AI is becoming increasingly pervasive in daily life, powering applications from playful voice assistants to medical diagnostic tools. The technology press has a keen interest in the latest models and algorithms. But what if the key to unlocking AI’s potential lies in data, not algorithms?
In industries that have already adopted AI, a quiet shift in focus is underway with the rise of data-centric AI. This approach avoids the assumption that bigger models and better algorithms are the main drivers of AI progress.
Instead, data-centric AI emphasizes that the quality and relevance of the data used to train a model is as important, if not more important, than the model architecture. This focus on data has a profound impact on how models are developed, deployed, and managed with the aim of making AI resilient, customizable, and widely applicable.
The role of algorithms and data in AI
Over the past decade, the dominant paradigm in AI has been algorithm-centric, with the assumption that the development of more sophisticated algorithms and models is the key to improving AI systems. This strategy has led to impressive advances, from language generation in GPT-4 and image generation in Midjourney to protein structure prediction in AlphaFold, but its focus on model complexity has overshadowed the important role of data. I dropped it.
In contrast, data-centric AI recognizes that the performance and reliability of an AI system fundamentally depends on the quality and relevance of the data on which the system is trained. AI pioneer and computer scientist Andrew Ng describes data-centric AI as “the discipline of systematically engineering the data needed to build successful AI systems.”
This means carefully curating, labeling, and structuring your training data to ensure it is consistent, unbiased, and representative of real-world use cases. Masu. Data-centric AI initiatives shift priorities and resource allocation in model development. Data-centric AI invests heavily in data engineering, annotation, and management, rather than focusing all resources on algorithm development.
A key implication of this methodology is that it requires a multidisciplinary team of domain experts, data scientists, and AI engineers who collaborate to create high-quality, application-specific datasets. You may also need new tools and platforms for data versioning, validation, and monitoring.
Nevertheless, data-centric approaches are critical for sectors such as healthcare and manufacturing, where data is often limited, cumbersome, and expensive to collect. In these situations, a small amount of carefully curated data can yield better results than a large, noisy data set. By prioritizing quality over quantity of data, data-centric AI promises to make AI more practical and effective for a wider range of industries and applications.
Practical application of data-centered AI
As Ng's definition suggests, data-centric AI requires a disciplined and systematic data management and engineering approach. This broad field includes several important practices such as data curation, labeling, versioning, and validation.
data curation
Curation is the process of identifying, selecting, cleaning, and organizing data to ensure that it is relevant, accurate, and usable for specific business purposes. This involves defining clear criteria for data quality, relevance, and representativeness, and systematically filtering, enriching, and structuring data from various sources to create high-quality datasets. Masu. Effective data curation involves close collaboration between domain experts, data engineers, and business stakeholders to ensure that curated data is aligned with business goals and use cases. is needed.
Labeling data
Labels provide meaningful descriptions of raw data, teaching machine learning models how to make accurate predictions and decisions. For example, label images with object categories, text with sentiment tags, and customer feedback with issue types. Correct and consistent labeling of data is critical to training effective AI models, but it can be time-consuming and expensive, and often requires human domain experts. By defining clear guidelines, using techniques like consensus labeling, and investing in tools that streamline the process, you can significantly improve label quality and efficiency.
Data versioning
Data versioning, similar to code versioning, involves tracking and managing changes to datasets over time. This involves systematically cataloging the different versions of a dataset and metadata about what changed, who made the change, and why it changed. Data versioning allows teams to collaborate effectively on evolving data sets and track model performance across different data versions, including the ability to roll back to previous versions if needed. This approach provides the transparency and reproducibility essential for building trust in AI systems and debugging data drift and quality issues.
data validation
Continuously assessing and monitoring the quality, consistency, and relevance of data helps train and evaluate AI models. This includes defining and measuring key data quality metrics such as accuracy, completeness, timeliness, and representativeness, as well as setting acceptable thresholds for each metric. Regular data validation helps identify and mitigate data drift, bias, or errors that can degrade model performance over time.
By curating high-quality and diverse data, models can learn more predictive and generalizable features. Consistent labeling reduces noise and ambiguity in the training signal. Versioning allows teams to iteratively audit and adjust datasets. Finally, validation helps you catch problems in your data before they degrade model performance.
For example, a data-centric approach in manufacturing involves carefully collecting and labeling images of product defects, consistently annotating defect types, and ensuring that datasets cover all relevant product versions and variations. This may include verifying that. In healthcare, multidisciplinary teams manage patient records with consistent codes, ensure demographic and clinical diversity, and regularly update datasets as new data becomes available. may be involved.
Beyond the model
The shift to data-centric AI has implications beyond model performance. Its purpose is to address today's most pressing AI challenges: trust, transparency, and accountability.
Resilient and reliable AI systems are essential for high-stakes applications such as healthcare and transportation, where safety is paramount. Data-centric practices such as comprehensive data validation and continuous monitoring can help keep AI systems safe and effective even as real-world conditions change.
Data-centric AI promotes transparency and accountability, and thus trust, by making data provenance and processing more visible and auditable. Additionally, data-centric AI intersects with responsible AI practices and AI governance frameworks. Documented data collection, labeling, and validation processes all help demonstrate compliance with privacy regulations and AI ethics frameworks. Involving diverse stakeholders in data curation can surface and reduce bias.
However, the shift to data-centric AI also raises challenges and questions. Collecting and annotating high-quality data can be resource-intensive. There are legitimate concerns about scalability and accessibility. Balancing data needs and privacy concerns is an ongoing challenge. Aligning data practices across organizations and domains requires alignment and standardization. But building more robust, responsible, and effective AI systems requires addressing these challenges.
Donald Farmer is president of TreeHive Strategy, which advises software vendors, enterprises, and investors on data and advanced analytics strategies. He has worked on several leading data technologies in the marketplace and at award-winning startups. Previously he led the Design and Innovation team at Microsoft and Qlik.
