Is data management the secret to generative AI?

Introduction

Patterns and relationships within vast amounts of data unlock entirely new possibilities. Sometimes, we learn about our past or discover something that helps us predict the future. For some time, organizations have been collecting data without fully understanding its potential. Over time, the volume of this data can become overwhelming. This is where the relationship between data and artificial intelligence (AI) gets particularly interesting.

In a recent discussion, Love Uger Wall, the Worldwide Sales Leader for IBM Watson X, and Edward Cisper, the Vice President of Product Management for the IBM Watson X platform, explored how generative AI (Gen AI) affects data management. One key insight is that there is no AI without data; data is the only sustainable source of competitive advantage in business.

Generative AI is changing perceptions of data in two significant ways:

1. Effective Use of Unstructured Data: Gen AI models can more effectively utilize unstructured data, which constitutes the majority of new data. This capability allows Gen AI to analyze large volumes of language data, primarily documents or software code, and identify patterns or make connections without needing extensive preparation or supervision.

2. Self-Management of Data: Gen AI can assist in addressing data management challenges. For example, if a client has various legacy applications with inconsistently formatted data, Gen AI can help make sense of that data, regardless of how it is scattered across systems. This potential leads to significant savings in time and energy, converting data management into a competitive edge for organizations.

Organizations differ in their ability to take advantage of their data. Some struggle with architectural issues, such as data being siloed across on-premises and cloud environments. Others face psychological barriers that prevent them from evolving their business models to turn data into a more integral aspect of their operations. Data monetization is considered a form of nirvana, allowing businesses to sell versions of their data effectively.

To improve data quality, organizations need to adhere to good traditional data quality practices. This includes cataloging or organizing data into a business glossary and developing thoughtful data access policies. Monitoring and enforcement are crucial to ensure effective governance—these policies should be set centrally and enforced locally while actively monitoring model inputs and outputs.

Quality and trusted data are prerequisites for the successful implementation of generative AI in business. Companies often struggle to move past initial prototypes to customizing models with their data and deploying them across the enterprise. There are two main ways organizations can tailor generative AI with their own data:

1. Tuning the Model: This process involves instructing the model using good examples from enterprise data, allowing the model to learn and adapt to the organization's language and structure.

2. Retrieval Augmented Generation (RAG): Unlike tuning, RAG leverages a knowledge base of quality enterprise data to enhance the accuracy of model responses, decreasing the risk of 'hallucinations.'

Training models require carefully curated data. IBM Research utilizes a data lake house architecture to source, catalog, filter, and transform data before using it for training. This comprehensive approach allows enterprises to maintain a complete lineage of data sets, pipelines, and AI models.

A well-stocked data lake house functions like a commercial kitchen, where all necessary ingredients (data) are readily available and organized. This architecture combines the best features of data lakes and data warehouses—offering flexibility, scalability, and excellent performance.

In summary, organizations need to adopt a rigorous data and AI governance framework early on, rather than as an afterthought, to facilitate efficient experimentation and deployment. Open-source technology plays a crucial role in this evolution by delivering superior transparency and security while enabling community-driven innovation.

Enterprises ready to embrace generative AI must ensure they have a solid foundation in data management to ensure that productivity gains translate into competitive advantages.

Keywords

• Data Management
• Generative AI
• Unstructured Data
• Data Governance
• Data Lake House
• Model Tuning
• Retrieval Augmented Generation
• Data Quality
• Data Monetization

FAQ

Q1: What is the relationship between data and generative AI?
A1: There is no AI without data; quality data is essential for the successful implementation of generative AI and serves as a sustainable source of competitive advantage.

Q2: How can organizations improve data management for generative AI?
A2: Organizations can enhance data management by adhering to good data quality practices, developing thoughtful access policies, and monitoring data usage.

Q3: What are two main ways to customize generative AI with enterprise data?
A3: The two primary methods are tuning the model using enterprise data and using retrieval augmented generation (RAG) to improve model accuracy.

Q4: Why is a data lake house considered beneficial for enterprises?
A4: A data lake house combines the scalability of data lakes with the performance of data warehouses, providing a well-organized, readily accessible pool of data.

Q5: How important is governance in AI and data management?
A5: Governance is critical in managing data risks and compliance; organizations must implement a governance framework early on to effectively manage AI and data.