Our video includes physical media (pictures of historical archives we took in the library), AI-generated background images, and some screenshots in the section on contemporary child labour issues.

Let's first talk about the physical media we used, the historical archive itself. We took pictures of all the contents of the historical archive in the library and carefully analysed the contents to try to find the problems reflected in it. After combining the content of the literature in the reading list, we quickly discovered that there were missing or hidden data in the historical documents. The use of these physical media contents is to give readers a basic understanding of the properties of archives and data. In other words, it can help us think more critically about what is missing from the archives when thinking about the question "How can historical documents help us better understand the current situation?" Instead of being limited to the data in the archives. In addition, we also found child labour issues at the time in the historical archives, such as two-year-old children entering factories to work. This is to return to our main research question, that is, we found hidden data and child labour issues at the time in the historical archives, so we have to see if there is a data hiding problem in contemporary child labour data.

We then used two AI tools, CoPilot and ChatGPT, to generate more than 40 background images. We used AI to generate images for two reasons. First, it was difficult for us to find "appropriate" real photos because they were old and involved copyright issues. More importantly, for ethical reasons, we seemed not to be able to use photos of real child labourers at the time (although we searched for many).

Second, using AI can greatly save the time of making videos, so we gave up the idea of ​​making animations. Because the former allows us to output more text content within the specified time. In addition, using AI to make pictures can highlight the elements we want to convey to the audience through instructions, so that the audience can focus more on specific issues instead of being distracted by other content.

However, the method of using AI tools to make pictures is not perfect, and the problems we encountered in this process will be explained separately later.

Finally, we also used screenshots in the part about the current situation of child labour. These screenshots are from videos about child labour in Bangladesh on YouTube. We were very careful to avoid the parts containing close-ups of their faces when choosing these pictures and focused on highlighting the poor working conditions of child labourers in Bangladesh and the harm they bring to children.

Our main purpose in using this part of the media is to alert the audience that the child labour problem still exists and to pave the way for the incomplete data on child labour in India and China in the following video. In other words, we want the audience to understand that the lack or incompleteness of child labour data reports does not mean that there is no child labour problem in reality, and it may even be hidden because it involves some extremely bad work (such as human trafficking, illegal mining, prostitution, etc.).

The use of digital media was crucial in helping us to uncover new meanings in the selected archival collections. While paper factory records from early 20th century Britain provided the initial material, it was through digital tools, visualization techniques, and narrative structures that we were able to analyse, question, and reconstruct these archives and relate them to contemporary issues.

First, digital media allowed us to visually highlight patterns of omissions in historical documents. For example, by creating graphic representations of accident record data and using background images that evoked the audience’s emotions, we were able to highlight blank fields and recurring omissions (e.g., missing information about rehabilitation, dismissal, or compensation). These omissions may be missed in a simple reading of the text, but through data mapping and image generation, they become visible, triggering analytical and emotional resonance.

Second, we used visualizations and symbolic images. By transforming abstract statistics into engaging visuals, digital media enables us to turn data into stories—making our findings more understandable, more relevant, and impactful to a wider audience.

However, in hindsight, there are some areas where we could have expanded the digital dimension further. For example:

We could have worked with digital design tools to create a more consistent visual aesthetic. And avoid information overload by more selectively curating visual content. Our peers have also pointed out that our videos may have information overload issues. In addition, we may have adopted user feedback mechanisms or real-time barrage mechanisms. That is, allowing viewers to comment on any part of the video to express their opinions or questions. This approach can deepen the dialogic nature of our video presentation, consistent with the critical interpretation emphasized by our video.

In our group video, we have analysed and reconstructed historical data and images related to child labour to gain a deeper understanding of the complexity of the issue of child labour. This is not only a historical phenomenon of the past, but also an ongoing reality in the context of globalisation today. While reviewing British factory archives, we noticed that managers often recorded only data that was favourable to themselves, ignoring or even intentionally censoring the real situation of workers. This made us realise that data is never neutral and what it reflects is often a power structure.

In the process of data visualisation, we tried to transform the original information into more communicative visual content. We redesigned the original charts, adjusted the colour scheme, structure and labels, not only for the sake of aesthetics, but also to improve the readability and efficiency of the message. We have also come to realise that visual design is not just a technical issue, but also a process of "re-expression", which affects how the audience understands and feels about a social issue.

We also used artificial intelligence tools such as ChatGPT and CoPilot to generate images and charts. This has provided us with a powerful supplement when information is lacking, but it has also created new challenges. For example, if the cues were not clear enough, the AI tended to generate images that did not match the historical context, such as overly idealised factory scenes. During this process, we have come to realise that technological tools are not neutral, and their use requires constant adjustment and judgement to maintain a high level of sensitivity to historical accuracy.

In this project, we used AI image-generation technology to recreate historical scenes, which provided new perspectives on our collection, especially in the absence of authentic visual materials. Compared to abstract data or written descriptions, visual representations made the historical context more intuitive and vivid, helping viewers to emotionally engage with and better understand the lived experiences of the time.

This process also revealed the limitations of AI-generated imagery. The accuracy of the visuals depends heavily on the clarity of the input prompts. If the descriptions are vague, the AI often produces idealised or unrealistic scenes. For example, when we failed to specify the factory setting, the system generated an image of a spotless, well-lit textile mill resembling a library, with neatly dressed workers—completely detached from historical reality.

Another aspect worth reflecting on was our use of CoPilot to generate images that complemented our narrative. This experience highlighted that while AI visuals can be highly persuasive, they may also obscure historical inaccuracies. Therefore, crafting precise prompts is essential to ensure the resulting images carry historical authenticity and analytical value. Anyway, we did our best to create the video in a short time. If there is enough time, we can use 3d models, augmented reality or other high-quality visual presentation techniques to create the video.

So, digital media expanded our expressive and methodological possibilities, allowing for richer historical representation. At the same time, it introduced new challenges. The process reminded us to approach such technologies with critical awareness, remaining vigilant about the ways they might oversimplify, beautify, or distort the past—and to always prioritise historical accuracy in the use of AI tools.

This part can be regarded as some of our reflections on feedback, namely that our videos contain a large amount of data but do not provide sufficient time for the audience to fully browse through these data. All the new data visualisations below were obtained by manually extracting metadata and organising them into CSV files, and then importing them into “Datawrapper” (a free online digital visualisation tool).

Furthermore, we have made adjustments to the presentation form of these data, aiming to enhance the viewers’ comprehension of the data content and taking into account the accessibility aspect by using more contrasting colours to highlight the contrasts among the data.

The following are four visualisations of data, along with comparisons to the previous charts.

In assessment 2, we went to see the archive we chose, filmed what we considered to be valuable content, and then through active discussion and interpretation of the archive, we reached a consensus on the material and decided what we wanted to focus on in the video. Once the preliminary work was complete, we worked together to refine the script to ensure that we could deliver what we wanted to say while retaining the contributions of our team members, and to ensure that the video reflected the decisions we had made together.

If changes can be made to the presentation of the video, perhaps we filter the original complicated data and turn it into visual charts to display. However, this would have its drawbacks, if the amount of data in the chart is too large, the audience may need some time to read and understand it, visually it would become monotonous, and if the chart lacks explanations, the audience may be confused instead. Also, after watching the videos of other groups, we may learn from the presentation of other groups, that is, to make the content more interactive, for example, to pick out specific cases from the data we collected and make them into a story, so that more humanistic cases may be able to emotionally resonate with the audience. However, this may not be able to present enough data-based content in the required time, other content may be compressed, and the video could be less persuasive and authoritative.