Since the dawn of technology, persistent change has been the norm.
The CEO of a Silicon Valley IT company once said that if you are uncomfortable with change, you are in the wrong business. Technology has improved productivity, resulting in greater efficiency. It has also disrupted existing processes, giving businesses and consumers new alternatives for creating and receiving value.
The latest technology to upend and reframe business methodologies is artificial intelligence (AI).
It’s not new. Consumers have been exposed to AI concepts for years through autocorrect, mapping services, and facial recognition. Businesses have adopted low-level AI in their use of automation.
So, why is AI suddenly attracting the attention of industry leaders?
Well, the underlying technology has gotten smarter and faster. Chips, or AI accelerators, are specifically designed to execute AI workloads efficiently. They can significantly improve the performance of AI algorithms compared to general-purpose CPUs or GPUs.
This performance improvement allows faster training and inference times, enabling real-time or near-real-time AI applications.
With the proliferation of data from various sources, including sensors, IoT devices, social media, and digital platforms, advanced AI algorithms are necessary to process, analyze, and derive insights from large datasets efficiently.
AI is now more accessible to anyone at home or in the office through various generative AI platforms.
So, with this heightened awareness, it’s time to clarify the importance of the relationship between data and AI.
Data drivers
The rise of data-driven decision-making not only influences the adoption of AI but also reinforces the importance of data itself. The DPP 2024 Predictions highlighted the importance of media organizations defining their data strategies. Throughout the media supply chain, data is critical to the effectiveness of various integrations and workflows. At the same time, high volumes of data are created at every stage, from creation to consumption.
The demand for cost reduction, streamlined workflows, and operational efficiency are the primary drivers of AI adoption across the media and entertainment industry. According to Grandview Research, the market value of AI in this sector is projected to reach $124.48 billion by 2028, a compound annual growth rate (CAGR) of 31.89 percent.
The application of AI will not only create efficiencies but also more data. The media and entertainment industry must ensure that the underlying data used to train AI engines is complete, validated, and unbiased. If the quality of the underlying data is in question, then the output is also problematic.
The importance of clean data
With the rising use of generative AI (GenAI), media organizations must ensure that the data underlying GenAI platforms integrated with media workflows is accurate, comprehensive, and complete. GenAI incorporates algorithms and data models to create new data. If errors exist in the data used in these models, the outcome will magnify those flaws.
On the other hand, clean data ensures that the generative model learns meaningful patterns and produces accurate and relevant outputs.
Generative AI has many applications, including image generation, text generation, music generation, and more.
In each application, clean and high-quality data is essential for training accurate and realistic generative models. For example, in image generation, if the input images are blurry or contain artefacts, the generated images may also suffer from the same issues.
Preparing data for use in a GenAI model requires ensuring the data is in a consistent format and data schema. Understanding the source, provenance, and copyright of the original data and any updates to that data is an element that is often overlooked.
As GenAI advances, tracking data lineage will increase in importance. Upon validation of data formats, schemas, accuracy, completeness, and consistency, the data is ready for normalization. This process ensures that data is suitable for use in a GenAI, machine learning, or more advanced AI training environment.
AI enhancing metadata management
When it comes to monetizing content catalogues, AI algorithms will analyze user behavior, consumption patterns, and preferences, enabling them to deliver personalized content recommendations. By analyzing the descriptive and collaborative data associated with content, content recommendation engines can strengthen personalization beyond genre or cast members. Using a broader data set when training AI algorithms, will provide greater relevance, thereby improving user engagement and satisfaction.
Another area ripe for AI is that of image generation. Images play a significant role in enticing consumers to watch a series or movie.
Audiences with different backgrounds or cultural beliefs are attracted to different types of imagery. Generating a wider range of images is appealing to many video service providers. However, using AI-generated images raises legal and ethical concerns, particularly regarding copyright and ownership. Video service providers must navigate these issues carefully to ensure they have the rights to use and distribute AI-generated content legally.
There are several ways in which a pragmatic approach to AI can add value to metadata management. Starting with quality control, AI models can identify inconsistencies or errors in metadata. In learning patterns such as date or text formats, AI can detect anomalies and flag data fields that are likely to be incorrect. Then, GenAI can make recommendations to improve data quality. AI can also identify opportunities for metadata enrichment by assessing existing descriptive metadata and suggesting relevant keywords, tags or captions.
As AI is trained to understand the content and context of data fields and content records, many use cases may be defined where it can standardize and accelerate outcomes. For example, as video service providers aggregate data from different sources, machine learning can streamline the matching and linking of metadata across different datasets or sources.
Another critical use case for broadcasters and streaming services is that of genre classification. Clustering algorithms help identify common characteristics within each genre and differentiate between different genres, even in cases where traditional genre definitions are ambiguous or overlapping. It is in these types of narrow use cases that AI can provide measurable value. However, this is only possible with a foundation of clean data.
MetaBroadcast has been consolidating and cleansing data for over ten years. It is a critical prerequisite to the automated equivalence process enabled by Atlas, our metadata management platform. Our metadata repository of over 140 million master MBIDs and their associated content records reflects millions of data fields ingested, cleansed, and equipped from many sources (e.g., ITV, Gracenote, IMDb, Press Association, Wikipedia, broadcaster CMS, etc.).
The records have been persistently updated as existing data changed or new data became available, giving MetaBroadcast enhanced capabilities to match content records successfully and deliver clean, consolidated metadata to our customers.
There is no argument that the data to train all AI models is present and available through the media supply chain. Yet, as all forms of AI are implemented across the media and entertainment industry, we must acknowledge that clean, validated, and verified data is imperative.
===================
** By Peggy Dau, Marketing Director, MetaBroadcast **