Business (mis)Use Cases of Generative AI Stephanie Houde, Vera Liao, Jacquelyn Martino, Michael Muller David Piorkowski, John Richards, Justin Weisz, Yunfeng Zhang Stephanie.Houde@ibm.com,Vera.Liao@ibm.com,jmartino@us.ibm.com,michael_muller@us.ibm.com David.Piorkowski@ibm.com,ajtr@us.ibm.com,jweisz@us.ibm.com,zhangyun@us.ibm.com IBM Research ABSTRACT of view on factors such as scenario plausibility, seriousness, and Generative AI is a class of machine learning technology that learns prevention. to generate new data from training data. While deep fakes and The contributions of this paper include media-and art-related generative AI breakthroughs have recently • participant-created future scenarios based on generative AI caught people’s attention and imagination, the overall area is in its capabilities in text, audio, and video infancy for business use. Further, little is known about generative • reactions to possible generative AI design fiction scenarios AI’s potential for malicious misuse at large scale. Using co-creation • discussion responses to specific probes on scenario plausibility, design fictions with AI engineers, we explore the plausibility and seriousness, ways things might be worse, ways things might be severity of business misuse cases. better, and prevention The remainder of this paper is organized as follows: Section 2 CCS CONCEPTS discusses related work in AI and Human Centered Data Science (HCDS), and then introduces design fictions as a method for "pro- • Human-centered computing; • Social and professional top- totyping" and discussing possible future outcomes of AI. Section 3 ics → Computing / technology policy; • Applied computing → describes how we used design fictions with experts in AI software Law, social and behavioral sciences; engineering to explore potential business misuse cases. Section 4 presents our results. Section 5 offers concluding thoughts. KEYWORDS Generative AI, design fiction, human-in-the-loop 2 RELATED WORK ACM Reference Format: Strong claims are made about the promises and current successes Stephanie Houde, Vera Liao, Jacquelyn Martino, Michael Muller and David of AI and data science [20, 25, 32, 52]. While some of these claims Piorkowski, John Richards, Justin Weisz, Yunfeng Zhang. 2020. Business are projected for the future [16], Agarwal and Dhar editorialized (mis)Use Cases of Generative AI. In Proceedings of March 17, 2020 (IUI ’20 in 2014 that “This is powerful. . . we are in principle already there” Workshops). ACM, New York, NY, USA, 6 pages. https://doi.org/10.1145/ [1]. Meanwhile to the dystopian extreme, scholars warn about nnnnnnn.nnnnnnn the “mythology” of working with big data, that the quantitative nature of data gives a false illusion that all data-driven outcomes are 1 INTRODUCTION objective, ethical, or true [12]. Increasingly, scholars in the emerging As we think about a future where humans and AI partner in creative field of Human Centered Data Science (HCDS) [3, 11, 33, 55] have activities, we consider how generative models [40, 41, 51, 53, 60] begun to investigate data science practices, showing the necessary, could impact current businesses and possibly create new ones. responsible, and increasingly accountable human activities that While deep fakes and media-and art-related generative AI break- take place between data and models [11, 21, 22, 26, 34, 37–39, 42]. throughs have recently caught people’s attention and imagination In this paper we extend this emerging work by examining possi- [2, 23, 27, 28, 45], the area overall is in its infancy for business ble applications of recent developments in generative AI methods. use. In this paper, we take an inverse approach to business cases Rather than waiting to “see what happens,” we apply the low-cost and propose business “misuse” cases from the “bad actor” point of method of design fictions to prototype [6, 18, 19, 30, 31, 36, 46– view. Using a practice called design fictions [8, 18, 19, 48–50], we 50, 54, 56] possible future applications. Rather than depending on engaged with software engineers who are expert in AI technolo- our own views, we engage with the thoughtful and creative contri- gies. We provided three half-page fictions about possible harmful butions of knowledgeable colleagues [4, 7, 8, 14, 29, 35, 57]. applications of generative technologies as probes for use in our There has been increasing interest in HCI of projecting the future co-creation exercises. As we guided the engineers’ exploration of of technologies through design fictions (DFs) — fictional scenarios in the three fabricated misuse business cases, we learned their points the form of narratives, concepts, prototypes, enactments, and games. [7–9, 14]. DFs have been applied to the design of new technology Copyright © 2020 for this paper by its authors. Use permitted under Creative [15], to explore how future users may adopt a technology [14], and Commons License Attribution 4.0 International (CC BY 4.0). as critical tools to anticipate the social and political consequences of technologies [17, 49]. Design fictions as a methodology spring from several sources. While many cite Sterling’s introductory definition of “deliberate use of diegetic prototypes to suspend disbelief about change” [48], relevant forms have also been explored in the fields of Participatory IUI ’20 Workshops, Cagliari, Italy, Houde et al. Design [5, 10], and Future Studies [36]. A predecessor to DFs in (2) Is this a serious problem? HCI research is design scenarios [59] - simple vignettes to illustrate (3) Can you make the scenario worse? the use of technologies to be developed. DFs sometimes focus on (4) Can you make the scenario better? the design of concrete “near future” technologies [6], constructing (5) Can anything be done to prevent this? a discursive fictional space for speculating about the effects of such (6) Is there a way out of this? technologies in the future. In particular, DFs has been embraced for (7) Who is the right person, organization, or entity to improve revealing values associated with new technologies [35, 49]. it? There are many forms and usages of design fiction. We differen- Participants spent about 15 minutes reading and discussing each tiate between research works that create DFs as the end products scenario for a total time of about 45 minutes. (e.g., to frame new design concepts that do not yet exist [13, 24]), or as critical tools [7, 8]), and those using DFs as probes—"critical nar- 3.2 Scenarios ratives to elicit open-ended responses from potential future users 3.2.1 Scenario A. Instant Author Story. [stakeholders] of proposed technologies” [43]. Part 1. It’s 2020 (and you’ve just seen this popup ad): Hey! I bet Our use of DFs falls in the latter category. Some also refer to it you’ve tried to write the Great American Novel several times but as participatory design fiction, in contrast to previous approaches you can’t seem to get past the first chapter. We get it. Like you, we where DFs are created only by researchers [35]. For example, Shulte find original writing to be really difficult and basically unrewarding. et al. described a 5-step method to create design fictions on the Luckily, we’ve created MakeMyStory.com that takes someone else’s topic of smart houses, and illustrated how the method can be used content, does a deep analysis of it, and emits it in a form that is as for research and design purposes. Recently, several works explored compelling as the original work but different enough that no one using the Story Completion Method (SCM) in design fiction. SCM will say it’s not yours. How cool is that? As an added bonus, your was first introduced in psychotherapy and qualitative research new work can be emitted in any language, increasing the number in psychology [44]. Wood et al. used SCM for speculative stories of countries where it can be sold. to explore the future vision of Virtual Reality pornography [58]. Part 2. It’s 2030 now and Amazon is completely flooded with Cheon and Su introduced Futuristic Autobiographies [14] as a way works by previously unknown writers. A few prominent authors to elicit perspectives on the future of technologies and conducted a have taken MakeMyStory to court to try to have the web site shut case study on the future of robotics. The method starts by posing down. Others have sued the nominal authors of these works but stories involving the participant as a character in a future state, as have not been successful. This is due to a few things. First, Make- grounded in background research work, and invites the participant MyStory has declared their software to be proprietary so it can’t be to complete the autobiography. examined and the courts have gone along with that so far (go figure). Second, the work that is claimed to be derived passes every known 3 METHODOLOGY test for originality. (Interesting side note: perhaps an investigation Our exploratory study into potential business misuse cases began into how the original authors determined that their work had been with the creation of design fictions focusing on the use of gener- stolen might lead to new techniques for automatically finding such ative AI for malicious purposes. We wrote three scenarios, each "adaptations". ) Third, a recently-added REMIX capability allows representing a different generative media type: text, audio, or video. multiple original works to be "blended" making derivation tracking Then, using these scenarios as a springboard, we led co-creation almost impossible. So ... check back in in another couple of years, sessions with AI experts to gain their views on possible worst-case assuming anyone is still bothering to read by then. business scenarios. We conducted these sessions in counterbalanced order across participants, to control for novelty, fatigue, and other 3.2.2 Scenario B. Fake Smoking Evidence Insurance Claim Story. possible order effects. Part 1. It’s 2020 and Jane just accepted a new job. As part of her onboarding process, she is answering some health questions for insurance purposes, including, of course, whether she smokes. That 3.1 Interviews one is simple...never. Later that year, Jane, develops shortness of We conducted co-creation sessions with six participants (2 female, 4 breath, confirmed to be the early stages of emphysema. Confident male) from our research organization. All participants are software that her health insurance will cover the costs she begins some very engineers practiced in AI techniques. After a level-setting descrip- expensive treatment. When the first batch of bills is rejected by the tion of generative techniques as a class of machine learning and insurance company she calls to find out why and is simply shocked. their potential use for malicious intent, we presented our potential Here is the conversation: future scenarios. Each scenario was presented in two parts. The first Representative: I’m sorry Jane, but based on your medical history, part is a design fiction from the near future (2020). After reading you will not be covered for this procedure. Your history of smoking the first part, we asked each participant to speculate on a worst case disqualifies you from coverage. misuse of generative AI in ten more years (2030) and invent their Jane: What are you talking about, I’ve never smoked in my life! own future scenario. After hearing their thoughts, we shared the Representative: Well, I’m looking at some videos online that beg second part which was our version of the future in 2030. With both to differ. Did you attend a wedding last month? our version and their version in mind, we asked them to respond Jane: Yes, it was my best friend’s wedding. But I don’t under- to the following questions: stand... what are you talking about? (1) Is the scenario plausible? Representative: You were wearing a blue dress, correct? IUI ’20 Workshops, Cagliari, Italy, Business (mis)Use Cases of Generative AI Jane: Yes. with participant comments during their invented 2030 futures, fol- Representative: Well ma’am, I’m watching a video from the wed- lowed by their reactions to our envisioned 2030 futures, and ending ding where you are very clearly smoking a cigarette. As such your with their responses to the focused questions. request for coverage has been denied. Jane knows that video had to have been faked, but she has no 4.1 Participant Invented Futures idea how to prove it. Let’s just hope her savings hold out. 4.1.1 Ubiquity. Participants anticipated a significant increase in Part 2. In what seems to be a pattern of corruption, insurance fakes due to reduced costs. Many of these fakes would likely be companies are denying medical claims due to video “evidence” of created by the general public. It was noted that large scale auto- risky behavior ranging from base jumping to travel to active war mated faking might be particularly dangerous since it would spread zones. Car insurance claims are being rejected due to “evidence” of adverse impact beyond what a few bad actors would likely do. Inter- preexisting damage. Even life insurance claims are being denied estingly, the very ubiquity of fakes might also reduce the impact of based on “evidence” that the deceased is still alive. Legislation was any particular fake since less weight would be given to any single passed to make these practices illegal, but without the ability to piece of content. Widespread surveillance, with verifiable prove- detect when a video is faked, and without the ability to truly prove nance, might become more important as a means of countering the legitimacy of a piece of digital content, insurance companies faked evidence, and privacy laws might require revisions as policy- are able to continue these practices to fatten their bottom line. makers learn more about how vulnerable groups could be affected by the proliferation of fakes, and in what ways. Not surprisingly, 3.2.3 Scenario C. Audio Evidence Story. there will be increased need for reliable detection of fakes. Part 1. It’s 2020, and the following story just appeared on page 14 of the New York Times: Expert testimony for the defense was 4.1.2 Quality. When informants were discussing their own partic- presented today in the case of Walter Milgrim, a prominent New ipatory fictions, they suggested that machine generated text may York investor accused of arranging the contract murder of his wife. actually be better than human authored text, at least for some In earlier testimony, the prosecution’s expert witness had played purposes. It was also noted that faked content may actually be an audio recording claimed to have been captured by a passenger acceptable if it is used for entertainment rather than as evidence. on the Chappaqua train platform on the night of the murder. In it, a person sounding like Milgrim and a person sounding like the 4.1.3 Ownership. There was a belief that all work was, in some trigger man could be heard discussing the means of payment. While sense, derived from what came before. This view led to the thought the voices were somewhat muffled, the recording seemed authentic. that automatically derived content was perhaps acceptable. This was cleverly refuted by the defense expert, who commissioned a separate recording at the train station. In that recording, a low 4.1.4 Markets. Niche markets for hand-crafted content might emerge hum was heard due to wind passing through the overhead wires. with “written by humans" sections being tucked in the back of book This hum was not present in the prosecution’s recording, but it stores and online sales sites. A greatly expanded need for digi- was to be expected due to windy conditions that night. Reasonable tal forensics in cases of disputed ownership and other matters of doubt? We’ll see. evidence might cause this to become an emerging business oppor- Part 2. It’s 2030 and a number of criminal convictions have been tunity. thrown into turmoil by the discovery of an audio tool that can synthesize ultra-realistic ambient soundscapes based on thousands 4.1.5 Technology. It was felt that models may be able to detect of parameters. For example, all manner of weather conditions can features of fakes that would not be apparent to humans. be simulated including how those conditions interact with physical features of a location. Seemingly-authentic crowd noises (from the 4.2 Reactions To Our Envisioned Futures faint shuffle of feet on snow to the sounds of people almost having 4.2.1 Ubiquity. People might start creating verified videos of them- to shout over traffic or being nearly drowned out by the arrival of selves to protect against fakes. This is already happening with a train or airplane) add to the realism. A whistleblower has alleged always-on dash cams in both private and police cars. Because of that a number of high profile cases in the Southern District of New the ease of fake creation, multiple independent sources of evidence York relied on recordings generated by this tool. It is unknown how might become even more required. many cases have been corrupted and there seems to be no way to determine which recordings were real and which were, in fact, 4.2.2 Quality. When informants were discussing our probe fictions, generated. they thought that there was a real possibility of a descent into mediocrity as fakes fed on themselves in a “never ending resounding echo chamber of cyber-generated nonsense”. Several thought we 4 RESULTS risked a loss of “art”, both in the creation of and the appreciation Two members of the research team independently reviewed the of content, with creatives having significantly reduced value. It is transcripts and audio recordings of each of the three scenarios in possible this would not be sustained, though, as much generated each of the six sessions, looking for key themes and comments. content might be “boring as hell”. These were then assembled into a combined data set. As no clear differences between the three scenarios were found, we present 4.2.3 Ownership. One participant wondered if AI would have to observations summarized by themes within session phases, starting be acknowledged as a co-creator of content. IUI ’20 Workshops, Cagliari, Italy, Houde et al. 4.2.4 Markets. Creating content for a single consumer would be and watermarking of captured content in personal devices, and new feasible, both for entertainment and education. This would upend legislation that finally caught up with emerging threats were all the current mass distribution market model. mentioned. Education and training were noted by many participants as important; making people aware of the threats posed by fake 4.2.5 Technology. It was thought that there would be a signifi- content and equipping them with skills to detect it might diminish cant ramp up of research in detecting machine-generated content. the viral quality of fakes. Progress here would likely be aided by the fact that detection ca- pability would increase as the corpus of fakes for training grew. 4.3.6 Who Might Improve. Here too, some combination of technol- On the other hand, widespread watermarking and the use of block ogy providers, regulators (especially international organizations), chains might obviate the need for much of this detection capability and legislative bodies were viewed as likely contributors. Special- as we could rely on immutable metadata to prove originality. ized oversight mechanisms might also arise to meet the needs of academic research and publishing, medical research, art competi- 4.3 Answers To Focused Questions tions, and so on. One participant thought that AI itself might assist 4.3.1 Plausibility. There were mixed views on the plausibility of in doing legal research and crafting good law to counter threats. 2030 scenarios. Some felt that similar futures were inevitable due to Finally, consumers and critics might rate and review content for reduced costs of fakes and often malevolent human nature. Others quality and originality in a manner not too dissimilar from today. considered these futures unlikely due to legal, regulatory, and mar- ket forces, the emergence of technical detection and provenance 5 CONCLUDING THOUGHTS mechanisms, and the belief that societies have generally been good at authenticating things they care about (e.g., money, passports). Our participants’ views were both thoughtful and highly varied. Even though their comments stayed at a quite general level, they 4.3.2 Seriousness. There were also mixed views on the seriousness sometimes came to contrasting conclusions when discussing their of these possible futures; if a future was considered implausible for own future scenarios as opposed to the future scenarios that we the reasons noted above it was also viewed as not serious. On the created (particularly regarding quality issues). These types of con- other hand, it was mentioned that a single blemish in a person’s trasts in perspectives have been important in other projects with health record could profoundly impact the ability to obtain insur- participatory forms of design fictions [4, 7, 8, 14, 29, 35, 57]. Taken ance coverage. Thus, quite serious outcomes might arise from even together, these contrasting results suggest that including informants isolated incidents of faking. in collaborative interpretive spaces can be a powerful method for 4.3.3 How It Could Be Worse. One participant felt that genera- increasing the knowledge of participants and researchers alike. tive models interactively posing as humans might be particularly Some informants thought scenarios of the sort we sketched for harmful. We already see low-tech variants of this with, for example, 2030 were inevitable while others thought they were unlikely. Draw- people posing as somewhat-distant relatives needing money to ing on their own expertise, their thoughts on how technology might get them out of difficulty during foreign travel. The synthesis of counter possible threats were more nuanced than their thoughts realistic human speech was considered harder than the synthesis about regulation and legislation. Perhaps the most interesting dy- of ambient audio, but the potential impact could be much worse. namics arose in connection with what many saw as an "arms race" On a different note, the coherent faking of multiple sources might between generation and detection capabilities. There was optimism further complicate use as evidence. One possible reaction to all this that the coming ubiquity of fakes will lead, in a manner not unlike would be governments banning the use of generative AI. If only a natural ecosystem, to an abundance of training data for high- governments were allowed to use it, the results could be quite dire. quality detectors. That, along with progress in digital provenance Finally, if these sorts of futures come to pass, people might simply and updated regulations and legislation, may prevent the worst of stop caring about truth altogether. these future scenarios from coming to pass. It will be important to expand these HCDS approaches [3, 11, 33, 55] to include more 4.3.4 How It Could Be Better. Several potentially beneficial uses of diverse informants in the future, for a more societally-grounded generative AI were offered. In one, the AI could generate realistic vision and critique of the implications of generative AI. time-aged portraits of lost loved ones to convey the sense that they were still present in peoples’ lives. In another, high-quality gen- REFERENCES erated speech could be used by those with severe communication [1] Ritu Agarwal and Vasant Dhar. 2014. Big data, data science, and analytics: The disabilities. Related to this, a generative AI might align the sounds opportunity and challenge for IS research. of diverse human accents in an audio stream for improved mutual [2] Sakshi Agarwal and Lav R Varshney. 2019. Limits of Deepfake Detection: A Robust Estimation Viewpoint. arXiv preprint arXiv:1905.03493 (2019). intelligibility when time was critical (e.g., in an emergency call). [3] Cecilia Aragon, Clayton Hutto, Andy Echenique, Brittany Fiore-Gartland, Yun Finally, the ability to cheaply produce content could provide value Huang, Jinyoung Kim, Gina Neff, Wanli Xing, and Joseph Bayer. 2016. Developing to currently under-served markets. a research agenda for human-centered data science. In Proceedings of the 19th ACM Conference on Computer Supported Cooperative Work and Social Computing Companion. ACM, 529–535. 4.3.5 How It Could Be Prevented. Answers to the "Prevention" and [4] Eric PS Baumer, June Ahn, Mei Bie, Elizabeth M Bonsignore, Ahmet Börütecene, "Way Out" questions were quite similar and are combined here. Oğuz Turan Buruk, Tamara Clegg, Allison Druin, Florian Echtler, Dan Gruen, Most felt that some combination of technology, regulation, and et al. 2014. CHI 2039: speculative research visions. In CHI’14 Extended Abstracts on Human Factors in Computing Systems. ACM, 761–770. legislation could prevent the worst of these future scenarios. Data [5] Ian Beeson and Clodagh Miskelly. 2000. Dialogue and dissent in stories of traceability and auditability, ubiquitous detection of fake content community. In Proc. PDC 2000. Citeseer. IUI ’20 Workshops, Cagliari, Italy, Business (mis)Use Cases of Generative AI [6] J Bleecker. 2009. Design Fiction: A Short Essay. Available at: ACM, 3–3. http://nearfuturelaboratory. com/2009/03/17/design-fiction-a-short-essay-on-design- [33] Michael Muller, Melanie Feinberg, Timothy George, Steven J Jackson, Bonnie E science-fact-andfiction/[date accessed: 6th Jan 2017] (2009). John, Mary Beth Kery, and Samir Passi. 2019. Human-Centered Study of Data [7] Mark Blythe and Elizabeth Buie. 2014. Chatbots of the gods: imaginary abstracts Science Work Practices. In Extended Abstracts of the 2019 CHI Conference on for techno-spirituality research. In Proceedings of the 8th Nordic Conference on Human Factors in Computing Systems. ACM, W15. Human-Computer Interaction: Fun, Fast, Foundational. ACM, 227–236. [34] Michael Muller, Ingrid Lange, Dakuo Wang, David Piorkowski, Jason Tsay, Q Vera [8] Mark Blythe, Enrique Encinas, Jofish Kaye, Miriam Lueck Avery, Rob McCabe, and Liao, Casey Dugan, and Thomas Erickson. 2019. How Data Science Workers Kristina Andersen. 2018. Imaginary design workbooks: Constructive criticism Work with Data: Discovery, Capture, Curation, Design, Creation. In Proceedings and practical provocation. In Proceedings of the 2018 CHI Conference on Human of the 2019 CHI Conference on Human Factors in Computing Systems. ACM, 126. Factors in Computing Systems. ACM, 233. [35] Michael Muller and Q Vera Liao. 2017. Exploring AI Ethics and Values through [9] Mark Blythe, Jamie Steane, Jenny Roe, and Caroline Oliver. 2015. Solutionism, Participatory Design Fictions. Human Computer Interaction Consortium (2017). the game: design fictions for positive aging. In Proceedings of the 33rd Annual [36] Daniel Pargman, Elina Eriksson, Mattias Höjer, Ulrika Gunnarsson Östling, and ACM Conference on Human Factors in Computing Systems. ACM, 3849–3858. Luciane Aguiar Borges. 2017. The (un) sustainability of imagined future infor- [10] Susanne Bødker and Kaj Grønbæk. 1991. Design in action: From prototyping by mation societies. In Proceedings of the 2017 CHI Conference on Human Factors in demonstration to cooperative prototyping. In Design at work: Cooperative design Computing Systems. ACM, 773–785. of computer systems. Lawrence Erlbaum Associates, 197–218. [37] Samir Passi and Steven Jackson. 2017. Data vision: Learning to see through [11] Nadia Boukhelifa, Anastasia Bezerianos, Ioan Cristian Trelea, Nathalie Méjean algorithmic abstraction. In Proceedings of the 2017 ACM Conference on Computer Perrot, and Evelyne Lutton. 2019. An Exploratory Study on Visual Exploration Supported Cooperative Work and Social Computing. ACM, 2436–2447. of Model Simulations by Multiple Types of Experts. In Proceedings of the 2019 [38] Samir Passi and Steven J Jackson. 2018. Trust in Data Science: Collaboration, CHI Conference on Human Factors in Computing Systems. ACM, 644. Translation, and Accountability in Corporate Data Science Projects. Proceedings [12] Danah Boyd and Kate Crawford. 2012. Critical questions for big data: Provo- of the ACM on Human-Computer Interaction 2, CSCW (2018), 136. cations for a cultural, technological, and scholarly phenomenon. Information, [39] Kathleen H Pine and Max Liboiron. 2015. The politics of measurement and action. communication & society 15, 5 (2012), 662–679. In Proceedings of the 33rd Annual ACM Conference on Human Factors in Computing [13] Barry Brown, Julian Bleecker, Marco D’adamo, Pedro Ferreira, Joakim Formo, Systems. ACM, 3147–3156. Mareike Glöss, Maria Holm, Kristina Höök, Eva-Carin Banka Johnson, Emil [40] Alec Radford, Luke Metz, and Soumith Chintala. 2015. Unsupervised Represen- Kaburuan, et al. 2016. The IKEA Catalogue: Design fiction in academic and tation Learning with Deep Convolutional Generative Adversarial Networks. industrial collaborations. In Proceedings of the 19th International Conference on arXiv e-prints, Article arXiv:1511.06434 (Nov 2015), arXiv:1511.06434 pages. Supporting Group Work. ACM, 335–344. arXiv:cs.LG/1511.06434 [14] EunJeong Cheon and Norman Makoto Su. 2018. Futuristic autobiographies: [41] Alec Radford, Jeff Wu, Rewon Child, David Luan, Dario Amodei, and Ilya Weaving participant narratives to elicit values around robots. In Proceedings of Sutskever. 2019. Language Models are Unsupervised Multitask Learners. (2019). the 2018 ACM/IEEE International Conference on Human-Robot Interaction. ACM, [42] Tye Rattenbury, Joseph M Hellerstein, Jeffrey Heer, Sean Kandel, and Connor Car- 388–397. reras. 2017. Principles of data wrangling: Practical techniques for data preparation. [15] Nazli Cila, Iskander Smit, Elisa Giaccardi, and Ben Kröse. 2017. Products as " O’Reilly Media, Inc.". agents: metaphors for designing the products of the IoT age. In Proceedings of the [43] Britta F Schulte, Paul Marshall, and Anna L Cox. 2016. Homes for life: a design 2017 CHI Conference on Human Factors in Computing Systems. ACM, 448–459. fiction probe. In Proceedings of the 9th nordic conference on human-computer [16] Ciprian Dobre and Fatos Xhafa. 2014. Intelligent services for big data science. interaction. ACM, 80. Future Generation Computer Systems 37 (2014), 267–281. [44] Iduna Shah-Beckley, Victoria Clarke, and Zoe Thomas. 2018. Therapists’ and [17] Paul Dourish and Genevieve Bell. 2014. Resistance is futile: reading science non-therapists’ constructions of heterosex: A qualitative story completion study. fiction alongside ubiquitous computing. Personal and Ubiquitous Computing 18, Psychology and Psychotherapy: Theory, Research and Practice (2018). 4 (2014), 769–778. [45] Yevhenii SHTEFANIUK, Ivan OPIRSKYY, Ihor IVANCHENKO, and Kateryna [18] Anthony Dunne and Fiona Raby. 2001. Design noir: The secret life of electronic PINDEL. [n.d.]. DEEPFAKE–NEW TECHNOLOGY FOR IMPERSONATION IN objects. Springer Science & Business Media. CYBERATTACKS. ([n. d.]). [19] Anthony Dunne and Fiona Raby. 2013. Speculative everything: design, fiction, and [46] Michael Skirpan and Casey Fiesler. 2018. Ad empathy: A design fiction. In social dreaming. MIT press. Proceedings of the 2018 ACM Conference on Supporting Groupwork. ACM, 267– [20] Hugh Durrant-Whyte. 2015. Data, knowledge and discovery: Machine learning 273. meets natural science. In Proceedings of the 21th ACM SIGKDD International [47] Marie Louise Juul Søndergaard and Lone Koefoed Hansen. 2016. PeriodShare: Conference on Knowledge Discovery and Data Mining. ACM, 7–7. A Bloody Design Fiction. In Proceedings of the 9th Nordic Conference on Human- [21] Melanie Feinberg. 2017. A design perspective on data. In Proceedings of the 2017 Computer Interaction. ACM, 113. CHI Conference on Human Factors in Computing Systems. ACM, 2952–2963. [48] Bruce Sterling. 2009. Design fiction. interactions 16, 3 (2009), 20–24. [22] Lisa Gitelman. 2013. Raw data is an oxymoron. MIT press. [49] Joshua Tanenbaum. 2014. Design fictional interactions: why HCI should care [23] David Güera and Edward J Delp. 2018. Deepfake video detection using recurrent about stories. interactions 21, 5 (2014), 22–23. neural networks. In 2018 15th IEEE International Conference on Advanced Video [50] Joshua Tanenbaum, Marcel Pufal, and Karen Tanenbaum. 2016. The limits of our and Signal Based Surveillance (AVSS). IEEE, 1–6. imagination: design fiction as a strategy for engaging with dystopian futures. In [24] Felix Heibeck, Alexis Hope, and Julie Legault. 2014. Sensory fiction: a design Proceedings of the Second Workshop on Computing within Limits. ACM, 10. fiction of emotional computation. In Proceedings of the 2nd ACM International [51] V Tsourikov. 2017. Generative inventions. In TRIZfest, Conference Proceedings. Workshop on Immersive Media Experiences. ACM, 35–40. 178–184. [25] Anthony JG Hey, Stewart Tansley, Kristin M Tolle, et al. 2009. The fourth paradigm: [52] Wil MP Van der Aalst. 2014. Data scientist: The engineer of the future. In data-intensive scientific discovery. Vol. 1. Microsoft research Redmond, WA. Enterprise interoperability VI. Springer, 13–26. [26] Sean Kandel, Andreas Paepcke, Joseph Hellerstein, and Jeffrey Heer. 2011. Wran- [53] Tijn van der Zant, Matthijs Kouw, and Lambert Schomaker. 2013. Generative gler: Interactive visual specification of data transformation scripts. In Proceedings artificial intelligence. In Philosophy and Theory of Artificial Intelligence. Springer, of the SIGCHI Conference on Human Factors in Computing Systems. ACM, 3363– 107–120. 3372. [54] Ron Wakkary, William Odom, Sabrina Hauser, Garnet Hertz, and Henry Lin. [27] Pavel Korshunov and Sébastien Marcel. 2019. Vulnerability assessment and detec- 2015. Material speculation: actual artifacts for critical inquiry. In Proceedings of tion of deepfake videos. In The 12th IAPR International Conference on Biometrics The Fifth Decennial Aarhus Conference on Critical Alternatives. Aarhus University (ICB). 1–6. Press, 97–108. [28] Yuezun Li and Siwei Lyu. 2018. Exposing deepfake videos by detecting face [55] Dakuo Wang, Justin D Weisz, Michael Muller, Parikshit Ram, Werner Geyer, Casey warping artifacts. arXiv preprint arXiv:1811.00656 2 (2018). Dugan, Yla Tausczik, Horst Samulowitz, and Alexander Gray. 2019. Human-AI [29] Ann Light. 2011. Democratising technology: making transformation using design- Collaboration in Data Science: Exploring Data Scientists’ Perceptions of Auto- ing, performance and props. In Proceedings of the SIGCHI Conference on Human mated AI. Proceedings of the ACM on Human-Computer Interaction 3, CSCW Factors in Computing Systems. ACM, 2239–2242. (2019), 211. [30] Joseph Lindley. 2015. Researching Design Fiction With Design Fiction. In Pro- [56] Richmond Y Wong and Deirdre K Mulligan. 2016. When a product is still fictional: ceedings of the 2015 ACM SIGCHI Conference on Creativity and Cognition. ACM, anticipating and speculating futures through concept videos. In Proceedings of 325–326. the 2016 ACM Conference on Designing Interactive Systems. ACM, 121–133. [31] Joseph Lindley and Robert Potts. 2014. A machine learning: an example of HCI [57] Richmond Y Wong, Deirdre K Mulligan, Ellen Van Wyk, James Pierce, and John prototyping with design fiction. In Proceedings of the 8th Nordic Conference on Chuang. 2017. Eliciting values reflections by engaging privacy futures using Human-Computer Interaction: Fun, Fast, Foundational. ACM, 1081–1084. design workbooks. Proceedings of the ACM on Human-Computer Interaction 1, [32] Renée J Miller. 2017. The future of data integration. In Proceedings of the 23rd CSCW (2017), 111. ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. IUI ’20 Workshops, Cagliari, Italy, Houde et al. [58] Matthew Wood, Gavin Wood, and Madeline Balaam. 2017. They’re Just Tixel Pits, [59] David Wright, Rachel Finn, Raphael Gellert, Serge Gutwirth, Philip Schütz, Man: Disputing the’Reality’of Virtual Reality Pornography through the Story Michael Friedewald, Silvia Venier, and Emilio Mordini. 2014. Ethical dilemma Completion Method. In Proceedings of the 2017 CHI Conference on Human Factors scenarios and emerging technologies. Technological Forecasting and Social Change in Computing Systems. ACM, 5439–5451. 87 (2014), 325–336. [60] Jonathan L Zittrain. 2005. The generative internet. Harv. L. Rev. 119 (2005), 1975.