=Paper=
{{Paper
|id=Vol-2736/paper5
|storemode=property
|title=Human Computation Requires and Enables a New Approach to Ethics
|pdfUrl=https://ceur-ws.org/Vol-2736/paper5.pdf
|volume=Vol-2736
|authors=Libuše H. Vepřek,Patricia Seymour,Pietro Michelucci
|dblpUrl=https://dblp.org/rec/conf/nips/VeprekSM20
}}
==Human Computation Requires and Enables a New Approach to Ethics==
https://ceur-ws.org/Vol-2736/paper5.pdf
Human Computation Requires and Enables a New
Approach to Ethics
Libuše H. Vepřek∗ Patricia Seymour
Human Computation Institute Human Computation Institute
Ithaca, NY 14850 Ithaca, NY 14850
l.veprek@lmu.de pannseymour@yahoo.com
Pietro Michelucci
Human Computation Institute
Ithaca, NY 14850
pem@humancomputation.org
Abstract
With humans increasingly serving as computational elements in distributed in-
formation processing systems and in consideration of the profit-driven motives
and potential inequities that might accompany the emerging thinking economy[1],
we recognize the need for establishing a set of related ethics to ensure the fair
treatment and wellbeing of online cognitive laborers and the conscientious use
of the capabilities to which they contribute. Toward this end, we first describe
human-in-the-loop computing in context of the new concerns it raises that are not
addressed by traditional ethical research standards. We then describe shortcomings
in the traditional approach to ethical review and introduce a dynamic approach
for sustaining an ethical framework that can continue to evolve within the rapidly
shifting context of disruptive new technologies.
1 Introduction
A new branch of artificial intelligence called “human computation” has emerged over the last 15
years that combines the respective strengths of humans and computers to tackle problems that cannot
be solved in other ways[2]. Information processing systems based on this approach often employ
online crowdsourcing to delegate to humans cognitive “microtasks” that elude the capabilities of
machine-based methods. Real-world human computation systems are already advancing cancer[3],
HIV[4], and Alzheimer’s[5] research, diagnosing malaria[6] in sub-Saharan Africa, reducing female
genital mutilation in Tanzania[7], predicting flood effects in Togo[8], endowing the blind with
real-time scene understanding[9], expediting disaster relief despite language barriers and failing
infrastructure[10], rewriting our understanding of cosmology[11][12] and improving predictions in
conservation science[13].
Four main paradigms exist for engaging people in human computation tasks. One of the first online
human computation systems, reCAPTCHA, operates on a quid-pro-quo basis, requiring a person
to digitize distorted text to demonstrate being human, which provides access to a website while
simultaneously contributing to a massive text digitization project[14]. Modern versions of this, such
as hCaptcha, require people to select all images from a provided set that contain some class of objects,
∗
Ludwig-Maximilians-Universität München
NeurIPS 2020 Crowd Science Workshop: Remoteness, Fairness, and Mechanisms as Challenges of Data Supply
by Humans for Automation. Copyright © 2020 for this paper by its authors. Use permitted under Creative
Commons License Attribution 4.0 International (CC BY 4.0).
such as traffic lights. This generates revenue by providing a data-labelling service that is used to help
train customers’ machine learning models[15]. Another paradigm, citizen science, entices public
volunteers to participate in the scientific process[16] through which they collect and analyze research
data in exchange for hands-on opportunities to learn about various research topics. Citizen science has
been steadily gaining popularity via community connectors like SciStarter.org and curation platforms
like Zooniverse. In a third engagement paradigm, “clickworkers” are typically paid via crowdsourcing
marketplaces, such as Amazon Mechanical Turk and ClickWorker.com to participate in various online
tasks that might be used for science, marketing, or product development applications.[17] Mooqita
represents a version of paid-crowdsourcing that uses massive open online courses (MOOC) as a way
to onboard new employees for salaried jobs involving online cognitive labor for a single organization
rather than a marketplace. In the final and most recent paradigm, popular online games such as
EVE Online and Borderlands Science have begun embedding microtasks within existing gameplay
to engage potentially millions of online gamers who opt-in to participate in exchange for various
advancement opportunities in the games[18].
Humans are squarely involved in human computation systems, and although most configurations do
not constitute human subjects research, they often pose new ethical dilemmas. Thus, traditional re-
search values and review practices typically do not apply to human computation and seem inadequate
to address the many new human-centered contexts produced by this growing field of study.
This becomes especially apparent in ethical review processes. Currently, online Citizen Science
research is reviewed by the same standards as clinical trials although the review board’s understanding
of “human subject” or “research participant” does not fit the role of participants in online citizen
science. Citizen scientists are not mere “human subjects” in a research study, but can be researchers,
human subjects, or sometimes both at the same time. This implies that their role has to be communi-
cated and elucidated to the traditional review boards. These new role allocations also come along with
different ideas and expectations for the different stakeholders, especially those performing tasks. For
example, David Resnik shows that if the role of citizen scientists exceeds being a mere human subject,
they may feel more connected and hence more ownership over their collected data[19]. Moreover,
in the arena of citizen science and human computation, traditional IRB is not clearly mandated and,
when used, it rigidly enforces ethical standards most often associated with biomedical research.
Ironically, human computation provides a potential solution to these ethical challenges. Herein we
argue for a new approach to ethical oversight that addresses the needs of online citizen science and
the new forms of human-computer collaboration in a digital age. First we explain how a distinction
between morals and ethics can be useful for this endeavor. Based on this distinction we explore
how we might appeal to the participatory methods of human computation to design a technosocial
platform that enables the curation of a “living” set of ethics and crowdsources the application of those
ethics to a suitable review process.
2 Morality vs. Ethics
In everyday life “morality” and “ethics” are used synonymously to describe the “good” in contrast to
the “bad” or to distinguish between “right” and “wrong”. Formally, however, Ethics is considered a
branch of philosophy that studies morality[20], or in the words of the German philosopher Dietmar
Hübner: “Morality is the object, ethics is science.”[21] This brings us to the question of: what
ismorality? According to Hübner, then is “a system of norms, whose subject is human behavior and
which claims unconditional validity”(ibid.: 13, transl. b.t.a.). This means that different “morals” exist
in different contexts such as different cultures or political currents, and there even exist different
morals for specific groups of people like doctors, journalists and scientists. To decide whether a given
moral is right or wrong therefore requires reflecting on the normative context in which it occurs. Thus,
the aim of differentiating between “morality” and “ethics” is not to enforce correct usage, but about
orienting our perspective as we consider ethical review in the new kinds of digital collaborations that
manifest in human computation systems such as online citizen science.
We can now relate to two different layers that need to be considered: The first layer consists of
the moral framework in the field. We need to understand what the moral values of the different
stakeholders in human computation and citizen science are: For example, what is important to citizen
science participants? What do they expect from their participation and how do they want to be
acknowledged? This understanding of “moral framework” could be related to what Lisa Rasmussen
calls the “citizen science ethos”[23]. To identify moral values in online citizen science, we introduced
an online discussion forum and invited both citizen scientists and citizen science practitioners to
contribute [22]. The analysis of the different discussion thread shows that mutual respect, inclusion
and transparency a.o. belong to the “moral framework” of citizen science. This could then inform the
second layer which consists of reflecting this moral framework in a set of ethical guidelines, which
could both inform the design and execution of citizen science projects as well as evaluation by ethical
oversight committees, who can then review citizen science research in accordance with community
values. Of course, these ethical guidelines would need to remain flexible as we may discover that
values that apply to most citizen science projects may still be ill-suited to certain specific studies.
3 Traditional IRB
The difference between our attitudes toward people who participate in research can perhaps be
seen best in the debate between using the term “participant” or “subject” to refer to someone who
volunteers their body, mind, and time to research. In 2014, Public Responsibility in Medicine
and Research (PRIM&R) put forth the following statement on this debate: “subject” is the most
appropriate title for those involved in research studies (recognizing, however, that in some instances
“participant” may be appropriate; for example, in community-based participatory research). In the
world of citizen science, however, participant activities may be more closely aligned with the work
that scientists do. This means that currently there is no existing model of review that appropriately
assigns autonomy to those engaged.
Moreover, the absence of ethical guidelines for citizen science creates a dilemma for indepen-
dent/institutional reviews (IRB) or ethics review boards (ERB) because they have to choose between
either applying ethical guidelines that do not fit the application or making their own decisions about
what’s right or wrong. The lack of consistent interpretation of research studies has prompted the
National Institutes of Health in the United States to mandate a single IRB review for some federally
funded research. The researcher, at the time of the grant application, identifies an IRB that will serve
as the only IRB for the project. This removes the multitude of review decisions, consent variation and
timing delays that routinely plague even the most prestigious and well-funded research institutions.2
Lastly, we build on our own experiences with IRB processes in the field of human computation-based
citizen science: Working with a traditional IRB highlighted the inconsistencies and infeasibility of
using the traditional US biomedical oriented approach of most IRBs. Fortunately, the experience was
mitigated by a knowledgeable and flexible liaison who helped to inform the members of the IRB
of the differences in human computation from biomedical research. For example, the concept of
risk and benefit, as defined by the U.S. regulations and guidance did not apply in a straightforward
manner to human computation. The protectionism of the US regulations may not be applicable to
human computation nor does it apply to the participant’s autonomy when performing tasks in this
arena. Each difference in approach and regulatory necessity had to be addressed, thus causing delays
and redundancies that was not conducive to efficient review.
Why should citizen science, and indeed the broader space of human computation, as a relatively new
approach to research, enter the fray of poorly organized and inefficient traditional review processes
just because review has always been done this way? Instead, we envision a triage-based approach
that determines whether traditional research ethics review applies to a research project based on the
project goals and participant task requirements and flips four key aspects of traditional IRB. It also
involves input from members of the human computation community at large.
4 Flipping IRB
Traditionally, IRB has often been compulsory and seemed like an adversarial process, where reviewers
were there to poke holes in the researcher’s ethical approach and eventually return a judgment. This
2
Throwing technology at the issue of IRB review was thought to be an answer to better organize and
streamline what had, for many years, been a paper-based process. The result of the development and use of
these electronic, online platforms has not been particularly helpful because it applies technology to an inefficient
process. To make matters worse, the technology is customized to the already-broken system, perpetuating the
pre-existing problems in an online context. Moreover, the systems available in the U.S. do not communicate
between institutions and do not solve the problem of inconsistent decisions and documents.
process could be recast as a collaborative one in which the role of the IRB expert is to help researchers
align their methods with established ethical guidelines toward achieving their research goals.
Traditionally, an application is filled out cautiously and word choices are carefully made to elicit an
approval outcome. The application then disappears into a black box called the IRB process, where
mystical and sometimes random-seeming things happen until a determination comes out the other
end. We believe full transparency is critical for building trust and working toward a shared goal. Once
aligned with an IRB expert who can direct the ethical review of a project as a collaborator, there is a
clear path to approval.
Traditionally, applications are isolated documents that are filed away and never seen again (until
it’s time to amend or renew). But that means that even if two projects by different researchers are
very similar, they each have to go through the same time-intensive, expensive, and laborious process.
We think that if your project is like someone else’s then we only need to consider the parts that are
different. We suggest that if we begin building a protocol repository, we can assist researchers to use
designs and approaches that have already been approved thus decreasing the issues that have to be
reviewed time and time again.
Traditionally, there was a ready roster of mostly the same people who were on tap to review applica-
tions as needed. We think the only time you need a panel is when there is something that isn’t clear
cut, and in that case, let’s enlist our community of peers, including human computation community
volunteers who have agreed to be on call.
5 A human computation approach to ethical review
Today, we have an opportunity to build a technosocial platform that turns this vision into reality.
We apply our own human computation approaches to building consensus and determining the right
division of labor between humans and machines. We also draw on our experience related to creating a
human computation platform that continues to engage over 30,000 volunteers to motivate community
participation in the review process. Indeed, we have begun to build this - it is called Civium, and its
main purpose is to make human computation research and applications more transparent, trustworthy,
and sustainable.3
For starters, we created an experimentation toolkit that makes it possible to clone a citizen science
project with a single click to create a sandbox version for running experiments without affecting the
live platform or data quality. This sandbox environment can be used to run an experiment that studies
the behaviors of the citizen science volunteers. In this case, volunteers are not working alongside
scientists analyzing data, they are being studied by scientists to help improve our understanding about
how to design effective citizen science platforms. For example, in one case study we investigated
human/AI-partnerships in the online game Stall Catchers. By including surveys in different stages of
the experiment and analyzing the collected (meta) data we could gain insights into when and why
users trusted the AI-assistant and when they questioned its’ skill. This configuration, however, goes
beyond traditional citizen science, might suggest the need for ethical review.
To make that process quick, informative, and painless, we are integrating a new IRB process (see
Figure 1) into the Civium environment. That means that instead of packaging up a description of
an experiment for the IRB expert, the expert can enter into the sandbox with the researcher and
examine the experiment to understand its design (see Figure 2 and 3 in the appendix) and see from the
standpoint of a participant how it will actually work (see Figure 4 in the appendix). The IRB expert
can make comments and suggest edits directly on the interface in support of aligning the research
goals with the community’s current ethical standards.This new procedure should minimize the time a
reviewer has to work on a proposal, since the submitted proposal could entail links to the sandbox
experiment allowing the reviewer to gain a better understanding of the design logic and the interface.
Traditionally, the reviewer would have to think about how the experiment would probably work based
on written material and some screenshots. The explicit role of the IRB expert is to shepherd the
review process and activate various assistive mechanisms as needed in service of that goal.
3
An effort based on a similar logic is OpenReview, a web interface with underlying database API that aims at
advancing openness in the scientific peer review process. We would like to thank the anonymous reviewer for
drawing our attention to this analogy.
Figure 1: Activity diagram of technosocial platform for ethical review
Meanwhile, the system logs the issues that arise and the ensuing dialog around those issues, as well as
any implemented solutions. These are made transparent and accessible in a repository that connects
these to a snapshot of the sandbox itself. This introduces something completely new and powerful for
ethical review - something we demand for our legal system and something we aim for in scientific
research, but that is missing in IRB, which is reproducibility. And when questions arise that are not
addressed or that are ambiguous under the current set of ethical guidelines, the IRB expert invokes
members of the community to review the issues. This is analogous to a section editor for an academic
journal finding reviewers for a manuscript. And the platform can help manage the recruitment of
these community contributors.
Finally, and critically, the outcome of this process can not only resolve the ethical dilemma for the
researchers, it can inform amendments to the ethical guidelines themselves, so that they can continue
to grow with our understanding of human-in-the-loop computing and fit the needs and circumstances
of this growing community.
Broader Impact
Herein we address the need for a general-purpose mechanism for AI governance that can evolve with
our understanding of the field. Not only does AI raise issues of autonomy, labor, and equitability, but
with increased reliance on systems that employ human cognition, the ethical waters become even
murkier. Our approach seeks to crowdsource the evaluation of the risks and rewards of situated AI
systems as well as human-computer collaborations, and aims at seeding and curating a set of related
ethics to ensure the fair treatment and wellbeing of humans in-the-loop. To ensure that the needs of
all stakeholders are taken into account we include diverse perspectives in a maximally transparent
process. For example, all reports will be stored in a searchable public repository. However, despite
our best intentions and planning, there are always the “unknown unknowns” that can arise when a
platform like this goes live. To address these risks, we will remain vigilant to the system’s behavior
and leverage community monitoring and feedback loops intrinsic to our approach.
Acknowledgments and Disclosure of Funding
We would like to thank Eglė Marija Ramanauskaitė for her great assistance and preparation of the
activity diagram and Percy Mamedy for his work on the implementation of the platform. We also
wish to show our appreciation to all participants of our discussion on reinventing IRB as well as the
contributors in the citizen science forum for their helpful insights and the fruitful discussions.
Libuše Hannah Vepřek, Patricia Seymour and Pietro Michelucci declare that they have no conflict of
interest.
References
[1] P. Michelucci, “How do we create a sustainable thinking economy?," Toward Data Science, Nov. 04, 2019.
https://towardsdatascience.com/how-do-we-create-a-sustainable-thinking-economy-4d77839b031e (accessed
Feb. 04, 2020).
[2] P. Michelucci and J. L. Dickinson, “The power of crowds," Science, vol. 351, no. 6268, pp. 32–33, Jan.
2016, doi: 10.1126/science.aad6499.
[3] F. J. Candido Dos Reis et al., “Crowdsourcing the General Public for Large Scale Molecular Pathology
Studies in Cancer," EBioMedicine, vol. 2, no. 7, pp. 681–689, Jul. 2015, doi: 10.1016/j.ebiom.2015.05.009.
[4] F. Khatib et al., “Crystal structure of a monomeric retroviral protease solved by protein folding game players,"
Nat. Struct. Mol. Biol., vol. 18, no. 10, pp. 1175–1177, Sep. 2011, doi: 10.1038/nsmb.2119.
[5] O. Bracko et al., “High fat diet worsens pathology and impairment in an Alzheimer’s mouse model,
but not by synergistically decreasing cerebral blood flow," bioRxiv, p. 2019.12.16.878397, Dec. 2019, doi:
10.1101/2019.12.16.878397.
[6] M. A. Luengo-Oroz, A. Arranz, and J. Frean, “Crowdsourcing Malaria Parasite Quantification: An Online
Game for Analyzing Images of Infected Thick Blood Smears," J. Med. Internet Res., vol. 14, no. 6, Nov. 2012,
doi: 10.2196/jmir.2338.
[7] M. O. Leaders, “Put Rural Tanzania on the Map," Read, Write, Participate, Apr. 19, 2018.
https://medium.com/read-write-participate/put-rural-tanzania-on-the-map-79d0888df210 (accessed May 02,
2019).
[8] P. Suarez, “Rethinking Engagement: Innovations in How Humanitarians Explore Geoinformation," ISPRS
Int. J. Geo-Inf., vol. 4, no. 3, pp. 1729–1749, Sep. 2015, doi: 10.3390/ijgi4031729.
[9] J. P. Bigham et al., “VizWiz: Nearly Real-time Answers to Visual Questions," in Proceedings of the 23Nd
Annual ACM Symposium on User Interface Software and Technology, New York, NY, USA, 2010, pp. 333–342,
doi: 10.1145/1866029.1866080.
[10] P. Meier, “Human Computation for Disaster Response," in Handbook of Human Computation, P. Michelucci,
Ed. New York, NY: Springer, 2013, pp. 95–104.
[11] A. J. Westphal et al., “Evidence for interstellar origin of seven dust particles collected by the Stardust
spacecraft," Science, vol. 345, no. 6198, p. 786, Aug. 2014, doi: 10.1126/science.1252496.
[12] C. N. Cardamone et al., “Galaxy Zoo Green Peas: Discovery of A Class of Compact Extremely Star-Forming
Galaxies," Mon. Not. R. Astron. Soc., vol. 399, no. 3, pp. 1191–1205, Nov. 2009, doi: 10.1111/j.1365-
2966.2009.15383.x.
[13] S. Kelling et al., “eBird: A Human/Computer Learning Network for Biodiversity Conservation and
Research," presented at the Twenty-Fourth IAAI Conference, Jul. 2012, Accessed: May 02, 2019. [Online].
Available: https://www.aaai.org/ocs/index.php/IAAI/IAAI-12/paper/view/4880.
[14] L. von Ahn, B. Maurer, C. McMillen, D. Abraham, and M. Blum, “reCAPTCHA: Human-Based Character
Recognition via Web Security Measures," Science, vol. 321, no. 5895, pp. 1465–1468, Sep. 2008, doi:
10.1126/science.1160379.
[15] “Moving from reCAPTCHA to hCaptcha," The Cloudflare Blog, Apr. 08, 2020.
https://blog.cloudflare.com/moving-from-recaptcha-to-hcaptcha/ (accessed Oct. 09, 2020).
[16] F. Heigl, B. Kieslinger, K. T. Paul, J. Uhlik, and D. Dörler, “Opinion: Toward an international definition of
citizen science," Proc. Natl. Acad. Sci., vol. 116, no. 17, p. 8089, Apr. 2019, doi: 10.1073/pnas.1903393116.
[17] J. Chandler, G. Paolacci, and P. Mueller, “Risks and Rewards of Crowdsourcing Marketplaces," in Handbook
of Human Computation, P. Michelucci, Ed. New York, NY: Springer, 2013, pp. 377–392.
[18] J. Waldispühl, A. Szantner, R. Knight, S. Caisse, and R. Pitchford, “Leveling up citizen science," Nat.
Biotechnol., vol. 38, no. 10, pp. 1124–1126, Oct. 2020, doi: 10.1038/s41587-020-0694-x.
[19] D. B. Resnik, “Citizen Scientists as Human Subjects: Ethical Issues," Citiz. Sci. Theory Pract., vol. 4, no.
1, Art. no. 1, Mar. 2019, doi: 10.5334/cstp.150.
[20] J. Deigh, An Introduction to Ethics. Cambridge: Cambridge University Press, 2010.
[21] D. Hübner, Einführung in die philosophische Ethik, 2nd ed. Göttingen: UTB GmbH, 2018.
[22] “Citizen Science Ethics" Human Computation Institute forum, 2020. https://forum.hcinst.org/c/citsci-
ethics/13 (accessed Nov. 18, 2020).
[23] L. M. Rasmussen, “Confronting Research Misconduct in Citizen Science,"Citiz. Sci. Theory Pract., vol.
4,no. 1, Art. no. 1, Mar. 2019, doi: 10.5334/cstp.207.
Appendix
Figure 2: Screenshot of nova backend: example experimental design (prototype)
Figure 3: Screenshot of nova backend: example experimental design (prototype)
Figure 4: Screenshot of example user interface for experiments (prototype)