typeface

Aa Bb Cc Dd Ee Ff Gg Hh Ii Jj Kk Ll Mm Nn Oo Pp Qq Rr Ss Tt Uu Vv Xx Yy Zz

1234567890

Iconography

Iconography

Google Material Icons

Google material design rounded icons were used to keep a harmony between the fonts and the icons, and to provide an overall friendly user interface for the target demographic.

Animations

Animations

Messages

In order to give our users the best possible experience, I reduced the messages section to floating messages, which can be toggled by the user, to a full messages screen if desired.

Graceful Degradation - THE REDESIGN

Responsive web design

The Pictionary application was designed under the theme of graceful degradation to optimally scale down the design when transitioning from a PC to mobile phones.

Development

Development

Responsive Web Development

I used a Django based framework to develop the Pictionary application. Working with HTML and CSS to convert my design into code.

Database Integration

We integrated Django with PostgreSQL for data collection and analysis. PostgreSQL adds an extra layer of security via password verification and role-based hierarchy.

Outcomes

Outcomes

Atypical behavior

Atypical behavior was observed more for texts rather than icons like arrows used for directing the guesser’s attention. Advancing the understanding of human-like AI agents.

Inclusion of AI

The acquired data from the experiments will aid in increasing the accuracy of the currently implemented AI agents, as they mimic human-like responses.

Learnings

Learnings

AI Agents

As I went ahead with the redesign it was insightful to learn about the capabilities of AI agents along with their limitations. We identified which framework is best for checking atypical behavior and laying future groundwork which explores ethics of AI agents.

Ethics in AI

I realized that how users may perceive AI is essential, raising questions on AI ethics. In the controlled experiment users were not made aware of the use of AI right away, but in general settings should they be?

Remote Collaboration

Most of my work was online, which helped me in developing skills for remote collaboration. Leading me to understand stakeholder expectations and communicating my thought process effectively and succinctly.

Designing Geospatial Experiences

Enriching Explorations - Geospatial

Next Project

Designing Geospatial Experiences

Enriching Explorations - Geospatial

Next Project

Designing Geospatial Experiences

Enriching Explorations - Geospatial

Next Project

Designing Geospatial Experiences

Enriching Explorations - Geospatial

Next Project