=Paper=
{{Paper
|id=Vol-2848/HAI-GEN-2020-Workshop-Preface
|storemode=property
|title=HAI-GEN 2020 : Workshop on Human-AI Co-Creation with Generative Models
|pdfUrl=https://ceur-ws.org/Vol-2848/HAI-GEN-Workshop-Preface.pdf
|volume=Vol-2848
|authors=Werner Geyer,Lydia Chilton,Ranhitha Kumar,Adam Tauman Kalai
|dblpUrl=https://dblp.org/rec/conf/iui/GeyerCKK20a
}}
==HAI-GEN 2020 : Workshop on Human-AI Co-Creation with Generative Models==
https://ceur-ws.org/Vol-2848/HAI-GEN-Workshop-Preface.pdf
HAI-GEN 2020: Workshop on Human-AI Co-Creation with
Generative Models
Werner Geyer Lydia B. Chilton
werner.geyer@us.ibm.com Columbia University
IBM Research AI New York City, New York
Cambridge, Massachusetts chilton@cs.columbia.edu
Ranjitha Kumar Adam Tauman Kalai
University of Illinois at Urbana-Champaign Microsoft Research
Urbana, Illinois Cambridge, Massachusetts
ranjitha@illinois.edu adam.kalai@microsoft.com
ABSTRACT due to breakthroughs in generative modeling using deep learning.
Recent advances in generative modeling will enable new kinds Ian Goodfellow’s work on face generation [5] and StyleGan [7], Ope-
of user experiences around content creation, giving us “creative nAI’s GPT-2 [9], or recent deep fake videos of Mark Zuckerberg [4]
superpowers” and move us toward co-creation. This workshop and Bill Gates [10]are prominent examples of content generated
brings together researchers and practitioners from both fields HCI by AI that is almost indistinguishable from human-generated con-
and AI to explore and better understand both the opportunities and tent. These examples also highlight some of the significant societal,
challenges of generative modelling from a Human-AI interaction ethical and organizational challenges generative AI is posing in-
perspective for the creation of both physical and digital artifacts. cluding security, privacy, ownership, quality metrics and evaluation
of generated content.
CCS CONCEPTS The goal of this workshop is to bring together researchers and
practitioners from both fields HCI and AI to explore the opportuni-
• Human-centered computing → Human computer interac-
ties and challenges of generative modelling from an HCI perspec-
tion (HCI); Interaction design; • Computing methodologies →
tive. We envision that the user experience of creating both physical
Artificial intelligence; • Applied computing → Arts and hu-
and digital artifacts will become a partnership of humans and AI:
manities.
Humans will take the role of specification, goal setting, steering,
high-level creativity, curation, and governance. AI will augment
KEYWORDS
human abilities through inspiration, low level creativity and detail
Generative modelling, artificial intelligence, generative design, user work, and the ability to test ideas at scale.
experience, collaboration, creativity Submissions in form of short papers, long papers and demos
ACM Reference Format: following the IUI paper and demo guidelines are encouraged but
Werner Geyer, Lydia B. Chilton, Ranjitha Kumar, and Adam Tauman Kalai. not limited to the following topics:
2020. HAI-GEN 2020: Workshop on Human-AI Co-Creation with Genera-
• Novel user experiences supporting the creation of both phys-
tive Models. In 25th International Conference on Intelligent User Interfaces
ical and digital artifacts in an AI augmented fashion
Companion (IUI’20 Workshops), March 17, 2020, Cagliari, Italy. ACM, New
York, NY, USA, 3 pages. https://doi.org/10.1145/nnnnnnn.nnnnnnn • Business use cases of generative models
• Novel applications of generative models
1 DESCRIPTION • Techniques, methodologies & algorithms that enable new
user experiences and interactions with generative models
Recent advances in generative modeling through deep learning and allow for directed and purposeful manipulation of the
approaches such as generative adversarial networks (GANs) [5], model output
variational autoencoders (VAEs) [8], and sequence-to-sequence • Governance, privacy, content ownership
models [6] will enable new kinds of user experiences around con- • Security including forensic tools and approaches for deep
tent creation, giving us “creative superpowers” and move us toward fake detection
co-creation and curation. While the areas of computational design, • Evaluation of generative approaches and quality metrics
generative design, and computational art have existed for some • User studies
time, content with unprecedented fidelity is now being produced • Lessons learned from computational art and design, and
generative design and how these impact research
Copyright © 2020 for this paper by its authors. Use permitted under Creative
Generative Design, Computational Design, or Computational
Commons License Attribution 4.0 International (CC BY 4.0).
Art are topics that have been around for a while (e.g. [1],[2],[3])
but for the most part have not been grounded in generative deep
learning approaches combined with a strong HCI perspective and
theory on co-creation. This workshop aims to lay the groundwork
for bringing this exciting area deeper into the field of HCI research.
IUI’20 Workshops, March 17, 2020, Cagliari, Italy Werner Geyer, Lydia B. Chilton, Ranjitha Kumar, and Adam Tauman Kalai
2 ORGANIZERS • Steven Wu, University of Minnesota
Werner Geyer is a Principal Research Staff Member and Research • Haiyi Zhu, Carnegie Mellon University
Manager at IBM Research in Cambridge, MA, where he is leading a
research team centered around AI Interaction technologies. He’s
been holding various roles as co-chair at ACM RecSys, including
4 WORKSHOP PROGRAM
general chair as well as a series of workshops and tutorials on
Social Recommender Systems. More recently, his team is exploring • 9:00 - 9:25 Opening & Introductions
generative modelling techniques in business settings. His website
is https://researcher.watson.ibm.com/researcher/view.php?person= • 9:25 - 10:10 Challenges in Building ML Algorithms for the
us-Werner.Geyer. He can be reached at werner.geyer@us.ibm.com Creative Community
Lydia B. Chilton is an Assistant Professor in the Computer Keynote 1 by Douglas Eck, Google AI
Science Department at Columbia University. For ten years she was Session Chair: Werner Geyer
a leader in the crowdsourcing research space of HCI, now she breaks
down problems for a combination of people and AI to solve. She • 10:10 - 10:50 Session 1 - Generative Music
organized the first CHI Workshop on Crowdsourcing and Human Session Chair: Lydia Chilton
Computation, which had over 100 attendees. She has lead the 2-
day crowdsourcing workshop and hackathon, CrowdCamp 3 times. [20 mins] Paper 1: Cococo: AI-Steering Tools for Music
Her website is http://www.cs.columbia.edu/~chilton/ She can be Novices Co-Creating with Generative Models Ryan Louie,
reached at chilton@cs.columbia.edu Andy Coenen, Cheng-Zhi Anna Huang, Michael Terry and
Ranjitha Kumar is an Assistant Professor in the Computer Sci- Carrie Cai
ence Department at the University of Illinois at Urbana-Champaign.
She develops data-driven design techniques for creating effective [20 mins] Paper 2: Latent Chords: Generative Piano Chord
user experiences, tying interface, interaction, and algorithmic de- Synthesis with Variational Autoencoders Agustin Macaya,
sign choices to user-centered goals. Her research has received best Manuel Cartagena, Rodrigo Cadiz and Denis Parra
paper awards and nominations at premiere conferences in HCI, and
been recognized by the machine learning community through in- • 10:50 - 11:20 Coffee Break
vited papers at IJCAI and ICML. She received her PhD from the Com-
puter Science Department at Stanford University in 2014, and was • 11:20 - 12:30 Session 2 - Generative Text, Images, and Draw-
formerly the Chief Scientist at Apropose, Inc., a data-driven design ing
company she founded that was backed by Andreessen Horowitz and Session Chair: Adam Tauman Kalai
New Enterprise Associates. Her website is http://ranjithakumar.net/,
and she can be reached at ranjitha@illinois.edu. [20 mins] Paper 3: How Novelists Use Generative Language
Adam Tauman Kalai is a Principal Researcher at Microsoft Models: An Exploratory User Study Alex Calderwood, Katy
Research working on machine learning and crowdsourcing. He has Ilonka Gero and Lydia B. Chilton
co-organized the conference on Crowdsourcing and Human Compu-
tation (HCOMP 2017), New England Machine Learning Day (NEML [10 mins] Demo Paper 4: Literary Style Transfer with Con-
2012-2018, with about 300 participants each), the Conference on tent Preservation Katy Gero, Chris Kedzie and Lydia B. Chilton
Learning Theory (COLT 2010), and hackathons on Crowdsourcing
and AI Fairness. He can be reached at adum@microsoft.com. Invited Guest Speakers
[20 mins] Invited Talk 1: Draw with Me: Human-in-the-Loop
3 PROGRAM COMMITTEE for Image Restoration Thomas Weber, Zhiwei Han, Stefan
• Nancy Baym, Microsoft Research Matthes, Yuanting Liu and Heinrich Hussmann
• Zoya Bylinskii, Adobe Research
• Carrie Cai, Google [20 mins] Invited Talk 2: Creative Sketching Partner: An
• Elizabeth Clark, University of Washington Analysis of Human-AI Co-Creativity Pegah Karimi, Jeba
• Sebastian Gehrmann, Harvard School of Engineering Rezwana, Safat Siddiqui, Mary Lou Maher, Nasrin Dehbo-
• Katy Gero, Columbia University zorgi
• Per Ola Kristensson, University of Cambridge
• Jacquelyn Martino, IBM Research AI • 12:30 - 14:00 Lunch
• Mauro Martino, IBM Research AI
• Michael Mateas, University of California, Santa Cruz • 14:00 - 14:45 Visual Human-AI collaboration tools
• Antti Oulasvirta, Aalto University Keynote 2 by Hendrik Strobelt
• Dafna Shahaf, Hebrew University of Jerusalem Session Chair: Ranjitha Kumar
• Akash Srivastava, IBM Research AI
• Hendrik Strobelt, IBM Research AI • 14:45 - 15:05 Session 3 - The Dark Side of Generative Ap-
• Michael Terry, Google proaches
HAI-GEN 2020: Workshop on Human-AI Co-Creation with Generative Models IUI’20 Workshops, March 17, 2020, Cagliari, Italy
Session Chair: Ranjitha Kumar creative process. However, there is a danger that with gener-
ative methods, creators will feel threatened by AI systems,
[20 mins] Paper 5: Business (mis)Use Cases of Generative which in turn might negatively influence productivity.
AI Stephanie Houde, Vera Liao, Jacquelyn Martino, Michael • The relationship between the Human and the AI system
Muller, David Piorkowski, John Richards, Justin Weisz and came up as another emerging topic. In particular, the ques-
Yunfeng Zhang tion when we think about AI as a collaborator versus a tool
we use. One aspect is how this relationship is formed, like
• 15:05 - 15:30 Discussion / Wrap-Up we might form an anthropomorphic relationship with a mu-
sical instrument, another aspect is how the AI system is
5 SUMMARY OF WORKSHOP DISCUSSION represented and how the interaction is modeled. For exam-
This workshop took place virtually on March 17 because of the ple, a conversational interface that is being used during the
global corona virus pandemic in 2020. Despite its virtual nature, the co-creation activity between Human and AI System might
30+ participants had very stimulating discussions after each paper strongly influence how this relationship is perceived by the
and also an extended discussion at the end of the workshop. The human during the creation activity. The perception of the
discussion evolved around tools, ideas, goals and grand challenges AI system may have an influence on the creative output.
that will help us, as a community make progress. The following
topics evolved during this discussion: REFERENCES
[1] [n.d.]. Introduction. https://neurips2019creativity.github.io/
• While generative approaches are an emerging topic in HCI [2] [n.d.]. MIT Quest for Intelligence. http://ganocracy.csail.mit.edu/
communities, for many years there have been existing com- [3] 2019. Making Art in the Age of Algorithms. https://engineering.columbia.edu/
munities around computational creativity (including a con- news/art-age-algorithms
[4] Samantha Cole. 2019. This Deepfake of Mark Zuckerberg Tests Facebook’s Fake
ference series) as well as AI & Creativity events (e.g. NeurIPS Video Policies. https://www.vice.com/en_us/article/ywyxex/deepfake-of-mark-
workshop) at AI events. For us to bring this topic into the zuckerberg-facebook-fake-video-policy
[5] Ian J. Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-
HCI communities, we need to make sure to establish and Farley, Sherjil Ozair, Aaron C. Courville, and Yoshua Bengio. 2014. Generative
maintain connections into the other communities. Adversarial Nets. In NIPS.
• During Doug Eck’s keynote the term fluency came up and he [6] Sepp Hochreiter and Jürgen Schmidhuber. 1997. Long Short-Term Memory.
Neural Comput. 9, 8 (Nov. 1997), 1735–1780. https://doi.org/10.1162/neco.1997.9.
was stating that while our AI systems can achieved fluency 8.1735
now in creating content, our systems still do not understand [7] Tero Karras, Samuli Laine, and Timo Aila. 2018. A Style-Based Generator Archi-
tecture for Generative Adversarial Networks. ArXiv abs/1812.04948 (2018).
intent or can properly deal with the meaning of the content [8] Diederik P. Kingma and Max Welling. 2013. Auto-Encoding Variational Bayes.
created. A major topic for future research is meeting the CoRR abs/1312.6114 (2013).
needs of humans in the creation process and providing the [9] Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, and
Ilya Sutskever. 2018. Language Models are Unsupervised Multitask Learners.
abilities to properly control algorithmic output. (2018). https://d4mucfpksywv.cloudfront.net/better-language-models/language-
• Another topic of interest that evolved during our discus- models.pdf
sion was the sense of ownership of the artifact created and [10] James Vincent. 2019. Listen to this AI voice clone of Bill Gates created by
Facebook’s engineers. https://www.theverge.com/2019/6/10/18659897/ai-voice-
the threat of loosing that sense of ownership. Research has clone-bill-gates-facebook-melnet-speech-generation
shown that the sense of ownership positively influences the