Three Hours on One Page of Orlikowski and Iacono (2001)

I had read Orlikowski and Iacono (2001) before. At least twice, maybe three times. I knew the headline claim: eighty eight percent of papers in Information Systems Research do not theorize the IT artifact. I knew the five views existed. I could name them if asked. Nominal, Tool, Proxy, Computational, Ensemble. That was enough for a conversation, enough for a study group, enough to nod along when someone mentioned it in class. But it was not enough for a comps exam, and I found this out last week when I sat down with the actual paper and a yellow highlighter.

Three hours on one page. Not the whole paper. One page. The page where the five views are defined. I read it. I highlighted a sentence. I stared out the window. I read it again. I wrote a note in the margin. I realized I had no idea what I had just read, not truly, so I read it a third time. This is a kind of reading that graduate school trains out of you. You learn to skim, to extract the contribution, to move on to the next citation. Slow reading feels like wasted time when you have a hundred and eighty four papers to cover. But the slow reading is where the actual understanding lives, and I had forgotten that.

Let me walk through what I found when I stopped skimming.

The Nominal view. This one seems easy. The paper names the technology but does nothing with it theoretically. The system appears in the title or the introduction and then disappears into the background while the real model runs on purely social constructs. I had always treated this as obvious, almost a throwaway category. But reading it slowly I realized something uncomfortable. The Nominal view is not a straw man. It is where most of IS sits, then and now. The paper mentions ERP or AI or blockchain, then proceeds to test hypotheses about trust, satisfaction, and intention that could apply to any context. Sarker et al. (2019) would later confirm this with the transport test: if you can move the model from an IT setting to a non IT setting without changing the claims, the technology was never doing theoretical work. Fifty six percent of IS research fails this test. The Nominal view sounds dismissive as a label, but it describes a majority of the discipline.

The Tool view gave me more trouble. On the surface it sounds reasonable: technology is a tool, people use tools to achieve goals. The problem is deeper than I had realized. The Tool view treats the technology as a stable, predictable instrument whose properties are fixed and whose effects are determined by how well it serves predetermined ends. This view is not wrong exactly. It is incomplete. It assumes the tool and the user are separate, that the tool does not change the user and the user does not reconfigure the tool through use. Anyone who has watched an organization implement a new system knows this is false. The tool gets bent and reshaped and worked around. The users develop new practices that the tool never anticipated. But the Tool view does not have the vocabulary for this, so its models stay clean and stay wrong.

The Proxy view was the one that clicked for me on this reading. I had never fully understood what Orlikowski and Iacono meant by proxy until I sat with the definition. The Proxy view uses the IT artifact as a stand in for something else. The paper does not theorize the technology itself. It theorizes what the technology represents: organizational capability, management sophistication, competitive pressure. When a study measures ERP implementation and calls it organizational integration, that is a proxy move. The artifact vanishes behind the abstraction. The danger is that you stop studying technology and start studying the abstraction, and the abstraction has a life of its own that may or may not correspond to what the technology is actually doing in the organization.

The Computational view was the most foreign to me because it is where I do not work. This view focuses on the internal logic of the system: algorithms, data structures, processing logic, computational power. It is the view of computer science and engineering, where the artifact is the entire object of study. Orlikowski and Iacono do not dismiss this view, but they note that it lacks organizational engagement. The computational artifact exists in a clean world of inputs and outputs, not in the messy world of social practice. This matters because a system that is optimal in the computational view can fail completely in the organizational one.

Then the Ensemble view. This is the one Orlikowski and Iacono argue for, and I had always nodded along without really feeling why it was necessary. The Ensemble view treats the IT artifact as embedded in social practice, tightly linked with human action, organizational context, and institutional structures. The technology is not separate from the context. It is shaped by the context and shapes it back. This is the emergent perspective from Markus and Robey (1988) applied to the artifact itself.

The reason the Ensemble view is harder is that it refuses to let you isolate the technology. If you take the Nominal view, you can use any theory from any discipline and call it IS research. If you take the Tool view, you can model technology as an independent variable with clean psychometric properties. If you take the Proxy view, you can work with abstractions that are easy to measure. If you take the Computational view, you stay in the lab where variables are controlled. But the Ensemble view requires you to study the technology in its full social entanglement, which means you need new constructs, new methods, and often a new ontology. It is the view that produces knowledge specific to IS, but it is also the view that is hardest to publish, hardest to design, and hardest to execute. That is why eighty eight percent of papers do not use it.

I sat with that number for a long time. Eighty eight percent. Not a minority. Not a bare majority. Almost nine out of ten papers in ISR, the field's own journal, did not treat the IT artifact as embedded in social practice. The paper was published in 2001. Twenty five years later, Sarker et al. (2019) found that only thirteen percent of IS research achieves genuine sociotechnical interaction. The number had barely moved.

This is what close reading revealed to me that skimming never could. Skimming gives you the conclusion: most papers do not theorize the artifact. Close reading gives you the mechanism: each of the four non Ensemble views is a different way of making the technology disappear, and they are not all the same. The Nominal view names it and forgets it. The Tool view uses it but flattens it. The Proxy view replaces it with an abstraction. The Computational view isolates it from organization. Each disappearance works differently, and each one requires a different corrective.

I wrote in the margin of my printout: each non Ensemble view represents a specific form of theoretical convenience. You pick the disappearance that makes your research design easiest. And the Ensemble view is the one that refuses convenience.

I think about this when I read current AI papers. The same pattern repeats. Many papers name the AI system in the title and then run a model on trust, satisfaction, or adoption that could describe any technology. Some papers use automation as a proxy for capability or intelligence. Some papers focus entirely on the algorithm. Very few treat the AI system as embedded in the organizational practices that shape it and that it reshapes. The Ensemble view is twenty five years old and we are still not doing it. That is not a critique of the field. It is a description of how hard the work actually is.