Academic conceptions of ‘internet meme’ typically assume a collection of files (e.g. jpeg or png files) that enact cultural memory: a defining interest of cultural heritage research (Assmann, 2008; Taylor, 1995; Viejo-Rose, 2015). As such, memes function as data of cultural heritage informatics: media imbued with the power of memory through iterative practices (Guan, 2008; Modrow & Youngman, 2023). Further still, memetics – the study of memes – studies dissemination, curation, classification, and organisation of culture (Dawkins, 1976; Shifman, 2014). In many cases cultural heritage informatics could easily borrow the name ‘memetics’ without altering much. Yet, this sort of immediate association of research seems unlikely. Why?

The overlap seems conceptually intuitive, but underlying the intuition is a tension in disciplinary goals. Internet memetics regularly seeks to capture descriptions of ‘new things’ in culture (Lankshear & Knobel, 2019; Smith et al., 2024): disjunction. Meanwhile cultural heritage seeks to find proper ways to canonise culture (Assmann, 2008) in relation to existing culture: conjunction. Internet memetics – despite conceptually being about memory and artefacts – is empirically resistant to the canonization of cultural connections by highlighting differences of interpretation (Lankshear & Knobel, 2019). Meanwhile, ‘cultural heritage’ implies something inherited culturally. Cultural heritage pragmatically seeks to make ‘something’ canon. Canonization of internet memes requires thoughtful responses within cultural heritage and memory institutions (Bratt & Smith, 2022; Rees, 2021; Tulloch, 2023), information studies broadly (Smith et al., 2024), but also within memetics (Pettis, 2021; Rogers & Giorgi, 2023).

In the internet research context, Pettis, Rogers, and Giorgi (2021; 2023) wonder how platform technology mediate, contextualise, and disjoin cultural memory. Instead, we turn to internet technology that conjoins the internet. Thus, we ask, ‘How can Google search algorithms mediate, contextualise, and conjoin memetic memory?’ This question allows us to return to how to canonise the dynamics and heterogeneity of the collective memory of material culture. Thus, we redeem a fundamentally connected dimension of culture that memetic research seems to resist (Lankshear & Knobel, 2019) and the cultural dynamics of memory which disciplinarily defines heritage of cultural memory (Assmann, 2008).

To illustrate the issue, consider a brief motivating example. An ‘internet meme’ is not a singular image-file nor a singular subjective memory. Rather, it is some arrangement (e.g., collection) of files, the way it moves through media, and the way it is remembered for every being who comes into contact with one or more of those image-files. Despite people frequently having claimed to have ‘seen/made/shared a meme,’ this reduces a meme to a subjective interpretation of singular, isolated files. Likewise, saying ‘I have curated a meme’ renders the meme dynamics invisible in the way they are technically and cognitively existing. Either case renders a meme without the context of how it is a meme, failing the conceptual objective of memes.

To properly ‘curate a meme’ requires representing a meme inclusive of dynamic cultural meanings, dimensions, and implications. Researchers and cultural heritage practitioners must inseparably and accurately represent both (1) dynamic cultural memory and (2) an accurate description of its digitally mediated context.

This paper approaches the dynamics of memory for static image-files and the technologies that interpret ‘memory’ for us as a kind of cultural heritage. Especially for memetics online, understanding how internet technologies interpret memetic content will shape the way we humans interpret culture within iterative expressions that emanate from memory practices. One of the most powerful technologies for the human interpretable internet is web search. Google, in particular, remains the most recognized search engine online (Bianchi, 2024). Understanding how Google conjoins the image-files that make ‘a meme’ empirically closes a gap in connecting memes’ static files to cultural heritage. In particular, the data that establishes regularities in web search and information retrieval are systematically organized through iterative, aggregative, networked, and collective habits of search. While perhaps one might suggest that ‘humans’ are absent in studying Google search, it is data humans ‘invisibly’ produce through search that provides a significant portion of Google’s regularized data for information retrieval.

Memetics is usually considered a kind of datafication of cultural information (Smith et al., 2024). Richard Dawkins (1976) coined the term meme to describe a unitary bit of cultural information which inherited from ‘mind-to-mind.’ He exemplifies this through architectural ‘arches’ as a cultural artefact in the world which is fundamentally connected to the mental representation of an ‘arch’. Later, Daniel Dennett integrated this into his theory of mind (1991/2017). Dennett illustrated how cognitive connections pass these idealised bits of culture by way of information, using a myth about how Marco Polo brought pasta from China to Italy through memory and documentation.

These early accounts of ‘meme’ were critiqued and re-theorised for digital contexts by Limor Shifman (2014) and others (see, Smith et al., 2024). Yet, memes do actively entangle artefacts and memory (Smith & Loewen-Colón, 2024), but few researchers have empirically shown how ‘internet memes’ and memory connect.

Often ‘mental’ perspectives – cognition, phenomenology, memory, knowledge, and so forth – are buried just under the data of technical objects (Rogers & Giorgi, 2023) or communication data (Shifman, 2014). Yet, meme’s information conjoins minds, albeit with more nuanced and multi-mediated ways than Dawkins’ or Dennett’s reductive ‘mind-to-mind’ descriptions. As contemporary memetic research clarifies, there are often a plurality of information technologies between ‘mind-to-mind’ mappings.

To redeem mental processes of memetics, how can we begin to understand what ‘memory’ is when the data are outside of human minds, and instead are in the world as artefacts? Previous research has suggested a relationship between cultural heritage and memes (Demsky, 2021; Modrow & Youngman, 2023). However, empirical interests of memetic cultural heritage (Demsky, 2021; Petrassi, 2024; Rees, 2021; Tulloch, 2023) usually investigate human-centric experiences without the nuance of technical mediations that conjoin minds. As such, there is a gap between internet memetic technological arguments and cultural heritage’s research on individual human experiences. Complex digitally mediated ecologies of memes (Shifman, 2014) and cultural heritage need to be re-joined to explain memetic memory (Smith & Loewen-Colón, 2024).

When memetic data are collections of image-files, the cultural heritage task is clarified. Relatedly, memory activates through these files. Despite being corollary, digital files and mental processes are not reducible to each other. Shifman (2014) suggests ‘social mindsets’ entangle inseparably with these files but remains open to how this occurs. She defines memes as ‘content, form, and/or stance’ which must travel across the internet by way of many users. Content is the ‘idea,’ form is the meme’s ‘media’, and stance is how the information enables unique contextual meanings for individuals as the meme is communicated. Despite the intentional entanglement of these concepts, internet memetics have tended towards turning ‘content’ of a meme into a plurality of meanings (Lankshear & Knobel, 2019; Smith et al., 2024) which tempts researchers into reducing Shifman’s definition to merely digital media’s immediate context (Wiggins, 2019). Doing so renders the transmission and dynamics of cultural heritage characterised within Shifman’s definition inaccessible. To redeem memory, Smith, and Loewen-Colón (2024) create initial steps in re-theorising memory for both Shifman and Dawkins.

While lots of theorisation of transmission of memetic information exists (Cappella et al., 2015; Dawkins, 1976; Hull, 2000; Smith & Hemsley, 2022; Smith & Loewen-Colón, 2024), observing transmission in digital contexts is a difficult task without near exact copies of artefacts such as ‘retweets’ or ‘share’ buttons (Sperber, 2000). This is partly why ‘viral events’ were distinguished from ‘memes’ in digital contexts (Nahon & Hemsley, 2013; Shifman, 2014). However, information transmission and meme might have nuanced connections if we could conceptualise what information constitutes a meme in some statistically regular image-file data. To fill the empirical gap between materiality of image-files and the ‘content’ of memetic memory, we must start by seeing information selection across image-files.

Theoretically, the relation of these image-files and memory have been recently updated as being connected to ‘traces’ (Smith & Loewen-Colón, 2024). Information selection of digital and cybernetic technologies are often called trace data (Geiger & Ribes, 2011; Østerlund et al., 2020). Trace data signifies movement or selection (Cappella et al., 2015; Heylighen, 1993) of information. Studying memetic traces empirically remains incompletely described, yet previous literature has plenty of suggestions.

Prior work suggests interpreting memetic data traces are digitally mediated and technologically situated (Rogers & Giorgi, 2023; Smith et al., 2022). However, Smith et al. suggests trace data can be understood through the datafication of Google image web search technologies (Smith et al., 2022; Smith & Hemsley, 2022) and across digital technologies more generally (Smith & Loewen- Colón, 2024). Google both provides traces of web indexing, but also artefactual traces ‘within’ each artefact. We distinguish artefact from jpeg or png files: memetic artefacts are data commonly identified across files. When Google identifies a frog or a cat in two different image-files, that frog or cat operates as artefactual trace data which Google considers culturally relevant for web search, indexing, and so forth. Google ‘traces’ a cat to a bunch of other cat artefacts. If Google considers this data relevant for web search, then it is this data that informs our cultural experience online.

By using Google as a method of recovering dynamics across image-files by tracing artefacts across them, we ask the following questions:

RQ1 considers technologically mediated conjunction using computational network methods and compares across three qualitatively selected memes based on what the authors deemed structurally different kinds of traces. RQ2 considers these three structurally different meme image-file collections to interpret similarity and differences (Smith & Hemsley, 2022) of artefactual traces across the files. The first question considers what the technology conjoins as the regular semiotic ‘structure’ of each collection. The second asks what we expected would be conjoined in relation to what actually was conjoined. Together these questions suggest how technologically mediated conjunction might be different from expected conjunction. We will explain how we provided these differences across three memes in the next section.

With interest in what data traces that sustain collective/distributed internet memory, we use relevant technologies that provide this datafication following Smith et al. (2022) who use a cyber- semiotic interpretation (Cannizzaro, 2016) of memetic traces. As such, our research design makes use of Google data gathered from Google Cloud Vision’s API. This provides us data that informs Google about what image files are affiliated with text search and reverse image search. These data play a large role in popular web search, shaping our digital cultural environment.

However, we are also interested in how a ‘meme’ that people curate into a collection of image documents ‘belong together’. For this we leverage the cultural ‘authority’ of Know Your Meme’s (KYM’s) image collections as other researchers have done (Katz & Shifman, 2017; Smith et al., 2022; Smith & Hemsley, 2022). KYM curates image-files into thousands of ‘meme entries,’ each of which operate as cultural heritage collections for ‘a meme.’ The following section shows how we use KYM’s meme entries to gather cultural heritage data from Google.

There are two types of networks: (1) feature networks and (2) image networks. Feature networks represent how the features (either label or web entities) are connected to each other based on common images that include both entities. Conversely, image networks show how individual images-files are connected based on common entities that are detected in both images.

In total, we see that Google’s ‘memory’ of memes as collections of image-files in some cases performs better at capturing what we expected of memetic cultural heritage, and sometimes it does not. In particular, Google understands what ‘Hello There’ and ‘Doge’ are about, however we have to see these two memes in different ways. Hello There, is understood more as a network of features by Google where Doge only seems to be understood through the dog, but what holds the meme together otherwise is not known to Google. Loss, however, seems to lose context in both cases. Essentially Google thinks ‘Loss’ is a webcomic or a generic meme on Know Your Meme called Loss. It does not give us any indication of its cultural relevance.

What our study suggests is that certain parts of cultural memory are not reducible to computationally reducible visual data. More context is needed to understand what makes a meme important. We see this in each of our meme’s network data. However, we do see that Google can provide a partial understanding of what artefactual trace data frequently ties them together. As such, we can begin to see what information selection is occurring across collections. However, in the case of Hello There and Doge, it seems Google cannot point to exactly what gets selected. That is, Google’s attempt to describe Doge leads it to various descriptions of Doge and not the cultural contexts in which Doge is used. It is even less precise about Loss. On the other hand, Hello There is more precise in its network, but it is less precise about its narrative content other than it is ‘funny’.

In relation to memetics, we have seen that some memes are understood through computational artefact networks through Google vision features. However, we have also given some insights into where it is limited. For example, it does not understand when cultural context is not within the data, or when the cultural context is strongly implied through probabilistic references in the case of Loss and Hello There.

Our literature review suggests that Google can actually play a significant role in what gets sustained in cultural heritage, and thus can shape cultural memory. As such a significant part of our cultural knowledge, how it understands or misunderstands this culture plays a part in a computationally sustained cultural heritage as trace data. Memes, when understood as ‘particulates’ of documented culture, provide a way to be granular about understanding computationally driven tools like Google Cloud Vision’s usefulness or limitations in sustaining the relevant parts of cultural memory. In order to be clear about what Google generates for cultural heritage, connections between cultural expectations of those that ‘know’ and an understanding of how the qualitative aspects of cultural heritage are bridging with computational tactics of making sense of them. While our results suggest that Google is ‘good’ at understanding certain parts of memetic meaning, we find that it needs to be augmented through human interpretation to make sense of a lot of what it provides. Google, in particular, can significantly aid in bridging new cultural heritage together however, since it pulls together a lot of information in the heterogeneity of the indexed web, so understanding such tools helps augment limited subjective views of cultural heritage work. Future work is needed to understand what computational aspects can be fully leveraged in augmenting cultural heritage work and what should be delegated to humans.