Rich Data Versus Quantity of Data in Code Generation AI: A Paradigm Shift for Healthcare

Area of Science:

• Artificial Intelligence
• Software Engineering
• Health Informatics

Background:

• Code Generation AI (Code Gen AI) relies on vast datasets for training.
• General-purpose code corpora may lack domain-specific context and quality crucial for high-integrity applications.
• The healthcare sector demands rigorous standards for software safety, auditability, and compliance.

Purpose of the Study:

• To evaluate the trade-offs between "rich data" and "data quantity" strategies in Code Gen AI.
• To focus on the impact of data quality in high-integrity sectors, specifically healthcare.
• To introduce a data selection matrix for Code Gen AI systems.

Main Methods:

• Comparative analysis of "rich data" versus "data quantity" approaches in Code Gen AI.
• Case study using Self-Evolving Software.
• Development of a weighted data selection matrix for Code Gen AI.

Main Results:

• Models trained on unfiltered, large-scale datasets can increase code duplication, churn, and error rates.
• Code Gen AI can enhance productivity by up to 55% in controlled environments.
• Rich, curated, domain-specific datasets yield more robust, compliant, and sustainable code.

Conclusions:

• Data quality and curation are paramount for Code Gen AI in healthcare.
• Balancing performance gains with reliability, maintainability, and ethical accountability is essential.
• Domain-specific, rich datasets are superior for developing high-integrity software systems.