Table 1
Summary of the categories of AI used in citizen science and their applications. (The list of categories is not ranked in terms of importance.)
| Description of instances where AI is applied | Types of AI | Examples of citizen-science software-applications |
|---|---|---|
 Applied use and impact: Assisting or replacing humans in completing tasks | ||
| Improving image or audio classification | Computer vision and computer hearing | Computer vision and computer hearing can be applied to photographic images (e.g., from cameras that are triggered by motion detection) or acoustic data, to assist or replace citizen scientists in classifying images or sounds for species detection and identification (Parham et al. 2018). Examples include citizen science biodiversity project iNaturalist (Joppa 2017; Van Hon et al. 2018); improvement of species monitoring and automatic annotation of previously collected data on undescribed or undiscovered species (Sun et al. 2017; Sullivan et al. 2018); and automatic detection of acoustic events such as bat vocalisations from audio recordings (Mac Aodha et al. 2018). |
| Accelerating the digitization of biodiversity research specimens | Computer vision and computer hearing | In digitising museum specimens, computer vision can assist citizens with tasks related to identifying labels, sorting handwritten versus typed labels, capturing label data, parsing information into field notes, normalising data, and minimising duplication. Examples include Leafsnap, for the identification of tree species in the North-eastern United States (Kumar et al. 2012); SPIDA, for the identification of one family of Australasian ground spiders (Russell et al. 2007). |
| Verifying the accuracy and consistency of contributors’ submissions | Automated reasoning and machine learning | The citizen-science biodiversity projects eBird (Sullivan et al. 2014) and iNaturalist. |
| Providing more rapid response to complex modern problems | Automated reasoning and machine learning | The citizen-science monitoring project Citclops for early warning of harmful algal blooms (Ceccaroni et al. 2018). |
 Applied use and impact: Influencing human behaviour | ||
| Extend social impact of citizen science | Robotic systems | A community-oriented robotic system designed to extend the social, educational, economic, and health benefits from citizen science to a more general public (Joshi et al. 2018). |
| Using social media for collaborative species identification and occurrence | Natural language processing, Knowledge representation and ontologies | Using specific social media to engage participants in contributing their observations over a long time-period (Deng et al. 2012). |
 Applied use and impact: Improving insights | ||
| Training of computer-vision and computer-hearing algorithms using citizen-science data | Computer vision and computer hearing | Data collected by citizens are used by knowledge engineers, people who integrate knowledge into computer systems to solve complex problems normally requiring a high level of human expertise, to train AIs. Examples include citizen-science biodiversity projects iNaturalist (Van Horn et al. 2018), Leafsnap and Pl@ntNet (as discussed in Bonnet et al. 2016). |
| Facilitating sharing the meaning of terms | Knowledge representation and ontologies | Citizen-science associations and projects based in the US, Europe, and Australia working together to design an ontology to represent knowledge in the domain of citizen science (Storksdieck et al. 2016). |
| Mining social-network data | Natural language processing | Citizen science projects can collect and analyse Twitter/Google data about health or the environment. An example is Aurorasaurus, a project to collect auroral observations (MacDonald et al. 2015). |
Table 2
Summary of new applications of AI in citizen science likely to appear in the near future.
| Description of instances where AI is likely to be applied | Types of AI | Examples of citizen science software applications |
|---|---|---|
| Applied use and impact: Assisting or replacing humans in completing tasks | ||
| Filtering out hard, repetitive, routine, or mundane tasks | Automated reasoning and machine learning | Software applications that allow citizen scientists to focus on more engaging tasks, for example, focusing on observations of interactions, or developing/contributing to innovative projects in the field. |
| Providing training/support | Automated reasoning and machine learning | AI systems that can be used in regions where citizen science training/support by humans is limited, such as when direct access to people with expertise is limited and/or human-language barriers exist. |
| Identifying species | Computer vision and computer hearing | AI tools that can instantly classify species based on images or sounds. |
| Applied use and impact: Influencing human behaviour | ||
| Describing and formally representing the domain of citizen science in all languages | Knowledge representation and ontologies | An ontology that can facilitate the creation of new citizen science applications in any language and the translation of existing applications into any language. |
| Making information and data more accessible in citizen science applications | Automated reasoning and machine learning; Natural language processing | Applications using machine learning and natural language processing to overcome information overload in citizen science platforms. |
| Providing an easy, engaging, and enjoyable citizen scientist experience with AI-based virtual assistance | Automated reasoning and machine learning | Virtual/simulated environments, in which citizens interact with AI to test tasks before real-world deployment. |
| Notifying citizens about what is likely to occur near them or what/when they could observe | Automated reasoning and machine learning | Mobile apps providing satellite-based information to citizen scientists (e.g., satellite-overpass maps). Applications that provide contextual information to citizens: What is measured, why, when, and where. |
| Adaptively managing and changing citizen science activities | Automated reasoning and machine learning | Trigger service for citizens to measure at certain times/frequencies (e.g., measuring at a satellite overpass or triggering a measurement for a certain monitoring request). Environmental data can be used to change the frequency or moment of monitoring by citizens, for example when an AI detects that there will be no satellite coverage due to cloud presence and alerts citizens to provide more observations in that particular time and location. AI models that benefit from information theory and statistics to help to prioritise effort in field work. |
| Motivating citizen scientists to participate | Automated reasoning and machine learning | Applications providing personalised reward models for making tools appealing to users. AI that optimises reward models to reflect the personality of the individual. Applications introducing context, information requirements, and gamification aspects. |
| Providing personalised notifications to increase engagement | Automated reasoning and machine learning | Notifications about collecting or analysing data, which are provided when and where appropriate and with personalised frequency. |
| Applied use and impact: Improving insights | ||
| Improving data quality control | Automated reasoning and machine learning | Applications that provide means to quality control data using cross checks between citizen science and other in-situ methods to address issues in the data that cannot be addressed by internal quality control (e.g., combining citizen data with satellite data). |
| Validating outputs through automatic procedures | Automated reasoning and machine learning | Machine-learning algorithms trained to filter out irrelevant data. |