Useful AI Requires Tons of Data

To operate successfully in a country, content providers and streaming platforms must comply with local regulations and perceive cultural sensitivities. This entails everything from editing prohibited content and assigning correct age ratings to accurately portraying religions and sub-cultures. With nearly 200 countries worldwide, it's almost impossible for any content creator to know what is or is not prohibited in certain countries. Cultural competence is crucial, and that's where Spherex is unmatched. We know how to handle all these issues and get them right the first time to reduce cost, mitigate risk, and accelerate time-to-market.

An Ounce of Prevention is Worth a Pound of Cure

Before releasing a title in a market, it is better to be aware of regulatory and censorship red flags in the content. Doing so allows the creative teams to proactively decide how to handle concerns on their terms and make edits when production schedules and costs are the most manageable and economical. With the number of titles released annually growing exponentially, it's impossible for humans alone to accurately and consistently prepare each title for global distribution. State-of-the-art machine learning (ML) and artificial intelligence (AI) systems now significantly augment human capacity to analyze and process millions of hours of video content for localization and regulatory compliance worldwide. Spherex is at the forefront of using AI/ML to provide age rating and cultural and regulatory insights gleaned from the analysis of millions of titles to identify the explicit scenes that may be problematic across global markets.

The traditional way of addressing these concerns is in post-production localization. Script and action translation have been part of the post-production process for decades. Problems arise when reliance on language translation misses cultural references, thus creating opportunities for unacceptable content to be overlooked and released to audiences. Violence, sexuality, drug use, and other events within a title can be perceived differently — even in neighboring countries. Therefore, knowing those differences are critical during localization.

Machine Interpretation of Content

Human or machine "intelligence" is obtained through "learning." For humans, learning starts when we're born and proceeds throughout our lives. We see, hear, feel and observe and, using our brains, put the input together to form words, thoughts, actions, and feelings. Machines, on the other hand, cannot. At their most fundamental level, what they know is narrowed down to zeros and ones, off and on, yes or no. Anything beyond that requires the development of programs and rules that govern what they can or cannot "do" based almost entirely on "true" or "false." The more we want them to know or do, the more complicated it becomes.

We've progressed significantly since the first tic-tac-toe computer game in 1952. Atlas and Spot, the famous AI dancing robots, required years of research, programming, development, and trial and error to enable them to walk, jump and balance on one foot without falling. At every development phase, they were taught to recognize surroundings and navigate objects to perform even the most mundane movements. Machines that analyze video and audio content must be trained in much the same way to "see" and "hear" objects and events. Simple tasks humans take for granted require machines to learn at the most fundamental levels.

What Spherexgreenlight™ had to Learn

Consider Spherexgreenlight™ and other Spherex AI technologies. Not only did the tools have to learn how to examine video and listen to audio, but they also had to be able to identify people, places, and things appearing in the video and combine findings to analyze and interpret the scene. For example, is a knife used for peaceful or harmful purposes? How does music impact or influence scene interpretation? What emotions are visible? What are the cues to determine the mood of a scene? How do animated and live scenes differ? When is drug use good versus bad? Are all curse words equal? It quickly becomes complex.

Training the Spherex AI/ML platform took years of development. It required terabytes of descriptive data covering every aspect of digitized video content to build the core intelligence of the system. We mined thousands of policy manuals, historical literature, local film/TV classifications, current affairs, judiciary decisions on sensitive topics (e.g., LGBTQ, sexual violence, self-harm, blasphemy and religious practices, drug use, and more), and consumer grievances in 100+ countries, affording a deep, extensive library of data to facilitate curation accurately. We developed a comprehensive graph database, an enterprise system for screening and annotating content, and an ML-based rules engine to produce precise and consistent age ratings for every country and territory worldwide. Our systems detect and analyze approximately 1,000 attributes in video scenes that link to rules for one or more regions. Our culture graph embodies 8.3 million potential feature combinations.

Our dedication to the industry and regulators is found across the entire Spherex ratings platform. As with all AI products and services, Spherex AI systems can perform tasks because they are designed and trained well. System training doesn't occur once and then end; it requires the constant addition of new data and improvement in the video and audio analysis components to ensure the platform is as thorough and accurate as possible.

Contact us today to see what Spherexratings™ and Spherexgreenlight™ can do for your content.