Shared (Mis)Understandings and the Governance of AI: A Thematic Analysis of the 2023-2024 Oversight of AI Hearings

1. Problem Statement

The paper addresses a critical gap in the governance of Artificial Intelligence (AI): the influence of industry narratives on early legislative deliberations. Despite the rapid proliferation of Generative AI (GenAI) and its associated risks (environmental degradation, labor exploitation, surveillance capitalism, and bias), regulatory frameworks are being shaped by "shared (mis)understandings" of the technology.

The core problem is that dominant discourses, largely driven by Big Tech, frame AI as an inevitable, universally beneficial, and uniquely "American" force. These narratives obscure the material realities of AI (such as its reliance on non-consensual data extraction and massive energy consumption) and delegitimize regulatory interventions. The author argues that these shared understandings function as a mechanism of power, allowing industry actors to shape the political landscape before concrete regulations are established, effectively prioritizing technological development over public safety and equity.

2. Methodology

The study employs a qualitative thematic analysis of legislative proceedings to deconstruct the power dynamics at play.

• Data Source: The author analyzed six hearings held by the Senate Judiciary Committee's Subcommittee on Privacy, Technology, and the Law during the 118th Congress (2023–2024). The hearings covered topics ranging from "Rules for Artificial Intelligence" to "Election Deepfakes."
• Scope: The dataset includes 15 hours of recorded testimony from 21 witnesses.
• Participant Demographics: The analysis highlights a significant imbalance in representation. Of the 21 witnesses, 10 were from the technology industry (e.g., CEOs of OpenAI, NVIDIA, Anthropic), while others represented academia, media, and government. Civil society and labor representatives were notably underrepresented.
• Analytical Framework:
  • Coding: The author performed a three-stage thematic analysis. First, a descriptive table categorized participants by affiliation. Second, testimony was coded into broad themes ("first-level coding") at the statement level (rather than word level) to identify recurring narratives.
  • Theoretical Lens: The analysis is grounded in Steven Lukes' (2005) concept of the "third dimension of power" (shaping perceived interests) and Langdon Winner's (1980) argument that preconceptions about technology can prevent the examination of its political qualities.
  • Research Question: "What role do shared (mis)understandings of AI play in early attempts at governing this technology?"

3. Key Contributions

The paper makes three primary theoretical and empirical contributions:

1. Deconstruction of the "AI" Term: The author challenges the term "AI" as a marketing construct (citing Emily Bender) that anthropomorphizes tools to dodge accountability. The paper argues that current "AI" is largely a data-extraction engine ("stochastic parrots") rather than a sentient entity, yet legislative discourse treats it as such.
2. Identification of Power Arrangements: The study reveals how industry actors successfully co-opt legislative hearings to establish a "shared reality" that favors deregulation. It demonstrates that governance is not just about writing laws, but about controlling the definition of the problem and the narrative of the solution.
3. Critique of "Innovation Amnesia": The paper connects current AI governance failures to historical patterns (e.g., the mechanical tomato harvester, the early internet), showing how "utopian" visions of technology are repeatedly used to obscure labor harms and environmental costs.

4. Key Results: The Three Dominant Understandings

Through thematic analysis, the author identifies three specific, interconnected narratives that dominate the hearings and shape the legislative outlook:

A. Understanding 1: AI Has Tremendous Potential Benefits

• Narrative: Industry representatives and many Senators frame AI as a universal good capable of solving existential threats (curing cancer, stopping climate change).
• Mechanism: Claims are often speculative and forward-looking (e.g., "one day we will...") rather than based on current evidence.
• Impact: This creates a "benefit bias" where current harms (bias, scams, environmental damage) are dismissed as temporary growing pains. The potential future value (e.g., Stuart Russell's estimate of $14 quadrillion) is used to justify any present risk, sidelining the disproportionate impact on marginalized communities.

B. Understanding 2: AI is Inevitable

• Narrative: The technology is portrayed as an unstoppable force ("the genie is out of the bottle").
• Mechanism: Industry argues that because AI algorithms are open-source and globally distributed, regulation cannot stop its spread. This creates a false dichotomy: either accept the technology as is, or face a "secret" arms race.
• Impact: This narrative renders legislative intervention futile. If AI is a "natural phenomenon," policymakers feel absolved of the responsibility to regulate it, leading to a "wait and see" approach that effectively amounts to deregulation.

C. Understanding 3: AI Nationalism

• Narrative: AI development is framed as a matter of national security and American exceptionalism.
• Mechanism: Industry and Senators contrast "American market-driven innovation" with "Chinese state-driven control" or "European rights-driven regulation."
• Impact: This framing equates deregulation with democracy. Any attempt to regulate AI is cast as an attack on American values, innovation, and small businesses. It forecloses critique by suggesting that slowing down AI development would cede global hegemony to adversarial nations.

5. Significance and Implications

The paper concludes that these shared (mis)understandings are not merely rhetorical flourishes but are active mechanisms of governance.

• Obfuscation of Material Harms: By focusing on speculative future benefits and inevitability, the hearings ignore the concrete, immediate costs of AI: the extraction of personal data (Surveillance Capitalism), the use of precarious labor for content moderation, and the massive environmental footprint of data centers.
• Stagnation of Regulation: The combination of "inevitability" and "nationalism" creates a political environment where meaningful regulation is viewed as impossible or unpatriotic.
• Policy Recommendation: The author argues that effective governance requires dismantling these narratives. Policymakers must reject the "inevitability" fallacy, recognize that AI is a choice made by powerful corporations rather than an autonomous force, and prioritize the protection of vulnerable populations over the preservation of "innovation" at all costs.

In summary, Leach's paper serves as a critical intervention, warning that without a rigorous deconstruction of the dominant techno-optimist narratives, early AI legislation will likely codify industry interests rather than protect the public.