Enterprise AI Analysis
Provocations from the Humanities for Generative AI Research
LAUREN KLEIN, Emory University, USA
MEREDITH MARTIN, Princeton University, USA
ANDRE BROCK, Georgia Institute of Technology, USA
MARIA ANTONIAK, University of Copenhagen, Denmark
MELANIE WALSH, University of Washington, USA
JESSICA MARIE JOHNSON, Johns Hopkins University, USA
LAUREN TILTON, University of Richmond, VA, USA
DAVID MIMNO, Cornell University, USA
Executive Summary: Humanity-Centered AI Development
This paper presents a set of provocations for considering the uses, impact, and harms of generative AI from the perspective of humanities researchers. We provide a working definition of humanities research, summarize some of its most salient theories and methods, and apply these theories and methods to the current landscape of AI. Drawing from foundational work in critical data studies, along with relevant humanities scholarship, we elaborate eight claims with broad applicability to current conversations about generative AI.
0
Key Provocations
0
Years of Interdisciplinary Research
0
Deeper Cultural Insight
0
Improved Research Potential
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
Meaning-Making
Culture Definition
Representativeness
Model Size
Training Data
Openness
Corporate Capture
Human Subjects
1. Models make words, but people make meaning
Generative AI systems excel at predicting word sequences, leading to linguistically fluent output. However, this output is devoid of inherent human intention or understanding, a critical distinction from how humans create and interpret meaning.
| AI Text Generation | Human Meaning-Making |
|---|---|
|
|
2. Generative AI requires an expanded definition of culture
Current AI approaches often rely on a narrow definition of "culture" based on European modernity (geographic region, nationality, language). Humanities research offers a broader understanding of culture, encompassing both "the way of life of a people" and "the works and practices of intellectual and artistic activity."
Understanding that training data both reflects cultures and consists of expressions of those cultures is crucial for intentional data curation. This expanded view allows us to analyze how power structures influence data capture and how AI models themselves become cultural objects reflecting contemporary tech culture.
3. Generative AI can never be representative
Inherent Bias
AI Systems Cannot Be Truly Representative
The issue of bias in AI systems reflects a deeper structural problem rooted in unequal power differentials, not merely "bad data." True representativeness is an unachievable goal when dealing with historical sources and diverse human experiences. Humanities approaches like critical fabulation and "null value" acknowledge and engage with these absences and silences, rather than attempting to model them away.
4. Bigger models are not always better models
The assumption that increasing parameters and data leads to better AI performance has diminishing returns. This "bigger is better" agenda is often driven by corporate interests and leads to unsustainable environmental and economic costs.
| Bigger, Universal Models | Smaller, Bespoke Models |
|---|---|
|
|
5. Not all training data is equivalent
Pretraining data is often treated as a homogeneous, undifferentiated resource, primarily valued for downstream utility. This approach overlooks the complex cultural, historical, and social contexts embedded within the data.
From Undifferentiated to Curated Data
Training Data as "Homogeneous Resource"
→
Evaluation by Generic "Toxicity" / "Quality"
→
Leads to "Documentation Debt" & Shallow Analysis
→
Humanities Perspective: Data as "Cultural Expression"
→
Requires Rich Metadata & Intentional Curation
→
Enables Transparency, Accountability & Innovation
6. Openness is not an easy fix
While open-source models are gaining traction, true "openness" remains a complex and unresolved issue. "Open" often applies only to model access, not to underlying data or training processes, and still requires significant computational resources, creating economic barriers.
Key concerns include copyrighted content (balancing creators' rights with research needs), community consent for data scraped from ostensibly "open" online spaces, and the ethical implications of synthetic data re-staging historical harms. Genuine openness must be continually reevaluated in changing contexts.
7. Limited access to compute enables corporate capture
The High Cost of AI: Fueling Corporate Capture
Current AI models' dependence on massive scale and exponentially growing costs for minimal performance gains presents a nightmare for corporate interests. This inverts the historical trend of computing efficiency and fuels a race for consolidated resources.
This dynamic aligns with the mechanisms of late capitalism, where corporations actively work to separate data from the people who produce it, transforming it into "surplus value" that can be monetized. This exploitation of resources and labor, paralleling historical colonial patterns, concentrates power in the hands of a few tech CEOs, driving further extraction and exploitation.
Key Takeaway: AI's economic mechanisms align with late capitalism, demanding strategies of resistance and alternative futures (e.g., Indigenous data futures, viral justice) to challenge corporate control.
8. AI universalism creates narrow human subjects
Generative AI operates within a "data episteme," where human identity and knowledge are validated solely through data acquisition. Culture is reduced to "content," transforming complex human experiences into abstract records suitable for extraction.
This reductionism, stemming from European modernity's premises of quantifying progress, actively generates narrow, often racialized, understandings of humanity. AI's claims of "universal knowledge" are thus a statistical enactment of existing ideologies, not a true representation of human diversity, making AI deeply anti-human in its current form.
Quantify Your Potential AI Impact
Use our interactive calculator to estimate the efficiency gains and cost savings your enterprise could realize by implementing a thoughtful, humanities-informed AI strategy.
Estimated Annual Savings
$0
Total Hours Reclaimed Annually
0
Your Humanity-Centered AI Roadmap
Our approach integrates humanistic understanding into every phase of AI development and deployment, ensuring ethical, effective, and equitable solutions.
Phase 1: Humanistic Assessment & Discovery
Conduct a deep dive into organizational culture, ethical frameworks, and societal impact. Identify potential biases in existing data and processes through a humanistic lens.
Phase 2: Contextual Data Curation
Develop tailored data strategies, prioritizing nuanced, diverse, and ethically sourced datasets. Implement rich metadata practices informed by humanities research to ensure accurate representation.
Phase 3: Bespoke Model Development & Tuning
Focus on smaller, purpose-built AI models that align with specific enterprise needs. Engage domain experts and humanities scholars in model design and fine-tuning to avoid universalist pitfalls.
Phase 4: Ethical Deployment & Monitoring
Establish robust governance for AI deployment, with continuous monitoring for unintended harms and biases. Integrate human oversight and feedback loops to adapt and refine models post-launch.
Phase 5: Future-Proofing & Cultural Integration
Foster an organizational culture that values interdisciplinary collaboration and ongoing ethical reflection in AI. Adapt strategies to evolving social, political, and technological landscapes for sustainable AI innovation.
Ready to Transform Your AI Approach?
Schedule a personalized consultation with our experts to discuss how integrating humanistic insights can drive innovation, reduce risk, and build more equitable AI systems for your enterprise.
Ready to Get Started?
Book Your Free Consultation.
Let's Discuss Your AI Strategy!
Lets Discuss Your Needs
Select a Date
Sun
Mon
Tue
Wed
Thu
Fri
Sat