While research and investigations share many similarities, there are also key differences that set them apart. An investigation aims to scrutinize the investigative subject, to discover something hidden or secret, and then tell a story about it. The see:set tool was developed to allow scrutiny and introspection.

When researchers investigate datasets, we are not doing this within the context or field in which the dataset was created, but instead, we engage with the dataset on our turf. Datasets are created with certain assumptions, values, and constraints in mind. More often than not, those are implicit rather than explicit. A properly structured investigation recognizes that nothing exists in a vacuum but always is the product of power relations and interests. Taking an adversarial standpoint will unearth the explicit qualities of datasets and the implicit ones.

From the beginning, it is necessary to acknowledge and continually revisit the following concerns:

Initially, all data is just information. How we engage with this information will shape our results. The same information, examined with different questions in mind, can yield other answers. To transform information into reliable data that can address our questions, we must carefully review and authenticate this information through a verification process.

Data verification processes start as a creative and iterative endeavor. It’s the time to learn and “get a feel” for what we are dealing with. Rarely can we precisely predict how this process will unfold from the outset. We adjust the verification process by using iterative feedback loops as we interact with the data and narrow down answers to our questions.

At the beginning of an investigation it is tempting to immediately jump to the concerns of which tools to use, almost as if they would solve all our problems and not create new ones. This is a mistake. While tools can be the start and output of an investigation, focusing on them too deeply can distract from the real goal – insight and facts. We have to model and pick our tools following our questions and process, and not the other way around. We place much prominence on technology and technological artifacts and often forget that they are means to an end.

The see:set tool started its life to solve challenges we faced – how to introspect and explore datasets and make them comparable. Datasets are meant for machine consumption, not human examination. We used see:set to engage with ten different datasets and 80 Million data points.

Not every engagement took place in see:set – one tool rarely satisfies all required functionality and interface needs. From the beginning, see:set focused on building a foundational information architecture to store and query data, allowing us to ask different questions based on search. While the user interface of see:set was aimed at exploring and browsing data, we also adopted a workflow where we used spreadsheets as a second interface to data. We always considered see:set a team player by fostering to export data that can be utilized in other tools.

Investigating datasets is a method as well as a critical practice. If anything, it is an antidote to suspending one’s disbelief, which has been the common reaction recently to the ubiquity of generative models.