Why Generative AI is No Good Without Traditional Machine Learning

AI is nothing new, it’s been around since the 1950s and has traditionally been utilised for user behavior modeling through a wide range of algorithms, such as neural networks and collaborative filters. In comparison, Gen-AI, in its current form, does not classify behavior, users, or content but focuses on the generation of human-like text, speech and art, and is therefore struggling to find real use in today’s classic AI applications. Gen-AI is not as intelligent as AI, it’s mainly a tool that helps create content that is understood in a way humans speak and write, so therefore easier to interpret for people. And that is where it can add the most value to important personalisation use cases. It can help generate responses that feel most natural to TV users, offering conversation-like interactions and explanations to make the service more ‘second nature’ for the user. The beauty of AI, in general, is the ability to understand a lot of data in a short period of time, Gen-AI translates that into an interesting outcome through the contextual knowledge it builds from the data. Users are more likely to trust and stay loyal to platforms, as the reasoning behind personalisation becomes understandable because the language used is similar to their own.

However, Generative AI comes at a cost - quite literally - and around scalability and the real-time aspect. There is a risk of overwhelming the user with too much, very rich, but not necessarily very accurate media. The generative technology is limited by its response time/processing power ratio and its reliability, factors that are crucial when it comes to personalisation within TV services in particular. There is a risk of striking the wrong balance between AI and Gen-AI and delivering everything from Gen-AI, as it seems to be the more universal tool, but sacrificing financial viability and reliability along the way. This is a risk because, without properly controlled input, Gen-AI can generate as much wrong content as it does useful. To realise the full potential, Generative AI has to either be applied in a controlled environment and only for use cases, where it can add premium value today, or there needs to be human intervention to ensure that the information is correct and accurate.

The automation of processes that enhance customer journeys and experiences is a strong area for Gen-AI. For example, Gen-AI can be utilised to deliver impressive user interfaces with engaging and resonating row titles to fit the preferences of individual viewers. The creation of such dynamic UIs that suit the preferences of audiences requires rich metadata, which describes each piece of content in detail. AI-powered algorithms identify trends within user data, including viewing history, preferences, social interactions and more. To be able to make the connections between different content items, metadata needs to be accurate and enhanced to deliver unique recommendations with natural language explanations or row titles.

AI has been applied within monetisation strategies for years, with the ability to analyse vast amounts of data, it’s a great fit to help match users with the right ads for the best ROI. Does the industry think that Generative AI can add value here? Honestly, as the technology is today, no, but it is increasingly becoming applied in adjacent areas, like swapping of branded products in video footage as part of sponsoring and product placement.

As mentioned, Gen-AI is not affordable for generic use cases, the running costs can be extraordinarily high and despite being more complex than traditional AI, it’s not as intelligent when it comes to capturing user intent or preferences. Also, Generative AI models may produce biased or inaccurate content based on the data they are trained on. There are some risks; however, they can be mitigated by OTT platforms adopting a comprehensive approach that combines AI-driven personalisation with human curation, regular auditing of AI recommendations, robust content moderation, continuous bias monitoring and mitigation, and transparent communication with users about AI usage.

There is no doubt that Gen-AI is opening new ways for broadcasters, service providers and operators to bring their services closer to the consumer. With better ways for them to understand what is happening in the service and how it is relevant to the viewers through the extra engagement delivered by the genAI-driven dialog. On the other hand, Gen-AI is still not as ‘intelligent’ as required to be used in the most effective way, as quality issues, costs and scalability are standing in its way. Alone, Generative AI is not strong enough to deliver the outcome that is needed, it’s a tool that allows a machine to hallucinate in a way that is nearly indistinguishable from the way humans speak and write, helpful in improving customer experiences, but should be used with caution and not be mistaken for a panacea.