potential futures. instance. Using design fiction, we develop a series of possible generative AI features and applications that could be developed in the future of humans' roles in software engineering. We use the fiction to highlight choices and value-tensions among these Generative AI; Software engineering; Workplace; Design fictions; Code-as-instance; Activity-as-instance; Employee-asCEUR
work in software engineering [ 1, 2, 3, 4, 5, 6, 7, 8, 9].
ing - and increasingly diverse - potential risks of applying generative AI to human work and human outcomes [10, 11, 12, 13, 14, 15]. In a paper at the 2020
problems with generative AI through the use of participatory design fictions [ 10]. In those three fictions, we invited our colleagues to speculate on possible societal harms from generative AI applications.
Here, we shift our strategy toward workplaces, and we try to take a more balanced view, considering both potential benefits and potential risks of generative applications in workplaces of the future. Workplace adaptations due to the COVID pandemic have accelerated sociotechnical trends of changed work-practices and changed technolognEvelop-O (J. D. Weisz)
0000-0001-7860-163X (M. Muller); 0000-0002-2533-9946 (S. Ross); 0000-0002-0246-2183 (S. Houde); 0000-0002-8442-2651 (M. Agarwal); 0000-0001-7172-4805 (F. Martinez); 0000-0001-8489-2170 (J. Richards); 0000-0002-4628-3785 (K. Talamadupula); 0000-0003-2228-2398 (J. D. Weisz) and tional needs • an interrelated set of speculations regarding future generative AI applications in workplaces, • a consideration of some value tensions that may emerge between employee needs and organiza
describe possible ways to complete the work in the near future.
plications is their ability to create instances from a ”learned” class of examples, including projects that involve images [18, 19, 20], videos [21], music [22, 23], molecules [24, 25], texts of many types [26, 27, 28, 29, 30], and diverse other media and categories (e.g., [31, 32, 33]).
patterns can function as predictions of (e.g.) the next word in a text or a software program; this ”next-insequence” aspect is a major focus of this paper, here futures for their human implications and impacts [64, 65]. extended to the ”next” user actions. We hope that design fictions can contribute to the mixed
While there is not space in this position paper to review initiative discussions mentioned earlier, focusing attenthe rich history of generative algorithms, we note briefly tion the types of relationships that we envision between that many reports describe generative AI applications humans and AIs [47, 66]. with both potential benefits [ 34, 35, 36] and potential Design fictions may take several forms, including text, risks [37, 38, 39, 14, 12, 10] to individuals or societies. images, film, video, theatre, and physical objects [ 67, Our team have been studying how generative application 68, 69, 70, 71, 17], and may be particularized into recogscan provide significant value to software engineers [ 26, nizable genres such as fictional job adverts [ 72], enact8, 28, 40], and our IUI 2022 paper reports measurable ments [73], technology probes [74, 75, 76], product catasuccesses [27]. In this position paper, we ask what types logs [74], autobiographies [77], and even clearly-fictitious of instances might be created in the future, and whether conference papers [78, 79]. More familiarly, design ficorganizations or societies may wish to place limits on tions can take the form of stories [10, 66, 80, 81, 82, 83, 84, the generation of certain types of instances. 85, 86, 71, 87, 88, 89]. We adopt the method of telling a
In the history of theorizing mixed-initiative dynam- story, through which we explore trends and possibilities ics between humans and AIs, most scholars have imag- at the intersection of technology, skilled human work, ined batch-like processes in which one party (human and values.1 or AI) takes a first major step, and the second party (AI or human) completes the work through a second major step [41, 42, 43, 44, 45, 46, 47, 48]. While Conversational 3. Design Fiction User Interfaces (CUIs) provide limited forms of on-going interactions, the sheer computational requirements of 3.1. Crafting the Design Fiction generative AI algorithms have made it dificult to de- Design fiction scholars emphasize that a fiction should sign and build generative dialogic applications until re- provide a perceptual bridge [90] between the reader and cently (e.g., [49, 50, 51]). An important problem in these novel concepts that may be challenging [64, 91] or even dialogs will be how to control the generative AI execution upsetting [74, 66, 10]. The temporary suspension of disfrom moment-to-moment - i.e., how to ”tune” [52, 53] or belief is considered important [92, 90], and this is accom”steer” [22, 23] the algorithm toward outcomes of value plished through narrative integrity and consistency of to their human partners. We speculate on how possible the imagined world [85, 93] and through empathy with control-parameters might be provided to human users, its protagonists [94, 95]. Auger states that ”careful manand how particular values of those parameters might agement of the speculation” is important, because ”if it be recommended through content-based or social-based strays too far... the audience will not relate to the proposal algorithms. resulting in a lack of engagement or connection” [90]. Design fictions use methods such as diegesis (the creation 2.2. Design Fictions of a story-world) to communicate new possibilities and to discuss their consequences [96, 97, 16].
For more than a decade, scholars have used design fic- We apply these principles by setting our story in a rections as a core research method [16, 17]. Based on the- ognizable software engineering workplace - albeit with orizing by Peirce [54] over a century ago, Dunne and futuristic tools and the gradual unveiling of an unsettling Martin summarized the potential of design fictions as workplace culture. To avoid aspects of the ”uncanny abductive methods, stating that valley” that can occur when AI agents behave in eerily human ways [98, 15], we designed the conversational style of the generative AI assistant - ”Your Programming Partner (YPP)” - to be recognizably non-human - even robotic in some ways. “The designers who can solve the most wicked problems do it through collaborative integrative thinking, using abductive logic, which means the logic of what might be. Conversely, deductive and inductive logic are the logic of what should be or what is… [55] (italics added)
Abductive reasoning has also been claimed as a major intentions, it is also important that they act as cultool for creatively building theory in approaches such as grounded theory [56, 57, 58, 59] and thematic analy- 1Methodological treatments of how to approach or write design sis [60, 61]. We apply future-oriented design fictions to if[c1ti7o]n,sHmuuasykobeetfoaul.nhduiunskMoa2r0k1u8ssstreuncatunrdinKgn,uBtlyzt[he1a6n],dSWturridgehet e[6t5a]l,. extrapolate current phenomena and trends into possible and Cheon and Su [77]; see also Blythe and Encinas [68] and Baumer futures [62, 63] and to begin to interrogate those possible et al. [78] for approaches to assessing or evaluating design fictions. tural probes [99], actively engaging the reader to form their own interpretations and to draw their own conclusions [100, 101, 75, 16, 102]. Coulton et al. note that ”it seems that Design Fiction has ambiguity ‘baked in’ ” [103], and researchers have tried to balance among diegetic factors such as familiarity and consistency, vs. the ambiguity that is needed to encourage new (e.g., abductive [55]) thinking among their readers [66, 68, 104].
The strategy of ambiguity can be particularly important if the fiction addresses value tensions [ 105, 84, 80].
Ambe [83] and Huusko [64] argue that many technology applications may have both utopian and dystopian implications, and these implications may be diferent for diferent stakeholders [106, 107]. Feminist technoscience convergently urges us to consider each person’s perspective, and to question power diferentials [ 108, 109, 110, 111].
We noted above that design fictions may be designed to raise challenging or upsetting questions. Using a formulation from Haraway’s feminist theorizing [112], Sondergaard et al. suggest that we need to ”stay with the trouble” of value tensions in AI: ”Might we allow the [AI assistant] to be not just good or bad, submissive or dominant, but a complicated, contradictory being?... The future world serves as a projection of current issues and conflicts, and thus the future becomes a way of looking at ourselves and our culture.” [70]
We tried to apply these rather open-ended concepts by deliberately leaving certain details unspecified, including the gender-identities of the actors, the nature of the actors in certain workplace roles, and the reasons for the loneliness that is experienced by the protagonist. We also presented certain potential value tensions in a relatively neutral way, to emphasize the questions rather than to impose premature closure on complex ethical topics.
For brevity, we now proceed directly to the design ifction. Drinking Chai with Your (AI) Programming Partner <1> Tikaani had been postponing work to translate the UX of the enormous BiggerFin application that their team was modernizing from its legacy Cobol code. So much had changed in the decades since BiggerFin was originally deployed. Everyone who had been on the original BiggerFin team had retired or moved on to other jobs.
Generative translation technologies had become increasing accurate to convert the back-end code, but modernizing the UI was still a challenge for AIs. The UI remained a challenge because web-based UI technologies were now much better, and corporations’ expectations for what was considered ”good design” had also changed. Today was the day to do this! And it would distract Tikaani from the loneliness of the ofice, with so many unoccupied desks. Fortunately, Tikaani had a conversational generative AI (genAI) assistant that could help, named ”Your Programming Partner” (YPP). <2> Tikaani opened the spec from the design team, and called up YPP.
YPP displayed ”Greetings follow. Good morning, Tikaani. Today we will have a good day, a very good day,” and followed with ”Advertisement follows. Try BerryBytes in the ByteBar. They’re brainfood.
They can reduce any unwanted aftertastes...” <3> During the previous modernization project, Tikaani had told YPP that its nickname was going to be ”Y”. Tikaani pressed the speech-totext key, and said ”Y, please review the spec,” and made a gesture to tell YPP which document was the spec. <4> “Drink chai,” said YPP. Tikaani took a sip of chai. <5> After a brief pause, YPP displayed, “To perform a generative translation from spec to architecture. a set of examples is required. Recommendations follow: Finance/big-institution; Finance/small-institution; Corporate; SMB; or say ‘other’ for non-recommended domains.” Each recommendation was displayed as a selectable button, so that Tikaani’s choice would immediately lead to action by YPP.
ducing the likelihood of uncanny valley efects [ 113, 114].
Messages from multiple entities may be delivered through the YPP conversational interface.
<6> While it was true that BiggerFin was indeed a large financial
client and would have been a partial match to “Finance/big-institution,”
the clients of BiggerFin for this service would be people and small
businesses seeking micro-loans. Tikaani wanted to use a more social
search strategy to make their choice. There was no one nearby to
ask, so Tikaani typed, ”Display team members along with the number
of financial projects they have worked on.” YPP produced the list,
again making each row of the displayed table into a button-for-action.
Tikaani saw that Yu had worked on 20 similar projects. Yu had been
Tikaani’s mentor when they joined the company, and Tikaani missed
the days when they had traded ironic comments over tea. Tikaani
gestured to the “Yu – Finance projects” row, and YPP accepted this
gesture as a choice+command, and began to work.
<7> YPP displayed, “Further guidance is required. Recommendations
follow. (
display functions as both informational and actionable.
engineering-style passive voice. Show multiple ”tunable parameters” of the generative algorithm [115, 116, 117].
Human can revise recommendation or accept as-is.
model of Tikaani’s actions from past projects, and has predicted Tikaani’s most likely ”next action” today, saving Tikaani’s time in the notebook.
<10> As YPP began to display the coding strategy into the shared Slack channel, Tikaani finished their sip of chai and put their cup down. Now that YPP was doing most of the work, Tikaani could relax and watch it unfold. They knew that they would need to review YPP’s code, but they hadn’t had to correct YPP’s architectural assumptions during the past year. YPP called Tikaani’s attention to lower-confidence code, and Tikaani made edits as needed. Tikaani was impressed that ”Y” had learned so much of Tikaani’s individual coding practices during their last year together. It was true that Tikaani missed the days when people actually had to solve their own coding problems without genAI support. But work with ”Y” was easier, and much faster, and actually produced fewer bugs. This day of work with ”Y”, while maybe a little bit boring, was going to be a highly productive and also pleasurable perhaps, indeed, a very good day. <11> YPP displayed, ”Recognition follows. You earned 5 BerryByte points. Points were downloaded to your ID chip. Don’t forget to redeem them at the ByteBar.” And after a moment, ”CyberHR hope that this recognition will help you to have a good day, a very good day.” <12> “Drink chai,” YPP said. Tikaani took a sip of chai. <13> Tikaani began to perform a detailed review of the code that YPP had generated. As Tikaani opened a generic class, YPP displayed, ”There are alternative modules to consider. Actionable Explanation follows. Class QuikClientPortfolio() from commercial package QuikFinance would be 20% faster, but has license fees that are usually $5k to $8k. Do you want to use QuikClientPortfolio() and submit a request for approval?” <14> Unsure of how to make the decision, Tikaani typed, ”Inquire through YourPartnetNet for colleagues who have knowledge of QuikClientPortfolio. Poll ’Which release of QuikClientPortfolio is stable?’ and indicate my status as Blocked. Use probable-knowldege feature. Anonymous responses are acceptable.” <15> A minute later, YPP displayed, ”6 colleagues responded. 5 said Release 15.3 is stable. 1 said Release 15.1 is stable. Two non-anonymized names are available.” <16> Tikaani decided to accept the majority opinion. They typed, ”Estimate likelihood of approval for QuikClientPortfolio Release 15.3, based on data from the current year only.” <17> YPP displayed, ”Estimation follows. Package QuikFinance has been approved on 82% of projects this year. Based on your personal history with requests, likelihood of approval for you is 90%.” Then YPP repeated, ”Do you want to use QuikClientPortfolio() and submit a request for approval?”
tions as a Transactional Memory System (TMS) [118]. It stores records of ”who knows what” [119, 120], and it ”learns” each employee’s knowledge trajectory, so that it can generate probable knowledge-states based on past activities [121].
<18> Tikaani pressed the speech-to-text key and said, ”Paste QuikClientPortfolio, generate documentation of the decision rationale via GPT-neo and display it.” YPP generated the rationale. Tikaani reviewed, and made two corrections. Tikaani said “Send request for approval, and add to my personal log.” YPP displayed each step as it completed it.
for-approval.
Generative AI provides partial
results for the human to
complete [26].
<19> “Drink chai,” YPP said. Tikaani sipped.
<20> Tikaani found a serious domain-related problem in one of YPP’s
generated modules. Apparently YPP didn’t understand that
microifnancing often involved clients with little collateral and incomplete
credit histories. Tikaani knew that there would need to be additional
factors added to the model. They began to rewrite the generated class
from scratch. YPP issued a first warning: ”The recommended module
for this functionality is based on the MonthProjection() class from the
FinBlast library.” Tikaani continued to write code.
<21> YPP continued, “Explanation follows. The estimated cost in
work time for writing your own version is 3-5 hours for you, plus
2-4 hours for the QA team to test your new code, with a heightened
risk of bugs of 34%. Required action follows. (
action... ... and the approvals process that the human must follow... ... including a policy-based risk.
sensing algorithms [125] to determine that the human may be persuadable to take the buy-out that will benefit the company? [ 126].
that allowed YPP to predict the human’s preferred JupyterGen implementation environment <8>, can be used to create v-Tikaani. <24> After the frustration of having to justify a necessary rewrite, Tikaani was almostly ready to take the buy-out ofer. But no. When the work was good, it was still interesting – even though it was lonelier now that Yu had taken the buy-out and had been replaced with v-Yu. Tikaani declined the ofer, and went back to coding. <25> “Drink chai,” YPP said. Tikaani glared at the screen. <26> “Drink chai, YPP repeated, “You are at 46%. Explanation follows. Having a good day is an employee responsibility. Chai helps. Below 40%, level-1 reporting to management is mandated. Required action follows. Drink chai.” Was that a robot joke? Tikaani had heard rumors of an experimental emotion module. Just before Yu had retired and been replaced with v-Yu, Yu had suggested that – <27> YPP interrupted Tikaani’s reverie. “Drink chai. You are at 42%.” <28> Tikaani gulped down a full mouthful of chai, and braced against its metallic aftertaste. Tikaani thought back to the days when chai was just an optional beverage that they drank with Yu’s team during tea-breaks. Now chai had become a mandatory delivery vehicle for the Computer-Human Adaptive Intracellular. The the short-lived CHAI virobots were linked to the cyberHR department through near-field communications. The virobots monitored and corrected employees’ health and psychological engagement, based on a generated target bio-labor profile from a ”learned” dataset of biosignals from employees with demonstrated productivity and suficient job-tenure. Tikaani prepared for the brief moment of dizziness as the virobots crossed their blood-brain barrier to reach the emotion centers of the limbic system. YPP’s warnings had left Tikaani is a very grim mood. But now, Tikaani felt the Intracellular beginning to activate. Tikaani experienced a familiar calm. Despite the loneliness of the mostlyempty ofice, it was going to be a good day after all, a very good day.
and is sensed by the AI.
Similarly to policy-based campaigns to improve employee engagement[127, 128, 129], the company helps employees to work happily and productively through biological interventions... ... which are highly efective, perhaps benefiting company productivity and perhaps employee mental health?
ifction by reference to numbered paragraphs - e.g., ” <3>.”
The ”Drinking Chai” story explored a series of increasingly futuristic applications of generative AI. We briefly review them here: research will be required to specialize these patterns to an individual human’s pattern through a smaller set of personal data. We leave open the question of whether this would be a desirable outcome. • Again controversially, we considered that organizations might use generative technologies to ”learn” a ”best” set of employee attributes, and might seek to impose those attributes on lesscompliant employees <28>.
In the preceding discussion, we listed an escalating series of generative features, and we provided evidence • We began with the well-understood domain of (where it exists) that current research may be trending generative software translation in paragraphs toward those features. It may be useful to address the <1, 3, 5, 7, 8, 9, 10>, and we emphasized the need plausibility of the concept of virobots to influence emfor human-AI partnerships in translation <10, ployees’ emotions <26, 28>. While the control aspects 13, 16, 20> [28, 26, 27, 130], sometimes guided are futuristic, we note that in-dwelling digital devices by the AI’s flagging of low-confidence transla- have been part human medical audiology, endocrinology, tions <10> [26, 27] and by the AI’s recommen- and neurology for at least a decade [134], now as part dation of alternate classes and libraries <10, 13>. of a medical approach to Internet of Things [135] with Applied inflexibly, the principle of generative an- concomitant privacy and security risks [136, 137]. Some ticipation of the ”best” coding usages will lead to of these devices are already being used to modify the trouble in paragraphs <20-22>. patient’s brain state [138, 139]. In the commercial space, • We also included currently-available capabilities Applied Digital company’s Verichip is marketed as a subto generate documentation for the generated dural injection of an RFID tag for building access[140]. and modified code, as has been done in limited Further, we note that organizations have for years ways for source code <18> [40, 8] and also for played white noise [141] or curated music [142, 143, 144, certain sub-genres of journalism (e.g., [131]). 141] to afect employees’ mental status for specific organi• We proposed capabilities for the human to con- zational purposes such as enhanced concentration [143], trol, steer, and tune specific aspects of the gen- job-engagement [142], and even reduction in costly emerative processes and outcomes <5, 7> [115, 22, ployee-controlled overtime hours [144]. Thus, employers’ 132]. interests in afecting employee mental state are already • We included GUI style transfer as a way of re- part of conventional ofice technologies, and implantable ifning the generic translations <6> [133]. digital devices are already a medical and industrial reality, • We proposed a more futuristic capability to parse as is the use of signals from these devices to modify a a specification document into an architecture person’s internal milieu. The only questions that sepplan, and then into a high-level class structure <3> arate the existing state-of-the-art technologies and our with necessary human guidance <5, 7> . paragraphs <26, 28> are: What kinds of psychological • Further into the future, we proposed that a gener- modifications might be implemented in the future? and
How would they be ”delivered?” and Who controls those ative assistant could learn the work-practices modifications? and of course Is this a desirable future? of its human partner, and could save the human’s efort by suggesting and then implementing the anticipated ”next steps” <8, 9, 17>. 4.2. Value Tensions • More controversially, we considered that organi- In keeping with theory and practice of design fictions [ 83, zations could develop virtual versions of par- 64, 107, 91] and feminist technoscience [109, 110, 108, ticular employees through ”learning” their in- 111], we hoped to raise questions of personal, organizadividual work-practices and then implementing tional, and societal values through our fiction. Our use of those patterns into a virtual replacement for the the principle of strategic ambiguity [103, 106, 70] helps employee <23, 24, 26>. We acknowledge that this to highlight some of the tensions: idea is futuristic. GANs and related algorithms require large amounts of data. While organizations • What are the trade-ofs of panoptic [ 145, 146] may be able to use existing generative algorithms surveillance on employees’ work-practices to for the general case of human actions, further achieve organizational goals of productivity <8,
17] to develop conjectures about possible futures of generative AI features and applications, and to raise valuesbased questions about those possible futures. It is fair to ask, ”what did we learn from this research method?” In the Discussion, we explored two topics.
We considered current generative AI approaches, and we projected them into possible futures. Generative software translation and generative documentation are current capabilities [28, 26, 27, 130, 40, 8, 131]. Further, we note that there is already research under way to provide finer controls over generative algorithms citelouie2020novice, louie2020cococo, zhou2020generative - although there is much work yet to be done. We used those concepts as ”starting points,” to explore more futuristic ideas.
If multiple generative AI applications predict the ”next
token” in a sequence of tokens, then we speculated about
the nature of possible future tokens, and what those
tokens might be used for. We considered what might be
possible - for good or ill - if human actions were treated
as be tokens. We were then able to imagine helpful
scenarios, in which an AI could beneficially anticipate the
ACM Conference on Supporting Groupwork, 2018, [103] P. Coulton, J. G. Lindley, M. Sturdee, M. Stead,
pp. 257–266. Design fiction as world building (
M. Muller, Teaching robots to act and converse in [105] D. Simons, Design for fairness in ai: Cooking a
physical spaces: participatory design fictions with fair ai dish (
tion, Digital Creativity 24 (2013) 11–35. [107] A. Morrison, R. Tronstad, E. S. Martinussen, De[91] M. L. J. Søndergaard, L. K. Hansen, Periodshare: A sign notes on a lonely drone, Digital Creativity 24 bloody design fiction, in: Proceedings of the 9th (2013) 46–59.
Nordic Conference on Human-Computer Interac- [108] M. L. J. Søndergaard, Troubling design: A
detion, 2016, pp. 1–6. sign program for designing with women’s health,
[92] B. Sterling, Shaping things, MIT Press, 2005. ACM Transactions on Computer-Human
Interac[93] M. Blythe, Research fiction: storytelling, plot and tion (TOCHI) 27 (
ifction (
interactive systems, 2014, pp. 1075–1078. [114] M. Chefitz, J. Austin-Breneman, N. Melville,
De[99] B. Gaver, T. Dunne, E. Pacenti, Design: Cul- signing conversational interfaces to reduce
dissotural probes, Interactions 6 (1999) 21–29. nance, in: Proceedings of the 2018 ACM
ConferURL: https://doi.org/10.1145/291224.291235. ence Companion Publication on Designing
Interdoi:10.1145/291224.291235. active Systems, 2018, pp. 219–223.
[100] K. Boehner, J. Vertesi, P. Sengers, P. Dourish, How [115] R. Louie, A. Coenen, C. Z. Huang, M. Terry, C. J.
hci interprets the probes, in: Proceedings of the Cai, Novice-ai music co-creation via ai-steering
SIGCHI conference on Human factors in comput- tools for deep generative models, in: Proceedings
ing systems, 2007, pp. 1077–1086. of the 2020 CHI Conference on Human Factors in
[101] W. Gaver, Cultural commentators: Non-native Computing Systems, 2020, pp. 1–13.
interpretations as resources for polyphonic assess- [116] D. Dablain, G. Siwo, N. Chawla, Generative ai
dement, International journal of human-computer sign and exploration of nucleoside analogs (
That’s what friends are for: facilitating’who [133] T. Zhao, C. Chen, Y. Liu, X. Zhu, Guigan: Learning knows what’across group boundaries, in: Proceed- to generate gui designs using generative adversarings of the 2007 international ACM conference on ial networks, in: 2021 IEEE/ACM 43rd InternaSupporting group work, 2007, pp. 379–382. tional Conference on Software Engineering (ICSE), [120] B. Biancardi, L. Maisonnave-Couterou, P. Renault, IEEE, 2021, pp. 748–760.
B. Ravenet, M. Mancini, G. Varni, The wonowa [134] J. Ahmad, F. Zafar, Review of body area network dataset: Investigating the transactive memory sys- technology & wireless medical monitoring, Intertem in small group interactions, in: Proceedings of national Journal of Information and Communicathe 2020 International Conference on Multimodal tion Technology 2 (2012).
Interaction, 2020, pp. 528–537. [135] A. K. Teshome, B. Kibret, D. T. Lai, A review
[121] S. Liang, Y. Luo, Z. Meng, Profiling users for ques- of implant communication technology in wban:
tion answering communities via flow-based con- Progress and challenges, IEEE reviews in
biomedstrained co-embedding model, ACM Transactions ical engineering 12 (
on Information Systems (TOIS) 40 (
MIT Press, 2022. [137] U. Ahmad, H. Song, A. Bilal, S. Saleem, A. Ul[123] T. Ye, Improving Worker Performance with lah, Securing insulin pump system using deep
Human-Centered Data Science, Ph.D. thesis, 2021. learning and gesture recognition, in: 2018 17th [124] Y. Toyoda, G. Lucas, J. Gratch, The efects of au- IEEE International Conference On Trust, Secutonomy and task meaning in algorithmic manage- rity And Privacy In Computing And Communiment of crowdwork, in: Proceedings of the 19th cations/12th IEEE International Conference On International Conference on Autonomous Agents Big Data Science And Engineering (TrustCom/Bigand MultiAgent Systems, 2020, pp. 1404–1412. DataSE), IEEE, 2018, pp. 1716–1719. [125] Z. Wang, C. S. Chong, L. Lan, Y. Yang, S. B. Ho, [138] H. Kassiri, N. Soltani, M. T. Salam, J. L. P. Velazquez, J. C. Tong, Fine-grained sentiment analysis of R. Genov, Battery-less modular responsive neusocial media with emotion sensing, in: 2016 Future rostimulator for prediction and abortion of epilepTechnologies Conference (FTC), IEEE, 2016, pp. tic seizures, in: 2016 IEEE International Sympo1361–1364. sium on Circuits and Systems (ISCAS), IEEE, 2016, [126] M. Skirpan, C. Fiesler, Ad empathy: A design fic- pp. 1298–1301.
tion, in: Proceedings of the 2018 ACM Conference [139] K. Abdelhalim, H. M. Jafari, L. Kokarovtseva,
on Supporting Groupwork, 2018, pp. 267–273. J. L. P. Velazquez, R. Genov, Neural
synchrony[127] J. Anitha, Determinants of employee engagement monitoring wireless brain implant for intractable
and their impact on employee performance, Inter- epilepsy neuromodulation, in: 2013 6th
Internanational journal of productivity and performance tional IEEE/EMBS Conference on Neural
Engineermanagement (
19wempen.pdf, accessed: 2021-12-22. [141] S. J. Schlittmeier, J. Hellbrück, Background music
[129] W. H. Macey, B. Schneider, The meaning of em- as noise abatement in open-plan ofices: A
laboraployee engagement, Industrial and organizational tory study on performance efects and subjective
Psychology 1 (2008) 3–30. preferences, Applied Cognitive Psychology: The
[130] M. Agarwal, K. Talamadupula, F. Martinez, Oficial Journal of the Society for Applied Research
S. Houde, M. Muller, J. Richards, S. I. Ross, J. D. in Memory and Cognition 23 (2009) 684–697.
Weisz, Using document similarity methods to cre- [142] A. B. Haake, Individual music listening in
workate parallel datasets for code translation, arXiv place settings: An exploratory survey of ofices in
preprint arXiv:2110.05423 (
Labour Review (
Shaping technology with moral imagination, Mit
Press, 2019. [148] L. V. Nägele, M. Ryöppy, D. Wilde, Pdfi: participatory design fiction with vulnerable users, in: Proceedings of the 10th Nordic Conference on
Human-Computer Interaction, 2018, pp. 819–831. [149] K. Baumann, B. Caldwell, F. Bar, B. Stokes, Participatory design fiction: community storytelling for speculative urban technologies, in: Extended Abstracts of the 2018 CHI Conference on Human
Factors in Computing Systems, 2018, pp. 1–1.
[150] L. Forlano, A. Mathew, From design fiction to
design friction: Speculative and participatory design
of values-embedded urban technology, Journal of
Urban Technology 21 (