Accepted Talks

Session 1: Humanizing AI

Toward Trustworthy Chatbots Through AI-Generated Characters

Yucheng Jin, Hong Kong Baptist University; Zhaolin Lu, Beijing Institute of Technology


Despite the proliferation of chatbots, a lack of trust negatively influences users’ engagement and satisfaction with chatbots. A recent study has shown that chatbot character design plays an important role in user trust toward chatbots. Existing research in human-AI interaction has demonstrated the importance of anthropomorphic cues for building trust in chatbots. Most chatbots employ a “one-size-fits-all” approach to offer a uniform design of characters to all kinds of chatbot users. However, prior work has shown the individual differences in users’ anthropomorphism needs and preferences of demographic characteristics (e.g., age, gender, and ethnicity). Therefore, we propose to leverage deepfakes to generate chatbot characters according to task types and user characteristics. We will briefly envisage our approach and discuss the potential benefits to chatbot users.

Deepfakes and AR - A partnership to build empathy

Nina Lyons, TU Dublin


Your content goes here. Edit or remove this text inline or in the module Content settings. You can also style every aspect of this content in the module Design settings and even apply custom CSS to this text in the module Advanced settings.

Breakup Simulator: Reducing the Friction of Difficult Conversations with Deepfakes

Maggie Chen, Harvard GSD Noah X. Deutsch, Harvard GSD/SEAS Erica Luzzi, Harvard GSD/SEAS


The rise of digital communication platforms has made it easier than ever for people to connect in new ways; however, digital interactions lack the intimacy and authenticity of in-person interactions. Recently, we have seen an epidemic of ghosting: when a person suddenly cuts off all communication with someone. Ghosting is a lose-lose situation – the experience may be painful, confusing, and frustrating for both parties. Our solution is the Breakup Simulator, a web-based platform that uses deepfakes to create an interactive breakup conversation between the user and their partner. The tool eliminates the fear of the unknown and eases anxieties about the difficult conversation of a breakup. Harnessing machine learning models including the First Order Model for Image Animation, we animate user images to create a realistic, interactive conversation. We hope that this platform will mitigate ghosting by alleviating the anxiety surrounding difficult conversations through rehearsal.

Session 2: Technical Advancements

Using Augmented Face Images to Improve Facial Recognition Tasks

Shuo Cheng, ByteDance; Guoxian Song, ByteDance; Wanchun Ma, ByteDance; Chao Wang, ByteDance; Linjie Luo, ByteDance


We present a framework that uses GAN-augmented images to complement certain specific attributes, usually underrepresented, for ML training. This allows us to improve inference quality over those attributes for the facial recognition tasks.


Generative 3D Animation Pipelines: Automating Facial Retargeting Workflows

Julius Girbig, LMU Munich; Changkun Ou, LMU Munich; Sylvia Rothe, LMU Munich


We discussed use cases for deepfakes within the 3D content creation domain and prototyped a tool for 3D facial retargeting as a case study. Our interviews with experts in this field were unexpectedly positive regarding our tool, which utilizes a deepfake algorithm.


Session 3: Positive Applications for AI-generated Characters

AI as a messenger to transport the message through out time

Pink Siripipat, Deus Works


AI as a messenger to transport the message through out time like the “Message in the bottle” No matter times goes by, the memories, knowledges and love messages will be here forever and for every generations. AI messenger can bring someone back from the death not only their thought but also their voice and their attitude. Imagine talking with your grand mother in AI version How many cooking recipe that she would love to share? Although negative comments about “Deep Fake” and “Digital Right” is the hot issue for all industries, in this Perspective the AI messenger came from the good intention to slove the real painpoint for all human and create the new envoriment in supporting learning and well being.

Designing for Assistive Self-Visualization

Kate Glazko, National Coalition of Independent Scholars; Yiwei Zheng, USC


Aphantasia is a visualization disorder that affects around 3.9% of people. Individuals with aphantasia are either wholly or partially unable to generate mental pictures, making on-demand visualization difficult or impossible. Self-visualization exercises, such as guided meditations, have been shown to help individuals improve their mental health and achieve their goals. However, self-visualization remains inaccessible to individuals who can’t visualize. In this paper, we propose an approach for designing assistive self-visualization tools and detail our preparations for our future experiment.


Txt2Vid: Ultra-Low Bitrate Compression of Talking-Head Videos via Text

Pulkit Tandon, Stanford University; Shubham Chandak, Stanford University; Pat Pataranutaporn, MIT Media Lab; Yimeng Liu, UCSB;  Anesu M. Mapuranga, Stanford University; Pattie Maes, MIT Media Lab; Tsachy Weissman, Stanford University; Misha Sra, UCSB


Video represents the majority of internet traffic today, driving a continual race between the generation of higher quality content, transmission of larger file sizes, and the development of network infrastructure. In addition, the recent COVID-19 pandemic fueled a surge in the use of video conferencing tools. Since videos take up considerable bandwidth (~100 Kbps to a few Mbps), improved video compression can have a substantial impact on network performance for live and pre-recorded content, providing broader access to multimedia content worldwide. We present a novel video compression pipeline, called Txt2Vid, which dramatically reduces data transmission rates by compressing webcam videos (“talking-head videos”) to a text transcript. The text is transmitted and decoded into a realistic reconstruction of the original video using recent advances in deep learning based voice cloning and lip syncing models. Our generative pipeline achieves two to three orders of magnitude reduction in the bitrate as compared to the standard audio-video codecs (encoders-decoders), while maintaining equivalent Quality-of-Experience based on a subjective evaluation by users (n = 242) in an online study. The Txt2Vid framework opens up the potential for creating novel applications such as enabling audio-video communication during poor internet connectivity, or in remote terrains with limited bandwidth.The code for this work is available at this https URL.

Session 4: Ethics & Social Impact

Are deepfakes concerning?

Dilrukshi Gamage , Tokyo Institute of Technology; Kazutoshi Sasahara, Tokyo Insitute of Technology


Deepfakes are synthetic content generated using advanced deep learning and AI technologies. The advancement of technology has created opportunities for anyone to create and share deepfakes much easier. This may lead to societal concerns based on how communities engage with it. However, there is limited research available to understand how communities perceive deepfakes. We examined deepfake conversations on Reddit from 2018 to 2021—including major topics and their temporal changes as well as implications of these conversations. Using a mixed-method approach—topic modeling and qualitative coding, we found 6,638 posts and 86,425 comments discussing concerns of the believable nature of deepfakes and how platforms moderate them. We also found Reddit conversations to be pro-deepfake and building a community that supports creating and sharing deepfake artifacts and building a marketplace regardless of the consequences. Possible implications derived from qualitative codes indicate that deepfake conversations raise societal concerns. We propose that there are implications for Human Computer Interaction (HCI) to mitigate the harm created from deepfakes.

Voices of Tomorrow: Augmenting refugee resettlement decision-making with Metahumans.

Claude P. R. Heath, Royal Holloway University of London; Matt Falla, parallel systems; Lizzie Coles-Kemp, Royal Holloway University of London;


Machines are making decisions today that will have a profound impact on people’s lives tomorrow. All too often, automated systems are opaque, with little explanation of how they work and how they will impact on people’s lives. Such systems often represent people reductively, without addressing what is important to them. Refugee resettlement system design is one area where algorithmic automation is beginning to play a limited role, although without representation of refugee preferences. For example, in the US, refugee resettlement agency HIAS has successfully trialled an algorithmic approach optimising the chances of finding employment within a six-month period, showing that short-term policy goals can be achieved with the help of automation even when based on sparse data. In the UK, however, the process of finding locations for the resettlement of refugees remains a manual procedure aimed at meeting acute housing and care needs via a complex, painstaking and time-consuming series of steps. These complex decision-making processes requires multiple stakeholders to agree upon optimal matching of individual cases with particular resettlement locations. Voices of Tomorrow is a speculative design initiative designed to probe the potential for algorithmic interaction within the complex ecosystem of resettlement decision-making, and assess what role this might have amongst the situated knowledge and practice of resettlement professionals. The designs are the result of a multidisciplinary collaboration, and use a mix of storytelling, augmented data analytics, agent-based modelling, and the simulated human face and voice of Unreal’s Metahuman toolkit, to imagine a future refugee resettlement decision-making system. The paper describes how these creative technologies might help decision-makers establish which conditions will allow the best possible outcomes for both refugees and their host communities, and to examine the different futures stemming from different present policies, thus helping to advance this complex and disputed area of policy implementation.

The Value Chain for any AI 4 Good Project and the Law as Gatekeepers for Commercialization

Albert Domingo, Universitat Pompeu Fabra; Paz Soler, Universitat Pompeu Fabra; Brisa Burriel, Universitat Pompeu Fabra


Analyze the positive applications of AI-Generated characters by providing certain examples that could contribute to describe and conceptualize what the future is already bringing for our society. Special point is made in the Business Model and the related legal structure (Data and IP), that will apply to the value chain. Determining the value of this Technology for the different stakeholders is crucial for it to deploy its full potential. The interests of the stakeholder’s interests have to be fulfilled. For that to be feasible, the proposal is a solution consisting of the definition of a new social contract with a balancing approach, based on the principles of proportionality, the preservation of data owners’ dignity, and containing a fiduciary duty to one another, which provides a win-win system food for the society as a whole.