1. Research Question
What changed that made artificial intelligence different from the technologies that came before it?
Experiment 04 established that technology was becoming labour. Across successive technological periods, technology had moved progressively further into the labour process. Tools extended human capability. Machines absorbed physical activity. Computing absorbed calculation and information processing. Digitisation converted increasing portions of human and institutional activity into machine-readable information. Software encoded processes, and automation connected those processes.
Technology was no longer simply assisting humans in performing labour. Increasingly, technological systems were performing parts of the labour themselves.
But humans remained substantially positioned around the technology. Humans initiated processes, interpreted information, authorised decisions and executed actions. Technology was performing increasing amounts of labour, but humans remained substantially responsible for initiating, authorising and interpreting execution.
Experiment 05 examined what changed when artificial intelligence entered this trajectory.
2. Research Context
The initial hypothesis was that the distinguishing characteristic of artificial intelligence was intelligence. Contemporary AI systems can perform activities historically associated with human cognitive labour. They can read, write, analyse, classify, generate, translate, code and increasingly perform forms of reasoning.
This appeared significant because the historical progression identified in the preceding experiments had moved from the mechanisation of physical labour towards the computational absorption of cognitive labour. If technology had progressively absorbed physical, computational and procedural labour, AI appeared to represent the next step: the absorption of cognitive labour.
However, intelligence alone proved insufficient to explain the change.
An AI model could produce an extraordinary answer without possessing any independent capacity to produce a consequence in the world. A model could recommend that a payment be made, but recommendation did not move money. It could identify the strongest candidate for employment, but identification did not hire that candidate. It could determine that inventory was running low, but determination did not replenish the inventory.
The distinction was therefore between cognition and consequence.
For cognition to produce consequence, something had to connect the computational output to systems capable of acting upon it. This shifted the research away from the AI model itself and towards the technological environment surrounding it.
3. Research Method
The experiment used a systems-mapping approach. Rather than treating artificial intelligence as an isolated technology, I examined the infrastructure required for AI systems to operate and the systems through which their outputs could interact with institutions, markets and the physical world.
The analysis traced the development and convergence of semiconductor infrastructure, data centres, telecommunications and fibre networks, cloud computing, databases and data infrastructure, digital identity, enterprise software, application programming interfaces, payment infrastructure, smartphones and connected devices, sensors, automation systems and robotics.
These technologies did not emerge simultaneously or for a single purpose. They developed across different industries and historical periods. The relevant question was therefore not whether any individual technology represented a transformation in execution, but what became possible when these previously separate systems were connected.
The analysis produced a four-layer model of the technological environment surrounding contemporary artificial intelligence.
The first layer was the physical layer. This consisted of the material infrastructure without which computational systems cannot exist, including energy systems, semiconductors, data centres, telecommunications networks, devices and satellites. Artificial intelligence therefore begins with material infrastructure. Although AI is frequently discussed as software, its operation depends upon electricity, processors, cooling systems, buildings, transmission networks, minerals, manufacturing capacity and communications infrastructure.
The second layer was the compute layer. This consisted of CPUs, GPUs, specialised accelerators, servers and cloud infrastructure. Advances in machine learning are inseparable from the availability of increasingly large quantities of computation. The development of modern AI has therefore depended not only upon improvements in algorithms but upon the industrial capacity to train and operate increasingly computationally intensive models.
The third layer was the data layer. This consisted of databases, storage systems, pipelines, records, identity systems, datasets and logs. Decades of digitisation have produced machine-readable representations of increasingly large portions of human, organisational and economic activity. Transactions, communications, identities, locations, inventories, financial records, organisational processes and behavioural information can increasingly be represented computationally.
The fourth layer was the execution layer. This consisted of the systems through which computational outputs can become actions. These included APIs, payment infrastructure, enterprise software, automated workflows, digital identity systems, compliance systems, software integrations and robotics. These technologies create pathways between computational determination and external consequence.
4. Research Findings
The central finding of the experiment was that AI did not arrive alone. It arrived inside an infrastructure capable of giving cognition reach.
A model that can reason is powerful. Reasoning alone, however, does not produce consequence. When a model is connected to data, it can interpret a situation. When it is connected to identity, it can determine who or what it is interacting with. When it is connected to software, it can participate in organisational processes. When it is connected to APIs, it can interact with other computational systems. When it is connected to payment infrastructure, economic transactions become possible. When it is connected to robotics, computational decisions can produce physical effects.
The significance of artificial intelligence therefore cannot be understood by studying model capability alone. The relevant unit of analysis is increasingly the model embedded within an infrastructural system.
The infrastructure surrounding artificial intelligence also displayed a recurring set of characteristics. It was resource-intensive, requiring substantial energy, computational capacity and physical infrastructure. It was environmentally coupled because computational expansion depends upon electricity generation, water, cooling, minerals, land and physical supply chains. It was capital-concentrated because the construction of advanced semiconductor, cloud and data-centre infrastructure requires levels of investment available to a relatively small number of corporations and states. It was geopolitically entangled because semiconductors, minerals, manufacturing capacity, energy systems, telecommunications infrastructure and supply chains cross jurisdictions and increasingly form part of strategic competition. It was interdependent because the functioning of each layer depended upon the availability and reliability of others.
Finally, it was path-dependent. Once organisations and societies reorganised important activities around these systems, withdrawing from them became increasingly costly and difficult. This connected the experiment to an earlier observation from the study of labour systems: irreversibility. Systems become difficult to reverse when other systems reorganise themselves around their existence.
I called this combination of technological capability, material dependency, systemic interconnection and increasing irreversibility the Infrastructure Condition.
5. Research Implications
Experiment 04 established that technology was becoming labour. Experiment 05 showed that artificial intelligence was entering a technological environment capable of connecting machine cognition to systems of execution.
This meant that AI could not be adequately understood as an isolated model or as simply the next technology to absorb another category of human labour. Its significance emerged from the relationship between the model and the infrastructure surrounding it.
Compute gave AI capacity. Data gave it context. Networks gave it reach. Software gave it pathways. APIs gave it access. Digital identity gave it participants. Payment infrastructure gave it economic consequence. Robotics could give it physical consequence.
No individual component constituted the transformation. The transformation emerged from their convergence.
The implication was therefore that the relevant object of research was changing. The question was no longer only what a model could produce. It was what a technological system containing that model could produce.
This also changed the significance of human involvement. If technology was becoming labour while infrastructure was simultaneously creating pathways from machine cognition towards consequence, then the remaining human role around execution could no longer be treated as a permanent feature of the system.
The Infrastructure Condition therefore established the environment in which that boundary could begin to move.
6. Conclusion
The experiment began by asking what made artificial intelligence different from the technologies that preceded it.
The answer was not artificial intelligence alone. The difference was the infrastructural environment into which AI arrived.
Previous technological systems had progressively absorbed components of human labour. AI emerged at a moment when physical infrastructure, computation, digitised data, networks and execution systems had become sufficiently developed and interconnected for machine cognition to acquire increasing operational reach.
This produced a different question.
The question was no longer simply:
What can AI know?
Nor even:
What can AI do?
It became:
What happens when machine cognition is connected to infrastructure capable of turning its outputs into consequences?
That question became the basis of the next experiment.
7. References
Hughes, Thomas P. (1983). Networks of Power: Electrification in Western Society, 1880–1930. Johns Hopkins University Press.
Hughes provides the historical foundation for understanding the Infrastructure Condition because he treats technological development not as the history of isolated inventions, but as the construction of large technological systems. Technologies become consequential through their relationships with physical infrastructure, organisations, capital, institutions and other technologies. His account of electrification also demonstrates how technological systems acquire momentum as societies organise themselves around their existence, making their development increasingly difficult to reverse.
Arthur, W. Brian (2009). The Nature of Technology: What It Is and How It Evolves. Free Press.
Arthur’s account of technologies as combinations of existing technologies provides a theoretical basis for understanding the Infrastructure Condition as an emergent convergence rather than a single invention.
Edwards, Paul N. (2010). A Vast Machine: Computer Models, Climate Data, and the Politics of Global Warming. MIT Press.
Edwards demonstrates how knowledge and computational capability become inseparable from the infrastructures that collect, process, standardise and circulate information.
Star, Susan Leigh, and Karen Ruhleder (1996). “Steps Toward an Ecology of Infrastructure: Design and Access for Large Information Spaces.” Information Systems Research, 7(1), 111–134.
Star and Ruhleder’s relational conception of infrastructure is particularly relevant to the claim that infrastructure becomes visible through dependency, embeddedness and the relationships formed around it.
Brynjolfsson, Erik, and Andrew McAfee (2014). The Second Machine Age: Work, Progress, and Prosperity in a Time of Brilliant Technologies. W. W. Norton & Company.
Brynjolfsson and McAfee provide broader context for the movement of digital technologies into cognitive tasks and the economic consequences of increasingly capable computational systems.