Special Issue for Human Computation on
Exploring the Interplay between Human Learning and Machine Learning
The Citizen Science Perspective
- Laure Kloetzer (Assistant Professor in Psychology & Education at University of Neuchâtel, Switzerland)
- Marisa Ponti, Scientific Project Officer (European Commission, DG Joint Research Centre (JRC), Ispra, Italy)
- Sven Schade, Scientific/Technical Project Officer (European Commission, DG Joint Research Centre (JRC), Ispra, Italy)
- One additional co-editor
This special issue will be organised under the auspices of the COST Action 15212 Citizen Science to promote creativity, scientific literacy, and innovation throughout Europe and its Working Group 2 ‘Develop synergies with education’ and Working Group 4 ‘Enhance the role of CS for civil society’. The special issue is part of the project DigiTranScope: Digital Transformation and the Governance of Human Society (2018-2020) at the JRC Centre for Advanced Studies and is an output of the workshop held at the JRC Seville in April 2019.
Scope of the special issue and why it is opportune now
The possible interplays between human learning and machine learning has (so far) not been sufficiently addressed in the context of citizen science. This special issue will provide a milestone contribution to this quickly emerging field.
Over the past 20 years, citizen science (including online citizen science) and Artificial Intelligence (AI) have been developing immensely and growing more and more popular. This double success seems to happen on parallel tracks. Machine learning technology has advanced thanks to the cross-fertilization of computer and cognitive science, and sophisticated models have been proposed in computer vision, speech analysis, music processing and bioinformatics, to name a few areas. Citizen science has expanded among scientific disciplines, from ecology, biology and astronomy, to medicine, geography, mathematics, history and climate change, for example. Interestingly, although citizen science is usually defined by inclusion of members of the public as contributors and sometimes co-designers in scientific projects , a close observation of some highly successful citizen science projects displays a more complex structure, in which it is not uncommon that members of the general public collaborate, directly or indirectly, not only with professional scientists, but also with algorithms. Foldit, iNaturalist, Eyewire, Phylo, Stall catchers…, to name just a few, all rely on a complex interplay of human input and machine learning to reach their goals. One could even consider that this human learning-machine learning symbiosis is the most secret feature of a full range of recent citizen science projects – highly efficient, but largely overlooked by the research on citizen science so far. We suggest to call these projects hybrid or “centaurus” projects, meaning that the volunteers sit (visibly or not!) on the shoulders of algorithms to perform the scientific tasks designed for the project.
This interplay of human learning and machine learning, or human-machine collaboration, already takes a fantastic variety of forms. The most obvious is pattern and species recognition: it is possible to train an algorithm to develop specific image recognition skills, which can be used in projects that require classification of large amounts of image data. For example, automatic plant image identification is now receiving attention in both botany and computer communities could be used in citizen science. However, in other fields such as astronomy, the classification of galaxies structure is not yet considered a task for computers - the human eye is still seen as the perfect pattern recognition tool. A general problem in citizen science (and generally in science) is that data grows much faster than the number of research participants. Although human efforts will always be needed in citizen science, combining these efforts with big data techniques has been said to help researchers process more data faster and allow the participants to focus on the harder classifications.
This raises at least five main questions: can we document interesting examples of such human-machine collaboration in citizen science ? Can we then create a typology of current various uses of AI in citizen science ? Can we identify the conditions requested to make these centaurus projects successful (and how to define their conditions of success by the way)? Can we also identify their current problems and limits ? And finally, last but not least, what can we learn from the successful case of Citizen Science which could be useful for understanding and designing other situations of human – machine collaboration ? In the general public, AI may today largely be seen more as a threat, competing for jobs and jeopardizing private life and even possibly democracy, than a help and an ally for the future development of human life and culture. A Special Issue on successful cases of human learning-machine learning collaboration in our field, Citizen Science, could maybe also feed the collective reflection on the conditions and limits of this collaboration.