Watch out for these traps when using AI

EVERYTHING from mobile devices, to home appliances, and even in classrooms; artificial intelligence (AI) is powering a huge part of our daily lives.

Thanks to the reducing cost of hardware and services, AI has become more accessible to businesses. Cloud computing and neural networks are more readily available, driving adoption of technologies like facial recognition, computer vision, and natural language processing (NLP).

This is the same for the media industry. Even moderate-sized media companies are migrating to intelligent systems. However, implementing AI isn’t like plugging in a USB drive; there are many factors a broadcaster must consider before embarking on AI projects.

Arnaud Elnecave, Vice President of Marketing, Dalet, shared with Tech Wire Asia in an interview, about the common traps businesses fall into when dealing with AI.

Although AI design and research surfaced as early as 1943, it wasn’t until recent years that businesses have started to deploy it en mass.

“AI is coming fast, but there is a steep learning curve to its efficient adoption,” Elnecave said. “There are a wide variety of AI models that you’ll need to use in concert, and this requires some seamless orchestration, advanced data models and human input for fine tuning.”

Before embarking on any AI projects, Elnecave suggested for business to consider the costs.

The prices of AI vary greatly depending on the system and what a broadcaster wants to achieve. Running an entire archive through an intelligent video indexer, for example, would be expensive and achieve little for a business.

It might be worth reviewing what needs to be done, and what level of indexation would be needed. Working smarter can help users to reduce costs.

Next consideration is data alignment. Elnecave explained, “When you are using multiple AI engines – and you will be very shortly – you must make sure that elements like your named entities, topics, and key phrases are consistent across all your datasets.”

Many companies, over the years, would’ve accumulated a lot of data that follows different structures. In terms of broadcasting, it could be the naming convention of video files, or different categories used to file video archives.

Having a full cleanup of data is important to properly categorize datasets. Companies would have to streamline the data to fit with the taxonomy of a core library.

One of the most important factors is usability. For users to benefit from the AI-generated information, they need intuitive, transparent user tools. Insights, recommendations, notifications, alerts, analytics need to be clearly presented so that users can take appropriate actions in response.

Finally, AI models need to be trained and adapt to the domain of application. The right kind of data must be fed to the algorithm, to improve its accuracy and be relevant for its intended purpose.

Elnecave suggested users of the system to actively perform the correct feedback loops that will fine tune the models. This needs to be constant and rigorous, weaving into normal, daily workflow. Performing fine-tuning outside of production processes can be costly.

“If you don’t address all these issues and potential traps, you just won’t get all the benefits out of your AI-driven projects. And your organization runs the risk of seriously overspending,” he advised.

Beyond obvious uses of AI – such as speech-to-text, image recognition, sentiment analysis; media has the potential to drive even more value out of AI.

One suggested use is implementing recommendation services within production. For instance, an AI could suggest instant sources of content, research, and verification for journalists while they are writing stories.

Elnecave will be sharing more about the traps and potentials of AI in the media industry at the ConnecTechAsia Summit in Singapore later this month.

A useful rule of thumb, according to Elnecave, is “when you’ve identified actual business cases that you know will generate the right value for the costs incurred. Not before.”

Having said that, it is best for companies to identify those cases as soon as possible. The first step is to prepare, as early as possible, a technology foundation and platform that can effectively connect and apply AI to benefit your business.

---

---

---

---