Limitations and Future Research
Critical Integration
This product represents a unique piece of interdisciplinary work, integrating cultural analysis, psychology, and computational methods to develop a new way of examining how visual culture encodes the intuitive actions and needs of humans. By treating images as data and applying large-scale object detection to concepts drawn from developmental affordance theory, it demonstrates how visual media can be systematically interrogated to reveal patterns that may otherwise remain invisible. In doing so, it reinforces the long-held idea that ‘a picture tells a thousand words,’ while providing a structured method to read those words through a developmental and evolutionary lens.
However, the very same methodology that enables this innovation also brings substantial limitations and the list is, by design and in typical interdisciplinary fashion, almost endless. The most significant is that every object set, frequency count, and category pattern is mediated by AI object detection, meaning the dataset is an algorithmic interpretation of reality rather than reality itself. This reliance introduces possible misclassification, cultural blind spots, and underrepresentation of symbolic or abstract imagery. A manual calibration was performed to limit this bias, see Manual Calibration Design Page.
Limitations
-
- Google Image bias – Algorithmically ranked results reflect dominant visual culture, not a balanced sample.
- Western-centric training data – AI and search engines underrepresent non-Western visual traditions.
- Simplified classification frameworks – Binary DA and Maslow categories flatten overlapping human needs.
- Category size imbalance – Corrective weighting can distort real-world distributions.
- Temporal fixity – Single-point collection prevents analysis of change over time.
- Exclusion of personal archives – Public imagery omits more authentic, situated representations.
Future Research
-
- Social media scraping – Analysing organic, user-generated imagery from platforms like Instagram, TikTok, or Flickr.
- Longitudinal tracking – Repeating searches over months/years to capture cultural, seasonal, or political shifts in visual narratives.
- Large-scale human submission – Building participatory datasets where contributors’ own images are classified into DA/needs groups.
- Colour analysis – Mapping dominant colours by search category to reveal additional details.
- Expanded search categories – Testing whether patterns hold across a much broader conceptual field.
- Cross-platform comparison – Comparing results from different search engines, stock libraries, and community archives to surface platform-specific biases.
- Cultural perceptions of gender and clothing – Using the strong detection of human figures and clothing in this project to investigate how different cultures visually construct and signal gender identity, roles, and dress norms.
In this way, the project not only delivers original insights but also provides a framework for future interdisciplinary research into how visual media reflects and shapes our most fundamental human needs, instincts, and social categories.