Search

Citizen science projects set up in research fields such as astronomy, ecology and biodiversity, biology, and neuroimaging produce large datasets; thus they hold promise for applying artificial intelligence for the social and environmental spheres. Human-machine integration in citizen science can harness the contributions of many human observers and use machine learning (ML) to process their contributed data. Several citizen science projects have designed complex human-machine systems, taking advantage of the complementarity of the strengths of humans and machines, and aiming to optimize for efficiency and human engagement. Using document analysis of 12 citizen science projects deploying ML techniques to optimize classification tasks, we describe the distribution of work between citizens and researchers and between humans and algorithms, as well as configurations of human-in-the-loop. The results indicate that experts are involved in every aspect of the loop, from annotating or labeling data to giving them to algorithms to train and make decisions from such predictions. Experts also test and validate models to improve accuracy by scoring their outputs when algorithms are not able to make the right decisions. While experts are the humans mainly involved in the loop, citizens are also involved at various stages of the process. We present three main examples of citizens-in-the-loop: (a) when algorithms provide incorrect suggestions; (b) when algorithms do not know to perform classification, and (c) when algorithms are active learners. We contend that, unlike automated systems that tend to remove or reduce the need for humans, the examined projects are heteromated systems that do not function without the indispensable human mediation of engaged citizens.

Human communication is profoundly social. In social activities, it is not uncommon that people are understood in many different ways, which may have various consequences for social and interpersonal communication. This thesis aims to investigate how understanding is signalled, detected, handled, and resolved in social interactions of varying complexity in intercultural, multimodal, and video-mediated communication situations. The analytical focuses are on micro-feedback and meaning repair, using an interactional approach based on theories of social communicative activity type, meaning and implicature, contextualisation, and relevance. Understanding issues are explored in audio- and video-recorded data of a spontaneous communication activity in first encounters and an educational activity with collaborative learning tasks. The results show that unimodal head movement is exclusively used to signal sufficient understanding. Head forward, eyebrow and gaze movement, smile, chuckle, and laughter can indicate understanding problems. Sufficient understanding is associated with short and medium duration of micro-feedback, and non-understanding is usually associated with a rising pitch contour. Misunderstanding does not occur as frequently as predicted in intercultural communication or video-mediated communication and is difficult for the interlocutors to detect. When information is repeated, paraphrased, or responded to with unanticipated actions or when nod in combination with “yeah” is associated with hesitation and uncertainty, a misunderstanding may have occurred. All the detected understanding problems are handled by means of meaning repair either self- or other-initiated but always self-performed. Video mediating technology does not seem to affect understanding, however, face-to-face communication provides better chances of detecting, handling, and resolving understanding problems. Apart from enhancing the theoretical understanding of understanding in real-time communication, the empirical findings also add to the foundation for practical design of technology enhanced education and communication, for example, online and flexible learning and digital communication.