By Debra Kaufman
HPA Creative Tech UK convenes in London on June 27 with a program focused on navigating the frontier of media and technology. Finishing the day’s engaging sessions will be a panel on artificial intelligence, with Arvato project manager Yvonne Thomas and FeedForward AI co-founder/director of business strategy Lydia Gregory. Prior to co-founding FeedForward AI, Gregory was head of growth for an AI music startup and Thomas, an engineer, worked at the EBU where she was involved in 3DTV, UHDTV and other standardizations. We sat down with Thomas and Gregory to get a preview of their perspectives on AI.
Thomas notes thatAI is not a new topic. “But it’s the moment when AI services become more mature and reliable,” she says. “Together with open APIs and advanced software, it’s actually easy to integrate and use those AI services.” Gregory adds that AI is getting proficient at such tasks as object recognition. Because AI solutions are mostly hosted in the cloud, they allow for easy scability and accessibility, she says. But, because AI is new, organizations using AI solutions need specialists, who are hard to find and expensive, to operate their systems. “Since the AI results are not yet 100 percent accurate, training the system is an on-going job,” Gregory adds. “AI must be considered a part of the system landscape. If the system can’t handle or visualize AI data, AI services are of very low value.”
Thomas points out that the media and entertainment industry can benefit from several AI use cases, including using video-centric analytics to enrich metadata to find content more precisely and faster, which helps to monetize it. “Second, AI can be used for predictions, meaning that a server can be repaired before it even breaks down, because AI can predict a failure down time,” she says. “This not only reduces costs but also increases the reliability of a system, especially in a 24/7 operation. Third, AI can increase the level of security by identifying a hack attack early and preventing its continuation.”
An example of the first use case is Arvato’s MAM system, says Thomas, who is its product manager. “Metadata enrichment is essential for a MAM, so that its users can work even more efficiently,” she points out. “The data-based services we currently use focus on video analytics, including face recognition, speech-to-text, lipsync extraction, scene detection and so on.”
Likewise, FeedForward AI has just released Figaro.ai, a solution for intuitive audio search. “To find the right track for a video soundtrack, advert or personal listening, we usually have to search through a large database of tracks, and, to do this, we have to use words,” says Gregory. “There is therefore a semantic gap between the words we use to describe the track and the audio itself. We enable a user to find the right track by using deep learning to create a link between the raw audio and its metadata.” Gregory adds that FeedForward AI is working with audio platforms to implement Figaro into their search capabilities. The company also has plans to extend the service to audio with speech, images and video.
Many people in the media and entertainment industry want to get up to speed on artificial intelligence, and both Gregory and Thomas have suggestions on how they might do so. “AI is part of almost every conference and the big service providers offer interesting blog posts of their progress,” suggests Thomas, who also recommends that those who are curious ought to “experiment with AI and be amazed.” Gregory suggests, for those who want to delve into the technology, to check out Andrew Ng’s Coursera course on machine learning or follow AI market research on TechEmergence.
What does the future hold for AI in our industry? “Overall, I am optimistic,” says Gregory. “New technology has always caused disruption but so far, AI is being used in the media & entertainment industries to augment the creative process. In a world where more content is required more quickly, the ability of machine learning to power solutions that decrease time to market is important.” She points out that, “many of the current machine-learning-driven solutions focus on the content finishing part of the lifecycle,” such as localization, compliance, improving discovering as well as restoration.
Ethics should always be a part of the conversation when talking about AI, says Thomas. “Overall, people fear AI although they seem to find it fascinating,” she says. Those who look closer recognize its benefits. Even though this means that many jobs will be automated, it doesn’t mean an overall job loss. The machines need to be trained and maintained, therefore only the job tasks will change.” She also cautioned that, because AI results aren’t completely accurate, users must take this into account when examining the resulting data. As to whether machines will take over the industry, she notes that, “the limits are being set by humans.” “If we like them to become more intelligent and self-learning then this is our decision,” she says.