FAIR Science for Social Machines: Let's Share Metadata Knowlets in the Internet of FAIR Data and Services

In a world awash with fragmented data and tools, the notion of Open Science has been gaining a lot of momentum, but simultaneously, it caused a great deal of anxiety. Some of the anxiety may be related to crumbling kingdoms, but there are also very legitimate concerns, especially about the relative role of machines and algorithms as compared to humans and the combination of both (i.e., social machines). There are also grave concerns about the connotations of the term "open", but also regarding the unwanted side effects as well as the scalability of the approaches advocated by early adopters of new methodological developments. Many of these concerns are associated with mind-machine interaction and the critical role that computers are now playing in our day to day scientific practice. Here we address a number of these concerns and provide some possible solutions. FAIR (machine-actionable) data and services are obviously at the core of Open Science (or rather FAIR science). The scalable and transparent routing of data, tools and compute (to run the tools on) is a key central feature of the envisioned Internet of FAIR Data and Services (IFDS). Both the European Commission in its Declaration on the European Open Science Cloud, the G7, and the USA data commons have identified the need to ensure a solid and sustainable infrastructure for Open Science. Here we first define the term FAIR science as opposed to Open Science. In FAIR science, data and the associated tools are all Findable, Accessible under well defined conditions, Interoperable and Reusable, but not necessarily "open"; without restrictions and certainly not always "gratis". The ambiguous term "open" has already caused considerable confusion and also opt-out reactions from researchers and other data-intensive professionals who cannot make their data open for very good reasons, such as patient privacy or national security. Although Open Science is a definition for a way of working rather than explicitly requesting for all data to be available in full Open Access, the connotation of openness of the data involved in Open Science is very strong. In FAIR science, data and the associated services to run all processes in the data stewardship cycle from design of experiment to capture to curation, processing, linking and analytics all have minimally FAIR metadata, which specify the conditions under which the actual underlying research objects are reusable, first for machines and then also for humans. This effectively means that-properly conducted-Open Science is part of FAIR science. However, FAIR science can also be done with partly closed, sensitive and proprietary data. As has been emphasized before, FAIR is not identical to "open". In FAIR/Open Science, data should be as open as possible and as closed as necessary. Where data are generated using public funding, the default will usually be that for the FAIR data resulting from the study the accessibility will be as high as possible, and that more restrictive access and licensing policies on these data will have to be explicitly justified and described. In all cases, however, even if the reuse is restricted, data and related services should be findable for their major uses, machines, which will make them also much better findable for human users. With a tendency to make good data stewardship the norm, a very significant new market for distributed data analytics and learning is opening and a plethora of tools and reusable data objects are being developed and released. These all need FAIR metadata to be routed to each other and to be effective.