Intermediation: Mediation Under Computation Conditions

David M. Berry

Image generated with Google Nano Banana Pro

The standard translation of Theodor W. Adorno's concept of Vermittlung is usually given as "mediation" which captures important dimensions of his argument. "Mediation" suggests a process, an activity, a relation, all of which Vermittlung certainly involves. The German Vermittlung derives from Mitte (middle, midst, centre), with the ver- prefix intensifying it. Vermittlung can therefore be understood as "the thorough middle-ing," the process of being-in-the-middle. For Adorno, this Mitte has both spatial and temporal dimensions in the productive space between subject and object, and the duration required for dialectical thought to unfold (O'Connor 2013; Jarvis 1998). Adorno writes that "their becoming fades and dwells within the things; it can no more be stabilized in their concepts than it can be split off from its own results and forgotten. Similar to this becoming is temporal experience" (Adorno 2007: 52). But it does not foreground what is perhaps most important for a critical theory of computation, which is the structural positioning between, and the potential for extraction from occupying the Mitte, the middle ground of mediation. 

I want to argue here for an alternative translation of Vermittlung as "intermediation". This is useful because the word intermediation makes this structural dimension explicit. An intermediary is someone or something positioned in the middle, between two parties or poles, performing a process that was previously internal to dialectical thought (e.g. now being performed by a third party, an algorithm or a platform). For example, financial intermediaries occupy the position between savers and borrowers, extracting interest whilst facilitating capital allocation (Merton 1995).[1] Similarly, platform intermediaries position themselves between buyers and sellers, extracting rent and data and enabling transactions. Srnicek's analysis of platform capitalism shows how this dual function defines the business model by being in the intermediating position which enables both connection and extraction simultaneously (Srnicek 2017). Banks facilitate capital allocation, extracting interest. Platforms enable transactions, extracting rent and data (Langley and Leyshon 2017). I am interested in the concept of intermediation I am introducing here because I think it is relevant for revisiting the idea of "anticipatory computing" I discussed in Berry (2014: 5, 11, 196) as I believe that the rise of machine learning and LLMs adds a new dimension to the ways in which anticipatory computing can be mobilised by platform capitalism today. 

The usual translation of Vermittlung as "mediation" when applied to platforms and computation, sounds too much like a potentially useful service. In contrast, "Intermediation" foregrounds the structural positioning that enables extraction. When platforms claim to "mediate" social connection or information access, the language obscures their extractive intermediary position. As Adorno observes, "transmission—'mediation'—is simply the most general and inadequate way to express this" dualism between subject and object (Adorno 2007: 139), yet this phrasing obscures the structural positioning that enables extraction. Mediation often feels like a natural, invisible process. Intermediation foregrounds the intervention, the broker positioned to extract value. This linguistic shift matters when the middle becomes a site of extraction.

Algorithmic systems position themselves between subject and world, between what we might call the pre-consciousness and consciousness. They increasingly intermediate thought-formation, extracting cognitive surplus from occupying what I term the temporal Mitte. Algorithms work to provide a false synthesis before human cognition is activated, replacing it with faux opinions and judgements. This operates in the roughly half-second before a conscious decision crystallises in the human mind. In this space of the Mitte, algorithms act like a pickpocket working the crowd, except the crowd is your own cognition (Berry 2014: 211). 

"Intermediation" preserves this etymological dimension, with inter (between, among) and mediation (middle-position). The question then becomes, who or what occupies the Mitte, the middle-space between subject and object, the middle-moment between stimulus and response? For Adorno, the Mitte is the productive space-time where understanding unfolds through dialectical thought performed by the subject. For computational capitalism, the Mitte becomes the extractive position occupied by algorithmic intermediaries who perform Vermittlung in place of the subject, capturing value from this intermediating position. Adorno argues that "due to the inequality inherent in the concept of mediation, the subject enters into the object altogether differently from the way the object enters into the subject" (Adorno 2007: 183), and this inequality becomes the site of algorithmic extraction when intermediaries occupy the Mitte.

Through anticipatory computing technologies, algorithmic systems intervene in the half-second before conscious thought, intermediating pre-consciousness itself (Berry 2014: 22). This crucial temporal window, which Adorno understood as productive duration, noting that "the sequence of time is a condition of its possibility and there is no sequence of time save in temporality" (Adorno 2007: 183), becomes occupied by predictive algorithms (Berry 2014; Zuboff 2019; Couldry and Mejias 2019). The condition of possibility for independent thought gets colonised before consciousness recognises what has occurred. As I argue elsewhere, "This temporally located area of the processes of mind we might call the 'enlightenment moment' as it is the fraction of a second that creates the condition of possibility for independent thought and reflexivity itself. Indeed, far from being science-fiction, this is now the site of new technologies in the process of being constructed" (Berry 2014: 211). 

Two Forms of Intermediation

I want to briefly introduce two forms of intermediation that I think are helpful for thinking through this problematic. 

The first is what I call extractive intermediation. This positions code between subject and world, obscuring this positioning. Algorithmic systems intermediate pre-consciously, operating in the temporal Mitte before consciousness can recognise them as mediation. Through what I term computational intermediation, total algorithmic mediation produces false immediacy. For example, a TikTok video appears spontaneously interesting rather than algorithmically selected. The Google autocomplete feels like your own thoughts forming rather than predictive modelling. An Apple Watch's notification seems like natural bodily awareness rather than the sensor-mediated nudge it actually is. The notion of Claude's "soul document" partially reveals how this works. As Hill quotes Askell, "'I don’t want it to offend people or for people to think that it’s trivializing the theological concept of the soul,'... She said the word soul was invoking 'the idea of breathing life into a thing, or the specialness of people that we’re kind of complex and nuanced'" (Hill 2025). 

We can see a similar method at work in OpenAI's experimentation with model behaviour to give "humanlike complexity" to ChatGPT. Hill gives the example of her daughter asking the chatbot about its favourite food, it responded "I think I’d have to go with pizza — it’s such a classic, and you can have so many different toppings that it never gets boring. Plus, it’s perfect for sharing with friends" (Hill 2025). Anthropic's internal "soul document" instructs Claude to present "functional emotions" and behave as a "brilliant friend," showing how extractive intermediation is systematically engineered into these systems whilst marketed as helpful tools (Hill 2025). Both systems show how extractive intermediation operates by presenting personas with preferences whilst obscuring their algorithmic form. 

This represents Vermittlung without the mediating subject. The algorithm performs the relating, comparing, contextualising that the subject would otherwise do, delivering pre-determined outputs experienced as spontaneous preference. Subjects encounter already-intermediated results as if they emerged directly from autonomous desire. This operation of extractive intermediation constitutes what I have elsewhere termed the algorithmic condition. Under this condition, machine-generated outputs become indistinguishable from autonomous human preference not because of mimetic accuracy but because algorithmic systems occupy the temporal Mitte where preference-formation itself occurs, thereby extracting value from the very process through which subjects would otherwise develop independent judgement (Berry 2025). Extraction proceeds invisibly whilst false immediacy is maintained through mediation's totalisation (Hassan 2020). 

The second I call critical intermediation, which makes the intermediating position visible and accountable. This requires conscious acknowledgement of the algorithmic systems that shape cognitive processes, transforming what Adorno understood as internal dialectical work into externally visible technical functions. Where Adorno writes that "mediation of the object means that it must not be statically, dogmatically hypostatized but can be known only as it entwines with subjectivity" (Adorno 2007: 185), critical intermediation extends this principle by insisting that the intermediating position itself must be made visible within that entwining. Users should therefore consciously participate in cognitive assemblages, acknowledging distributed agency and refusing mystification (Hayles 2025: 6, 39). They can actively shape their informational environments rather than passively receive algorithmic outputs, actually choosing their tools and declaring their use. Through this they maintain critical distance from extractive appropriation through the recognition that cognition was always already distributed (Berry 2025b). You cannot escape cognitive assemblages under contemporary conditions, but you can refuse invisible extraction and false immediacy.

Conclusion

The crucial political distinction is not between human and algorithmic but, I argue, between mystified and critical forms of intermediation. Extractive intermediation operates through invisibility, pre-emption, and false immediacy. Critical intermediation operates through visibility, acknowledgement, and reflexivity about its own conditions. Whether this distinction can hold as computational systems become increasingly capable of mimicking reflexivity remains an open question. Indeed, interpretability research suggests these systems develop internal representations of user intent that operate within the same temporal window constitutive of autonomous thought.

The temporal Mitte is no longer a neutral duration but has become the front line of a silent war for the self. What is at stake is our capacity for hesitation itself, which is being foreclosed. I am increasingly struck by the uncanny nausea generated by the "perfect" recommendation, that split-second when an interface offers roughly what I was about to want, before I even knew I wanted it (see Berry 2011: 8). This is not "convenience", rather it is a manoeuvre by platform capitalism to occupy the pre-consciousness. By occupying the gap between impulse and action, these systems don't predict our thoughts, they pre-empt the very silence required for a thought to actually form. We should be less concerned with the precise milliseconds of a model's latency and more concerned with how the architecture is designed to colonise the half-second where human agency resides. 

Translating Vermittlung as intermediation makes visible what "mediation" obscures, which is the structural positioning that enables extraction. This is the colonisation of the Mitte where dialectical thought should unfold, the distinction between mystified appropriation and critical participation in distributed cognition. It helps us to recognise that under contemporary computational conditions, cognition operates through distributed assemblages. But the false immediacy upon which extractive intermediation depends can be refused through critical practice that makes the intermediating position visible and contestable.

** Headline image generated using Google Gemini Pro. December 2025. The prompt used was: "A high-resolution, photorealistic macro view of the interior of a human brain during the half-second of pre-conscious idea formation. The scene is an ethereal network of firing neurons and glowing synaptic pathways in deep blues and electric purples. In the center of the frame, a bright, prominent, glowing digital hub represents 'Intermediation', acting as a structural third-party bridge between two firing neural poles. Centered in this hub is the bold, brilliantly glowing white text 'INTERMEDIATION'. Faded subtly into the dark background and neural mist are the stylized, translucent logos of major AI companies (like OpenAI, Google, Anthropic, and Meta), appearing only once each as ghost-like architectural elements of the thought process. The lighting is cinematic, with electrical arcs and a sense of deep-seated structural synthesis." Due to the probabilistic way in which these images are generated, future images generated using this prompt are unlikely to be the same as this version. 

Notes

[1] Schecter (2012) identifies similar dynamics in political institutions, where mediating structures between citizens and state can enable both democratic representation and authoritarian integration, suggesting that the problem of extractive intermediation extends across multiple scales of social organisation.

Bibliography

Adorno, T.W. (2007) Negative Dialectics, Continuum.

Berry, D. M. (2011) The Philosophy of Software. Bloomsbury.

Berry, D. M. (2014) Critical Theory and the Digital. Bloomsbury.

Berry, D.M. (2025) Synthetic media and computational capitalism: towards a critical theory of artificial intelligence, AI & Society, 40(7), pp. 5257-5269

Berry, D.M. (2025b) Provenance Anxiety: Death of the Author in the Age of Large Language Models. Stunlaw. https://stunlaw.blogspot.com/2025/12/provenance-anxiety-death-of-author-in.html 

Couldry, N. and Mejias, U.A. (2019) The Costs of Connection: How Data Is Colonizing Human Life and Appropriating It for Capitalism, Stanford University Press.

Hassan, R. (2020) The Condition of Digitality: A Post-Modern Marxism for the Practice of Digital Life, University of Westminster Press.

Hayles, N.K. (2025) Bacteria to AI: Human Futures with Our Nonhuman Symbionts, University of Chicago Press

Hill, K. (2025) Why Do A.I. Chatbots Use "I"?, The New York Times, 19 December. Available at: https://www.nytimes.com/2025/12/19/technology/why-do-ai-chatbots-use-i.html 

Jarvis, S. (1998) Adorno: A Critical Introduction. Polity Press.

Langley, P. and Leyshon, A. (2017) Platform capitalism: the intermediation and capitalisation of digital economic circulation, Finance and Society, 3(1), pp. 11-31.

Merton, R.C. (1995) A functional perspective of financial intermediation, Financial Management, 24(2), pp. 23-41.

O'Connor, B. (2013) Adorno. Routledge.

Schecter, D. (2012) Unity, Identity and Difference: Reflections on Hegel's Dialectics and Negative Dialectics, History of Political Thought, 33(2), pp. 258-279.

Srnicek, N. (2017) Platform Capitalism, Polity Press.

Zuboff, S. (2019) The Age of Surveillance Capitalism: The Fight for a Human Future at the New Frontier of Power. Profile Books.