IT-Driven Transcriptions: The Hidden Politics of Language, Metaphor, and AI Objectivity

TL;DR

Language in IT is never neutral. This paper explores "transcriptions"—the process by which everyday metaphors move into computing and technical terms move into society. It argues that by calling software systems "intelligent" or "learning," we mask human subjectivity and allow systems to automate discrimination under the guise of mathematical objectivity.

Background: Beyond Code as Law

In the intersection of Gender Studies and Computer Science, Lawrence Lessig famously claimed "Code is Law." Britta Schinzel takes this further by exploring the language that builds the code. The paper positions itself as a critical deconstruction of how we name things in the digital age, moving from the purely technical into the "material-semiotic" realm.

The "Transcription" Trap: From Metaphor to Reality

The author uses the term transcription to describe more than just translation. It is a cultural process that changes the original meaning.

1. Everyday Language in IT

We use metaphors to understand the abstract:

• The Physical: "Bridges," "gateways," "clouds."
• The Biological: "Viruses," "evolutionary algorithms," "neural networks."
• The Social: "Clients," "daemons," "social networks."

These metaphors are anthropomorphizing. By giving machines human traits (like "understanding" or "seeking"), we grant them a person-like status that they do not possess. This hides the fact that these systems are essentially "syntax without semantics."

2. IT Terms in Everyday Life

Conversely, the term "Algorithm" has migrated from mathematics into the public sphere. While a mathematical algorithm is a finite, objective set of operations, modern "algorithmic systems" are actually complex software knots involving contingent data, heuristics, and subjective design choices.

(Image: Representative of the intersection between Gender and IT paradigms)

Methodology: Agential Realism and Machine Learning

Schinzel applies Karen Barad's concept of Agential Cuts. In Big Data, we "scoop" data from an infinite sea of utterances. Each time we analyze this data, we make a "cut" that defines what is visible and what is invisible.

The Death of Theory?

The paper critiques the "Digital Empirical Turn"—the belief that with enough data, we no longer need theories or models. This is dangerous because:

• Data is not a raw resource: It is not "oil" or "fuel." It is a transcription of living bodies and behaviors.
• Invisibility of Methods: Private firms hide their analytical methods, making it impossible to deconstruct why a system made a specific decision.

Experiments and Results: The Bias in the Machine

The paper highlights the "post-algorithmic" nature of Deep Learning. Unlike a traditional program, a trained neural network's code cannot be easily "re-engineered" to find the source of an error or bias.

Key Evidence of Bias:

• Linguistic Bias: Natural Language Processing (NLP) models trained on corpora like Wikipedia or Twitter inherit human-like prejudices regarding gender and race.
• Predictive Policing: Systems that "predict" crime often fail because they are trained on historical, biased arrest data, effectively "automating the past."

(Image: Visualizing the socio-technical feedback loops in IT systems)

Deep Insight: Algorithms are Tools, Software is Choice

Schinzel’s most crucial distinction is between the Algorithm and the Software System:

1. Algorithms are mathematical, universal, and essentially "neutral" instructions.
2. Software Systems are human-managed combinations of algorithms, data, and goals.

When we blame an "algorithm" for discrimination, we shift responsibility away from the human designers and the societal structures that produced the training data. By calling AI "objective," we make it harder to challenge its outputs.

Conclusion: A Call for Critical Literacy

The paper concludes that we must stop believing in the "objectivity" of AI. These systems are irreversible transcriptions of our existing world—including its flaws. To build a more ethical future, we must:

• Distinguish between mathematical tools and social software.
• Deconstruct the metaphors we use to describe "intelligence."
• Acknowledge that any data-driven system is a subjective "cut" through reality.

Takeaway: We don't just "use" IT; we are "transcribed" by it. Recognizing the gendered and moral weight of IT language is the first step toward reclaiming agency in a world governed by "objective" machines.