Bridging Technical AI, Societal Impacts, and Workforce Competencies in AI Education

Technical Summary: Bridging Technical AI, Societal Impacts, and Workforce Competencies in AI Education

Problem Statement
As artificial intelligence (AI) becomes embedded in critical sectors such as hiring, healthcare, and policing, the consequences of system failures extend beyond technical errors to include significant societal harms, including discrimination and structural inequality. Despite the proliferation of AI ethics modules and interdisciplinary coursework, AI education remains structurally fragmented. Technical concepts (models, optimization) are frequently taught in isolation from ethics, policy, human-centered design, and professional skill development. This separation creates a sociotechnical gap where technical abstraction obscures how design decisions translate into social consequences, and where fairness is often attempted through technical metrics alone without contextual understanding. There is a lack of empirical work examining whether technical instruction, societal harms, and workforce competencies are systematically aligned within AI-related curricula.

Methodology
This work-in-progress employs a qualitative curriculum mapping approach to analyze the alignment of three dimensions: (1) technical systems, (2) societal harms, and (3) workforce competencies. The study utilizes a three-part alignment model to operationalize these dimensions.

• Dataset Construction: The analysis began with three coordinated courses within an institutional AI degree program (AI Literacy, Human-Centered AI, and Social Technology Design). To mitigate single-institution bias, the dataset was expanded using public technology ethics syllabus registries (e.g., Casey Fiesler's collection) and screened using AI-related keywords (e.g., machine learning, LLM, algorithmic bias).
• Selection Criteria: Courses were included if they centered on AI systems, engaged with societal impacts (bias, privacy, governance), and provided sufficient detail (syllabi, learning objectives, assignments) for structured analysis. The final dataset for this initial report consists of six courses mapped in depth, drawn from a larger pool of 16 confirmed AI-relevant courses.
• Analysis: Course materials were iteratively coded to identify patterns of co-occurrence and gaps. The analysis focused on whether societal harms were linked to specific technical mechanisms and whether workforce competencies (e.g., communication, stakeholder reasoning, responsible innovation) were explicitly named and assessed.

Key Contributions
The paper makes four primary contributions:

1. A Cross-Course Curriculum Mapping Approach: An initial framework for linking technical systems, societal harms, and workforce competencies across diverse course types.
2. An Emerging Alignment Model: A case-based model for mapping curriculum structures.
3. Preliminary Empirical Findings: Evidence regarding the current state of AI education, specifically highlighting strong technical coverage, uneven integration of societal harms, and limited explicit assessment of workforce competencies.
4. Design Implications: Early recommendations for AI education design to better connect technical rigor with social critique and professional skills.

Results
Analysis of the six in-depth mapped courses reveals four recurring patterns:

• Variable Technical Depth: Technical coverage is generally strong but varies in depth. Some courses (e.g., Fairness in Machine Learning) connect specific technical components (embeddings, supervised learning) directly to harms. Others (e.g., AI Literacy) discuss AI systems at a broader level (platform governance, content moderation). The critical finding is not the depth itself, but whether the level of technical detail is connected to societal harms and professional competencies.
• Uneven Integration of Societal Harms: While all courses addressed harms such as bias, privacy, and misinformation, the integration was inconsistent. Some courses linked harms directly to design, data, or model choices. Others focused on platform-level critiques without tracing harms back to specific algorithmic mechanisms. The study emphasizes that structural connection requires clarifying how technical, organizational, and social decisions interact to produce harm.
• Implicit Workforce Competencies: Competencies such as communication, critical evaluation, and interdisciplinary collaboration appeared across courses but were rarely explicitly assessed. They were often implied through projects or discussions rather than being named as learning outcomes. A gap exists between the presence of these skills and their explicit evaluation.
• Misalignment of Rigor and Critique: A recurring mismatch was observed between technical rigor and social critique. Courses with strong algorithmic depth often provided clear explanations of harms but neglected communication and governance contexts. Conversely, courses with strong societal critique often discussed harms in depth without fully explaining the underlying technical mechanisms. The strongest alignment was found in case-based or project-based courses that integrated systems, harms, and competencies within a single assignment structure.

Significance and Claims
The paper argues that AI literacy efforts must evolve beyond mere awareness of AI concepts or harms. True literacy requires competencies for responsible action in AI-enabled contexts. The study claims that simply adding ethics content to technical curricula is insufficient; technical abstraction often obscures the production of harm, and fairness cannot be addressed through metrics alone.

The significance of this work lies in its demonstration that AI education requires deliberate structural alignment between technical systems, societal impacts, and decision-making skills. The authors posit that separating ethics from technical instruction limits students' ability to connect design decisions to real-world outcomes. By highlighting the gap between what students learn and what they are expected to do in professional settings, the paper advocates for embedding competencies directly into technical assignments. Furthermore, it suggests that case-based, cross-course approaches using shared projects can make the alignment of systems, harms, and skills more visible and actionable, moving AI education toward a more integrated, human-centered perspective.

Note: As a work-in-progress, the authors acknowledge that the current findings are based on an evolving dataset and do not constitute definitive claims about the entire landscape of AI education. Future steps include expanding the dataset, refining the coding scheme, and developing clearer measures of alignment across institutions.