Page 444 -

14.3 Human computation 435

avoiding the unfortunate outcome of a failed experiment leading to ill will (McInnis
and Leshed, 2016). It may not be possible to identify all problems in advance, but
working with the community of users to build trust and promote fairness may be an
important strategy for successful human computation studies.

14.3.3 FUTURE OF HUMAN COMPUTATION
Human computation has many promising applications. A 2014 workshop of the
Computing Community Consortium of the Computing Research Association out-
lined numerous possibilities for the use of human computation to meet pressing
social needs, including social support for people in need; combining training with
problem solving to improve the process of interpreting radiology images; to collect
river-level information and serve as an early warning for possible floods; and others
(Michelucci et al., 2015). As our engagement with our devices continues to occupy
much of our time and attention, attempts to channel this fascination in socially mean-
ingful ways are likely to continue to be a growing part of the landscape.
HCI research efforts have explored possible extensions to crowd source models,
designed to increase the utility of crowdsourced work. Possibilities include changing
task structures to include “handoffs” between workers, thus possibly increasing the
quality of the resulting work (Embiricos et al., 2014); using algorithmic approaches
to plan task workflow (Weld, 2015); exploring the impact of task ordering on speed
and mental demand during the completion of a sequence of small tasks (Cai et al.,
2016); and using new models to encourage participation, including leveraging par-
ticipant curiosity (Law et al., 2016), providing entertaining “micro-diversions” to
improve productivity of workers conducting many tasks (Dai et al., 2015), or using
“twitch” microtasks capable of being completed very quickly to lower barriers to in-
volvement (Vaish et al., 2014). Other efforts have explored paying crowd workers to
be ready to respond quickly, thus enabling real-time crowdsourcing (Bernstein et al.,
2011), applying algorithmic approaches to identify when tasks should be reassigned
because original workers have abandoned them (Kucherbaev et al., 2016), and using
models of increased error tolerance to increase the rate at which large tasks can be
completed (Krishna et al., 2016).
Another promising line of research asks a slightly different question—“how can
crowdsourced workers help with familiar, knowledge-intensive tasks?” As complex
tasks, writing papers, drawing figures and diagrams, and analyzing budgets require
significant cognitive effort and attention to detail, crowdsourced workers might help
writers, designers, and analysts with on-demand suggestions for improving the qual-
ity of their work. These possibilities drove the development of Soylent, a set of tools
for using human computation to improve the writing process. Developed as exten-
sions to Microsoft Word, Soylent provides writers with the ability to request human
computation assistance in shortening texts, grammar and spell-checking, and other
tasks not easily accomplished via existing word processing tools (Bernstein et al.,
2015). Although the possibilities of using human computation assistance to assist