Online media content moderation remains controversial. New regulations and public concern will likely keep it a priority for years. Weaponized AI and other tech advances make it harder to address. Cambridge, England-based Unitary AI believes it has found a better way to moderate content by using a “multimodal” approach to parse video, the most complex medium.
Unitary announced $15 million in funding today to capitalize on market momentum. Top European VC Creandum led the Series A, which included Paladin Capital Group and Plural. Unitary’s business is growing. The platform is classifying 6 million videos per day, up from 2 million last year (covering billions of images), and adding languages beyond English. It says ARR is in the millions but won’t name customers.
Unitary is expanding into new regions and hiring more people with the funding. Former Meta executive Carolyn Everson invested in Unitary, which raised under $2 million and $8 million in seed funding. Its valuation is unknown.
In recent years, dozens of startups have used AI to build content moderation tools.
The sheer scale of video challenge is an appropriate application. No army of people could parse the tens and hundreds of zettabytes of data created and shared on YouTube, Facebook, Reddit, and TikTok, let alone dating sites, gaming platforms, videoconferencing tools, and other video-based platforms that account for more than 80% of online traffic.
Investors liked that angle too. “In an online world, there’s an immense need for a technology-driven approach to identify harmful content,” Paladin Capital Group chief investment officer Christopher Steed said.
Still, it’s packed. Microsoft, Hive, Active Fence / Spectrum Labs, Oterlu (now part of Reddit), Sentropy (now part of Discord), and Amazon’s Rekognition are among the many AI systems in use.
Unitary AI believes video tools are ineffective. That’s because tools have traditionally focused on parsing text, audio, or image data, not both. It causes many false flags (or none).
He cofounded Unitary with CTO James Thewlis. “What is innovative about Unitary is that we have genuine multimodal models,” Haco said. Instead of analyzing a series of frames, you need to simulate a human moderator’s viewing experience to understand a video’s nuance and whether it’s artistic or violent. We analyze text, sound, and images.”
Customers specify what to moderate (or not), and Haco said they usually use Unitary with a human team, which will work less and be less stressed.
Why hasn’t “multimodal” moderation been done before?
Haco noted that “you can get quite far with the older, visual-only model”. This creates a market gap for growth.
Content moderation issues continue to plague social media platforms, games companies, and other digital channels where users share media. Social media companies have softened their moderation policies, fact-checking organizations are declining, and the ethics of moderating harmful content are unclear. The desire to fight has declined.
However, Haco has a curious history of working on difficult topics. Haco, a quantum physics PhD, worked with Stephen Hawking on black hole research before Unitary AI. (She appears in a Netflix documentary on the Event Horizon Telescope’s first black hole image.) Her desire was to work on earthbound issues, which can be as confusing as a spacetime gravity monster.
She said her “ephiphany” was that there were so many content moderation products and so much noise, but nothing yet matched what customers wanted.
Meanwhile, Unitary is using Thewlis’s PhD in computer vision from Oxford, which focused on “methods for visual understanding with less manual annotation.”
I believe ‘Unitary’ is a double reference. The startup is combining parameters to understand videos. It could also refer to Haco’s previous career: unitary operators describe quantum states, which are complicated and unpredictable like online content and humans.
AI multimodal research has been ongoing for years. But we seem to be entering an era where the concept will be applied more. In its Connect keynote last week, Meta mentioned multimodal AI several times to preview its new AI assistant tools. Unitary sits at the intriguing intersection of cutting-edge research and real-world application.
“We first met Sasha and James two years ago and have been incredibly impressed,” said Creandum principal and board member Gemma Bloemen. “Unitary has emerged as clear early leaders in the important AI field of content safety, and we’re excited to back this exceptional team as they accelerate and innovate in content classification technology.”
Unitary had powerful AI for classifying harmful content from the start. Ian Hogarth, Plural partner and board member, said the company has reached 7 figures of ARR this year, which is unheard of at this stage.