Unitary AI, $15M for Video Content Moderation

News
October 3, 2023

Unitary AI Secures $15 Million Funding for Innovative Multimodal Video Content Moderation Approach

Unitary AI, a startup based in Cambridge, England, has secured $15 million in funding to support its innovative multimodal approach to video content moderation. The funding round, led by top European VC Creandum, will allow Unitary AI to expand into new regions and hire additional talent. With the number of videos being classified by Unitary AI increasing to 6 million per day, the company’s platform is gaining momentum in the market. By analyzing text, sound, and visuals simultaneously, Unitary AI aims to provide a more effective solution to the challenges of video content moderation, which have been a priority for online platforms in recent years.

Content moderation remains a hotly debated issue in the realm of online media. With new regulations and growing public concerns, it’s expected to remain a top priority for years to come. However, the emergence of weaponized AI and other technological advancements is making the task increasingly challenging. Enter Unitary AI, a startup from Cambridge, England, which introduces a more effective solution: a ‘multimodal’ approach to parse content moderation, especially in the intricate medium of video.

Today, Unitary announces a $15 million funding round to seize the opportunities created by its recent market momentum. This Series A round, spearheaded by leading European VC firm Creandum and featuring contributions from Paladin Capital Group and Plural, coincides with Unitary’s robust business expansion.

In the past year alone, the platform’s video classification volume has surged from 2 million to an impressive 6 million per day, encompassing billions of images. Furthermore, Unitary is broadening its language support beyond English. Although specific customer names remain confidential, the company reveals that its Annual Recurring Revenue (ARR) has reached a multi-million-dollar figure.

Unitary will leverage the funding to expand its presence into new regions and bolster its talent pool. While the company has not disclosed its current valuation, it previously secured an initial funding round of under $2 million and an additional $8 million in seed funding, with notable investors such as Carolyn Everson, a former executive at Meta.

In recent years, numerous startups have emerged, each harnessing various aspects of artificial intelligence to develop content moderation tools. When you consider the immense scale of the challenge presented by video content, it becomes evident that AI is ideally suited for this task.

The sheer volume of data, amounting to tens and hundreds of zettabytes, generated and shared across platforms like YouTube, Facebook, Reddit, TikTok, dating sites, gaming platforms, videoconferencing tools, and other online spaces is staggering. Videos constitute over 80% of all online traffic, making it virtually impossible for a human workforce alone to effectively parse and moderate this vast digital landscape.

This perspective also caught the attention of investors. Christopher Steed, Chief Investment Officer at Paladin Capital Group, emphasized the significant demand in the online realm for a technology-driven solution to detect harmful content.

Nevertheless, it’s a competitive landscape. Numerous players, including OpenAI, Microsoft (utilizing its proprietary AI, not OpenAI’s), Hive, Active Fence/Spectrum Labs, Oterlu (now part of Reddit), Sentropy (now integrated into Discord), and Amazon’s Rekognition, are just a handful among the many operating in this space.

From Unitary AI’s perspective, existing tools have not delivered the level of effectiveness required for video content moderation. The reason behind this is that conventional tools have primarily been designed to parse data of singular types, such as text, audio, or images, rather than integrating them simultaneously. This approach often results in numerous false alarms or, conversely, missing problematic content entirely.

“What sets Unitary apart is our pioneering use of genuine multimodal models,” explained CEO Sasha Haco, who co-founded the company with CTO James Thewlis. “Instead of solely analyzing individual frames, understanding the nuances, such as whether a video is artistic or violent, requires simulating how a human moderator views the video. We achieve this by comprehensively analyzing text, sound, and visuals.

Customers can define their specific moderation parameters according to their needs. Typically, customers use Unitary in conjunction with a human moderation team, relieving them of a significant workload and reducing stress.

The concept of ‘multimodal’ moderation may seem self-evident, so why hasn’t it been implemented earlier? Haco explained that one reason is that the older, visual-only model could achieve a certain level of effectiveness. Nevertheless, this leaves an opportunity for further growth and innovation in the market.

The reality is that the challenges of content moderation continue to plague social platforms, gaming companies, and other digital channels where user-generated media is shared. Recently, social media firms have indicated a shift away from stricter moderation policies, fact-checking organizations are losing momentum, and ethical questions surrounding content moderation persist, especially in dealing with harmful content. The determination to combat these challenges has waned.

However, Haco brings a fascinating history of working on complex and enigmatic subjects. Prior to her role at Unitary AI, Haco, who holds a Ph.D. in quantum physics, conducted research on black holes alongside Stephen Hawking. (Interestingly, she can be seen in a Netflix documentary about the first image of a black hole, captured using the Event Horizon Telescope.) Her desire to shift her focus to earthly challenges highlights that these problems can be just as intricate to comprehend as the mysteries of spacetime and gravity monsters.

Her ‘epiphany,’ she explained, came from recognizing the abundance of content moderation products in the market, the noise surrounding them, and the absence of a solution that truly aligned with customer needs.

Thewlis, on the other hand, is putting his expertise to direct use at Unitary. With a Ph.D. in computer vision from Oxford, his specialization focused on ‘methods for visual understanding with less manual annotation.’

The name ‘Unitary’ holds a double reference. On one hand, the startup unifies various parameters to enhance video comprehension. On the other hand, it may allude to Haco’s previous career in quantum physics, where unitary operators are used to describe quantum states—a realm known for its complexity and unpredictability, much like the world of online content and human behavior.

Multimodal AI research has been in progress for several years, but we are now entering an era where the concept is poised to find numerous practical applications. A recent example is Meta, which prominently featured multimodal AI in its Connect keynote, unveiling new AI assistant tools. Unitary, positioned at the intersection of cutting-edge research and real-world application, exemplifies this exciting shift.

Principal at Creandum and board member, Gemma Bloemen, expressed their enthusiasm, stating, ‘We first met Sasha and James two years ago and have been incredibly impressed. Unitary has emerged as clear early leaders in the important AI field of content safety, and we’re excited to support this exceptional team as they continue to drive innovation in content classification technology.’

Ian Hogarth, a partner at Plural and also a board member, commended Unitary’s powerful AI capabilities for classifying harmful content, highlighting the company’s remarkable achievement of reaching seven figures in Annual Recurring Revenue (ARR) within this year, a feat rarely seen at such an early stage of a startup’s journey.

Conclusion

In conclusion, Unitary AI’s multimodal approach to video content moderation is a game-changer in the industry. By analyzing text, sound, and visuals simultaneously, Unitary is able to provide more accurate and nuanced results, reducing false flags and improving the moderation process. With their recent funding and growing customer base, it is clear that Unitary’s innovative solution is meeting the needs of technology companies in the ever-evolving landscape of content moderation. As the challenges of online media continue to persist, Unitary AI is paving the way for more effective and efficient moderation practices.

What are your thoughts on Unitary AI’s multimodal approach to video content moderation? How do you envision the future of content moderation evolving, considering the growing volume of user-generated content online and the need for more effective solutions? Share your insights below.