Conference PaperPDF Available

Wander: An AI-driven Chatbot to Visit the Future Earth

Authors:

Abstract and Figures

This artwork presents an intelligent chatbot called Wander. This work used knowledge-based story generation to facilitate a narrative AI chatbot on daily communication platforms, producing interactive fiction with the most accessible natural language input: text messages. On social media platforms such as Discord and WeChat, Wander can generate a science-fiction style travelogue about the future earth, including text, images and global coordinates (GPS) based on real-world locations (e.g. Paris). The journeys are visualised in real-time on an interactive map that can be updated with participants’ data. Based on Viktor Shklovsky’s defamiliarization technique, we present how an AI agent can become a storyteller through common messages in daily life and lead participants to see the world from new perspectives. The website of this work is: https://wander001.com/
Content may be subject to copyright.
Wander: An AI-driven Chatbot to Visit the Future Earth
Yuqian Sun
yuqiansun@network.rca.ac.uk
Computer Science Research Centre,
Royal College of Art
London, UK
Chenhang Cheng
koch3092@gmail.com
Unaliated
Shanghai, China
Ying Xu
yingxu2030@gmail.com
Department of Industrial Design,
School of Art and Design, Wuhan
University of Technology
Wuhan, China
Yihua Li
cory958014884@gmail.com
Department of Product Design,
College of Fashion and Design,
Donghua University
Shanghai, China
Chang Hee Lee
changhee.lee@kaist.ac.kr
Aective Systems and Cognition Lab,
Industrial Design Department,
College of Engineering, Korea
Advanced Institute of Science and
Technology
Daejeon, South Korea
Ali Asadipour
ali.asadipour@rca.ac.uk
Computer Science Research Centre,
Royal College of Art
London, UK
ABSTRACT
This artwork presents an intelligent chatbot called Wander. This
work used knowledge-based story generation to facilitate a nar-
rative AI chatbot on daily communication platforms, producing
interactive ction with the most accessible natural language in-
put: text messages. On social media platforms such as Discord and
WeChat, Wander can generate a science-ction style travelogue
about the future earth, including text, images and global coordinates
(GPS) based on real-world locations (e.g. Paris). The journeys are vi-
sualised in real-time on an interactive map that can be updated with
participants’ data. Based on Viktor Shklovsky’s defamiliarization
technique, we present how an AI agent can become a storyteller
through common messages in daily life and lead participants to
see the world from new perspectives. The website of this work is:
https://wander001.com/
ACM Reference Format:
Yuqian Sun, Chenhang Cheng, Ying Xu, Yihua Li, Chang Hee Lee, and Ali
Asadipour. 2022. Wander: An AI-driven Chatbot to Visit the Future Earth.
In Proceedings of the 30th ACM International Conference on Multimedia (MM
’22), October 10–14, 2022, Lisboa, Portugal. ACM, New York, NY, USA, 2 pages.
https://doi.org/10.1145/3503161.3549971
1 INTRODUCTION
In this project, we created an AI-enabled chatbot called Wander (Fig
1), an android that wanders on the ctional future earth, receiving
messages from the 21st century through widely used instant mes-
saging software (IM): WeChat (Chinese version) or Discord (English
version). This work aims to provide an intersection between real-
ity and virtuality: all ctional content, including text and images,
is generated in real time based on open data, real-world location
Permission to make digital or hard copies of part or all of this work for personal or
classroom use is granted without fee provided that copies are not made or distributed
for prot or commercial advantage and that copies bear this notice and the full citation
on the rst page. Copyrights for third-party components of this work must be honored.
For all other uses, contact the owner/author(s).
MM ’22, October 10–14, 2022, Lisboa, Portugal
©2022 Copyright held by the owner/author(s).
ACM ISBN 978-1-4503-9203-7/22/10.
https://doi.org/10.1145/3503161.3549971
knowledge. Each trip is like a text-based adventure, but there are
no xed choices, and every location can be easily accessed through
text messages.
Figure 1: Chatbot and character design of Wander
When chatbots are developed as social agents, they are mostly
considered assistants (like Siri) for certain tasks. When the AI agent
becomes a character, it is usually in an individual environment
within a game and has limited options and plots. Even the games
using a state-of-the-art natural language processing (NLP) model
that allows free input, such as AI Dungeon[
5
], are separate envi-
ronments and story settings with little connection to the real world.
We believe that the human–AI relationship can take a step further.
Thus, the primary motivation behind this work is to re-imagine
the role of AI as a storyteller. An AI character might not be limited
to systems or scripted characters in a game but might also have a
7250
MM ’22, October 10–14, 2022, Lisboa, Portugal Yuqian Sun et al.
Figure 2: Workow of Wander.
connection to our real world, as a messenger between our life and a
ctional world. The main contributions of this work are as follows:
(1) We designed a prototype interactive system to merge users’
input and real-world knowledge to procedurally generate human–
AI co-creative playable stories.
(2) We demonstrated AI-generated interactive ction through
an IM chatbot that can be easily accessed through text messages
with natural language inputs.
(3) We visualised players’ participation on a globe website, which
can be regarded as a speculative future chronicle developed by
humans and AI together.
The training data, the interaction system and the images are all
from reality and existing knowledge. Thus, at a higher level, the
world here can be seen dierently through the AI’s understanding,
like Shklovsky proposed for the defamiliarisation technique: ’impart
the sensation of things as people are perceived and not as they are
known’[4].
2 THE ARTWORK
2.1 Wander bot
The core part of this project is the conversational agent Wander
bot (Fig 1). According to the pre-set background story, she is an
android that wanders the future earth, contacting people in the 21st
century through IMs.
Her journeys (travelogue) on the future earth are realised through
two types of command: ‘Visit’ and ‘Action’. Each time a participant
sends a location message with the ‘Visit’ command, Wander will go
to that place in a random year between 3000 and 5000 AD and then
send back travel notes, including GPS location, futuristic photos,
and an environmental description. Then, with the ‘Action’ com-
mand, participants can ask Wander to explore the location by by
asking her to search for life or explore the ruins.
2.1.1 Image and style transfer. Through Google Map’s API, we can
obtain the GPS location and an image of the place (if an image ex-
ists). We used the arbitrary style transfer model from Runway.ml[
1
]
because it only needs the original image and a style image to trans-
fer, and the average calculation time is under 10 seconds. This
o-the-shelf technology allows Wander to give instant responses.
2.1.2 Text generation model: Dreamily.ai. The text generation model
we used is called dreamily.ai[
2
], and it has a large data set of Chinese
and English ction. This module is similar to other language gener-
ation models like GPT-3[
3
] but focuses specically on stories. We
did prompt engineering work with Google Knowledge Graph
1
to
extract descriptive sentences about the location. With the variation
in prompts, Wander will produce a rich and surprising response
corresponding to the destination.
2.2 Future Map
Wander’s Map (Fig 2 bottom ) website is a record of public partici-
pation. All records of ‘Visit’ commands, including text and photos,
will be sorted in the timeline. Users can rotate the globe and check
the journey histories of each location. Through public participa-
tion, knowledge of the future earth will continue to be updated and
become a speculative future chronicle developed by humans and
AI together.
3 CONCLUSION
Our work presents an AI chatbot-based work, Wander, that gen-
erates interactive ction with open data from real-world maps. It
introduces a narrative context for human–AI co-creativity and then
invites humans to immerse themselves in the ctional world with
widely used IMs. Using real-world information, we aim to create
novel experiences that resonate with people’s personal experiences
and lead them to gain a new perspective to see the world.
4 ACKNOWLEDGEMENT
Thank ColorfulClouds Tech for supporting us with the AI model
dreamily.ai[2].
REFERENCES
[1] RunwayML | Machine learning for creators. https://runwayml.com/.
[2]
ColorfulClouds. Dreamily-beta by colorfulclouds tech. https://dreamily.ai/,
2021.
[3]
OpenAI. GPT-3 Powers the Next Generation of Apps. https://openai.com/blog/gpt-
3-apps/, 2021.
[4] Shklovsky, V. Art as Technique. Literary Theory: An Anthology 3. 1917.
[5] Walton, N. Ai dungeon. https://play.aidungeon.io/main/home, 2019.
1https://developers.google.com/knowledge-graph
7251
ResearchGate has not been able to resolve any citations for this publication.
Art as Technique. Literary Theory: An Anthology 3 . 1917 . Shklovsky, V. Art as Technique. Literary Theory: An Anthology 3
  • V Shklovsky
  • Shklovsky V.
GPT-3 Powers the Next Generation of Apps
  • A I Open
  • Open AI
Dreamily-beta by colorfulclouds tech
  • Colorfulclouds
ColorfulClouds. Dreamily-beta by colorfulclouds tech. https://dreamily.ai/, 2021.
GPT-3 Powers the Next Generation of Apps
  • Openai
OpenAI. GPT-3 Powers the Next Generation of Apps. https://openai.com/blog/gpt-3-apps/, 2021.
Art as Technique. Literary Theory: An Anthology 3
  • V Shklovsky
Shklovsky, V. Art as Technique. Literary Theory: An Anthology 3. 1917.